The Ethics of Nudge: From Liberty to Autonomy · The Ethics of Nudge: From Liberty to Autonomy...
Transcript of The Ethics of Nudge: From Liberty to Autonomy · The Ethics of Nudge: From Liberty to Autonomy...
The Ethics of Nudge: From Liberty to Autonomy
by
Blake Austin Chapman
A thesis submitted in conformity with the requirements for the degree of Doctor of Juridical Science
Faculty of Law University of Toronto
© Copyright by Blake Austin Chapman 2016
ii
The Ethics of Nudge: From Liberty to Autonomy
Blake Austin Chapman
Doctor of Juridical Science
Faculty of Law University of Toronto
2016
Abstract
Behavioural insights – the way that individuals actually choose and behave in
real life, often contrary to the predictions of neoclassical economics – are increasingly
influencing law and policy. One of the policy tools developed in response to these
insights are nudges – a method of intentionally designing the context of choice so as to
steer individual choices in welfare-promoting directions while also preserving freedom
of choice.
Because of their potential to improve health and welfare in a libertarian manner,
nudges have exploded in popularity, deployed in disparate domains such as health
care, education, retirement saving, energy consumption, and tax compliance. As more
clever interventions are developed using behavioural insights, there is a tendency to
label them all as nudges. However, doing so misses the fundamental promise of
iii
nudging – improving welfare while respecting liberty (and autonomy). The first distinct
contribution of this thesis is to present a framework for identifying nudges as those
interventions in choice architecture that preserve the opportunity set of individual
choosers satisfying a liberty of choice constraint.
Once nudges have been properly identified, the second contribution is to provide
a framework for their ethical evaluation. To date, nudges have been critiqued on several
ethical grounds, for example, paternalism, manipulation, and for not being wholly
libertarian. However, most of these critiques suffer from a generalization problem – not
all nudges are the same and we should not try to evaluate them as a class. Instead, we
need a framework for evaluating the effects and functions of nudges individually. To
this end, I will offer a normative framework for the ethical evaluation of nudges that
uses autonomy (conceived of as participatory self-government) as its guide. Only those
nudges that preserve the autonomy of individual choosers will be found to be ethical.
The ethical framework will turn on a key distinction between participatory “process
autonomy” (actual participation as a decision-maker) and “outcome autonomy” (non-
participatory linking between motivations and choices).
iv
Acknowledgments
Over the last five years I've discovered that a doctorate is mainly a test in
perseverance. From a topic change two years in, to impostor syndrome, to the
motivational abyss of revisions, it's no surprise a doctorate is considered a "terminal"
degree. And then life outside the work changes too, and one's priorities take on new
life, and getting to the finish line takes on different meaning. The only way to survive is
with help, and that's why this page is the most important one in my entire thesis.
First, thank you to Bruce Chapman for taking on another "B. Chapman" all those
years ago and the constant support and psychotherapy since. I will never forget some of
the great discussions we had and the wacky hypotheticals we devised to test my
theories. Thank you as well to Hamish Stewart for his insightful comments and to
Colleen Flood for finding me the financial support to keep going and keeping me
grounded.
I'd be nowhere if it weren't for the incredible support of my family. Lindsay,
you've been simply amazing. You've carried our family on your shoulders long enough,
it's my turn now. And to my little girl Indy, when you were born I promised that I
would finish school before you started and, well, I made it (but just)! You two are all the
motivation I will ever need to get up every day and push forward.
v
Table of Contents
Acknowledgments.....................................................................................................................iv
Table of Contents.........................................................................................................................v
List of Figures.............................................................................................................................ix
List of Appendices.......................................................................................................................x
Chapter 1 Introduction...............................................................................................................1
1.1. Introduction.....................................................................................................................1
1.2. Nudging, Defined............................................................................................................3
1.3. Behavioural Insights & Nudge......................................................................................6
1.4. Nudge and the Law.......................................................................................................12
1.5. Two Goals & A Plan......................................................................................................15
PART I: IDENTIFYING NUDGES..........................................................................................21
Chapter 2 The Easy Cases of Opportunity Set Preservation...............................................23
2.1. Introduction...................................................................................................................23
2.2. Comparing Opportunity Sets......................................................................................24
2.3. Eliminating Options......................................................................................................27
2.4. Option Substitution.......................................................................................................28
2.5. Adding Options – Enhancing the Opportunity Set..................................................31
2.6. Perception of Options...................................................................................................34
2.6.1. Changing Salience by Affecting Option Awareness (#1-2)......................................39
2.6.2. Relative Changes in Salience (#3-5).............................................................................41
2.6.3. Option Comprehension (#6-7).....................................................................................42
2.7. Conclusion......................................................................................................................45
Chapter 3 The Hard Cases of Opportunity Set Preservation: Costs...................................46
3.1. Introduction...................................................................................................................46
3.2. Costs & Construing the Opportunity Set...................................................................47
3.3. Monetary Costs..............................................................................................................49
3.3.1. Local Budget Constraints.............................................................................................51
vi
3.3.2. Global Budget Constraints – Opportunity Costs......................................................54
3.4. Non-Monetary Costs.....................................................................................................57
3.4.1. Opportunity Costs & Finite Non-Monetary Resources............................................58
3.4.2. Identifying Finite vs. Infinite Non-Monetary Resources..........................................61
3.4.2.1. Case Study: Graphic Tobacco Warning Labels.........................................................65
3.4.2.2.Case Study: Defaults....................................................................................................69
3.5. Reducing vs. Increasing Opportunity Costs..............................................................73
3.6. Shifting the Burden of Costs........................................................................................77
3.7. Expected Costs & Benefits............................................................................................81
3.8. Conclusion......................................................................................................................85
PART II: EVALUATING NUDGES........................................................................................87
Chapter 4 Mapping Autonomy...............................................................................................89
4.1. Introduction...................................................................................................................89
4.2. The Value of Autonomy...............................................................................................91
4.3. Autonomy as Self-Government...................................................................................95
4.4. Authenticity: Higher-Order Motivations vs. Lower-Order Desires.......................97
4.5. Authenticity: Independence.......................................................................................103
4.6. Effective Self-Government...........................................................................................110
4.7. Process & Outcome Autonomy.................................................................................112
4.8. Conclusion....................................................................................................................119
Chapter 5 The Lexical Priority of Process Autonomy........................................................121
5.1. Introduction.................................................................................................................121
5.2. Participation.................................................................................................................122
5.3. Side-Benefits of Participatory Process Autonomy..................................................127
5.4. Heterogeneity of Motivations....................................................................................133
5.5. Non-Chooser-Regarding Motivations......................................................................136
5.6. Context-Dependent Motivations...............................................................................143
5.7. Undecided Motivations..............................................................................................156
5.8. Meta-Motivations........................................................................................................162
vii
5.9. Challenges of Process Autonomy..............................................................................169
5.10. Conclusion....................................................................................................................176
Chapter 6 Evaluating Nudge Effects on Process Autonomy.............................................177
6.1. Introduction.................................................................................................................177
6.2. Autonomy Preservation Standard of Evaluation....................................................178
6.3. No Aggregation Among Affected Choosers............................................................182
6.4. Independent Effects on Motivational & Decisional Autonomy............................184
6.5. Nudges & Manipulation.............................................................................................191
6.6. Nudging Motivational Autonomy............................................................................195
6.7. Nudging Decisional Autonomy................................................................................207
6.7.1. Deliberation..................................................................................................................210
6.7.1.1. Awareness & Active Choosing.................................................................................212
6.7.1.2.System 1 & System 2..................................................................................................220
6.7.2. Biases.............................................................................................................................229
6.7.2.1.Re-Biasing & Counter-Biasing..................................................................................235
6.7.2.2.De-Biasing....................................................................................................................241
6.7.3. Information...................................................................................................................245
6.8. Conclusion....................................................................................................................258
Chapter 7 Outcome Autonomy.............................................................................................261
7.1. Introduction.................................................................................................................261
7.2. Outcome Autonomy: Alignment Between Choices & Motivations......................262
7.3. Theoretical Outcome Autonomy Approaches.........................................................265
7.3.1. Libertarian Paternalism..............................................................................................266
7.3.2. Other Outcome Autonomy Approaches..................................................................272
7.4. Outcome Autonomy as a Tool of Last Resort..........................................................276
7.5. Conclusion....................................................................................................................279
Chapter 8 The Challenges of Outcome Autonomy.............................................................281
8.1. Introduction.................................................................................................................281
8.2. The Epistemic Problem...............................................................................................282
viii
8.3. Undecided Choosers...................................................................................................296
8.4. Short-Term Behaviour vs. Long-Term Motivations................................................300
8.5. Inapplicability of Outcome Autonomy to Some Choice Scenarios.......................303
8.6. Determining Effects of Nudges on Outcome Autonomy.......................................305
8.7. Aggregating Outcome Autonomy............................................................................313
8.8. Conditions on Using Outcome Autonomy as a Tool of Last Resort.....................320
8.9. Conclusion....................................................................................................................325
Chapter 9 Synthesizing the Autonomy Effects of Nudges................................................327
9.1. Introduction.................................................................................................................327
9.2. Table of Combinatory Autonomy Effects................................................................329
9.3. Unethical Nudges........................................................................................................330
9.4. Ethical Nudges.............................................................................................................339
9.5. Conclusion....................................................................................................................352
Chapter 10 Conclusion...........................................................................................................354
Bibliography.............................................................................................................................358
Appendix I – Glossary............................................................................................................374
ix
List of Figures
Figure 1 – Nudge Visualization ............................................................................................... 19
Figure 2 – Interventions Summary Table ............................................................................... 22
Figure 3 – Salience Effects on Opportunity Set Perception .................................................. 38
Figure 4 – Opportunity Set Intervention Summary Table, Revisited ................................. 86
Figure 5 – Simplified Decision-Making Process .................................................................. 114
Figure 6 – Defaults vs. Active Choosing ............................................................................... 215
Figure 7 – Active Choosing (accounting for meta-choices) ............................................... 217
Figure 8 – Combinatory Effects of Nudging on Autonomy .............................................. 329
Figure 9 – Combinatory Effects of Nudging on Autonomy, Unethical Only ................. 330
Figure 10 – Combinatory Effects of Nudging on Autonomy, Ethical Only .................... 339
x
List of Appendices
Appendix I – Glossary ............................................................................................................ 374
1
Chapter 1
Introduction
1.1. Introduction
In 2008, economist Richard Thaler and legal scholar Cass Sunstein
popularized a form of political theory called ‘Nudge’ in a bestselling book of the
same name.1 Nudge theory sought to bridge the political divide between classical
liberalism’s dogma of non-intervention and the interventionist ideal of welfare
paternalists.
Nudges, as forms of government intervention, offer the attractive promise to
policymakers of steering individual choices in a welfare-promoting direction, while
also preserving choosers’ liberty of choice. The seemingly conflicting ability of
nudges to change choices for the better, while also not interfering with individual
liberty, is why Sunstein and Thaler call nudge a possible “third way” in politics.
Nudges let policymakers have their cake (welfare improvements) and eat it too
(non-interference with choice and autonomy).
In the years since Thaler and Sunstein’s work on nudges, a cottage industry
has sprung up developing a seemingly endless list of clever nudges as solutions (or
1 Richard H Thaler & Cass R Sunstein, Nudge: Improving Decisions About Health, Wealth, and Happiness (New York: Penguin Books, 2009) [Thaler & Sunstein, "Nudge"].
2
partial solutions) to many serious and less-serious social problems on topics as
diverse as taxes, obesity and, yes, even urinal spillage in airport bathrooms.2
The reach of behavioural insights has even penetrated the civil service. The
UK government established and then spun off into a social purpose company, a
dedicated ‘nudge unit’, called The Behavioural Insights Team, to apply behavioural
insights to British public policy.3 The team has several ‘achievements’ to date
including instituting ‘prompted choice’ for organ donation, as well as boosting tax
repayment rates by 15% by altering the content of tax letters using social cues. A
similar team has been formed within the Obama Administration,4 with Canada not
far behind.5
However, nudging is not without its ethical concerns. Nudging has been
criticized for being manipulative, exploitative of individual cognitive weaknesses,
and as unjustifiably paternalistic.6 This thesis will show that nudging as a whole, is
2 See e.g. Ibid. 3 The Behavioural Insights Team, online: <http://www.behaviouralinsights.co.uk/>. For an excellent examination on the background of behavioural public policy, particularly in the UK, see Rhys Jones, Jessica Pykett & Mark Whitehead, Changing Behaviours: On the Rise of the Psychological State (Cambridge: Edward Elgar, 2013). 4 Francesca Gino, “Why the U.S. Government Is Embracing Behavioral Science”, (18 September 2015), online: Harvard Business Review <https://hbr.org/2015/09/why-the-u-s-government-is-embracing-behavioral-science>. 5 Kathryn May, “Government’s new innovation ‘Hub’ open to new thinking”, Ottawa Citizen (11 February 2015), online: <http://ottawacitizen.com/business/local-business/governments-new-innovation-hub-open-to-new-thinking>. 6 See e.g. Luc Bovens, “The Ethics of Nudge” in Till Grune-Yanoff & SO Hansson, eds, Preference Change: Approaches from Philosophy, Economics and Psychology (New York: Springer, 2008); T M Wilkinson, “Thinking Harder About Nudges” (2013) 39:8 J Med Ethics 486; Mark D White, The Manipulation of Choice: Ethics and Libertarian Paternalism (New York: Palgrave Macmillan, 2013).
3
neither ethical nor unethical.7 Instead, such determinations should be contingent on
the particular circumstances of a given nudge and its affected choosers. Some nudges
are ethical on autonomy grounds and some are unethical. Likewise, some
interventions that many think of as nudges are in fact not nudges. These twin efforts
– identifying nudges and developing and applying a framework for their ethical
analysis based on autonomy – are this thesis’s dual purposes and original
contributions to knowledge.
1.2. Nudging, Defined
There is considerable disagreement in the literature about what counts as a
genuine nudge.8 This confusion surrounding nudging creates two problems. The
first is the risk that nudging will collapse under its own weight.9 Practitioners,
academics and policymakers using the same term to mean different things suggests
that the term has lost, or is close to losing, its meaning, and thus its normative force.
The second issue is that we cannot possibly hope to evaluate nudges for their
ethical implications without an understanding of what counts as a nudge in the first
7 Throughout this thesis, reference to “unethical” and “ethical” should be read as “unethical on autonomy grounds” and “ethical on autonomy grounds,” respectively. We are only conducting a modest ethical analysis using autonomy as our guide, not a comprehensive multi-principled evaluation. 8 See e.g. Evan Selinger & Kyle Powys Whyte, “What Counts as a Nudge?” (2012) 12:2 Am J Bioeth 11. 9 J.S. Mill’s famous Harm Principle arguably has suffered that very fate. See Hamish Stewart, “The Limits of the Harm Principle” (2010) 4 Crim Law Philos 17.
4
place. We are interested in examining the ethics of nudging, not behaviour change
through law and policy generally. This is partly self-serving – to contain the scope of
this thesis – but it is also important to restrict our ethical analysis to just nudging
because, despite the confusion, nudging does represent a new way of achieving
behaviour change, and thus it implicates autonomy and ethics in a manner and
depth that traditional tools do not.
In their seminal book Nudge, Sunstein and Thaler define a nudge as, “any
aspect of the choice architecture that alters people’s behaviour in a predictable way
without forbidding any options or significantly changing their economic
incentives.”10 At a minimum, a nudge seems to requires three things. First, the
intervention has to affect part of the choice architecture. This is an easy hurdle since
choice architecture is defined as the, “the context in which people make decisions.”11
Choice architecture is the context of choice. This is an extremely broad concept,
encompassing every external influence on choice, including price, non-financial
costs, information, the manner of presentation, cognitive cues, social cues, etc. As
Sunstein writes: “[c]hoice architecture specifies when, whether, and how we
10 Thaler & Sunstein, "Nudge", supra note 1 at 6. 11 Ibid at 3.
5
choose.”12 However, as we will see, not every change in choice architecture counts as
a nudge.
The second requirement for an intervention to count as a nudge is that it has
to change people’s decisions/choices – what options they select – in a predictable
way. Choice architecture that had no impact on a chooser’s choices would not count
as a nudge, nor would choice architecture that had an uncertain impact on the
direction of choices.
Finally, as we will see in Part I, to be a nudge, an intervention in choice
architecture must meet a liberty of choice constraint – the opportunity set (the set of
options from which choosers can select) faced by choosers must be the same before
and after a choice architectural intervention.13 Eliminating options or adding
significant taxes to certain options materially alters the opportunity set, thereby
violating the liberty of choice constraint. To Sunstein and Thaler, nudges are every
element of choice architecture except for forbidding options or significantly changing
economic incentives.14 These exclusions are mentioned not because they are
avoiding economic measures, but because of what those economic approaches mean
for liberty of choice. Sunstein writes that nudges should be, “understood as
12 Cass R Sunstein, Choosing not to Choose: Understanding the Value of Choice (New York: Oxford University Press, 2015) [Sunstein, “Choosing”] at 5. 13 Cass R Sunstein & Richard H Thaler, “Libertarian Paternalism Is Not an Oxymoron” (2003) 70:4 Univ Chicago Law Rev 1159 at 1161. 14 Ibid at 6.
6
interventions that maintain freedom of choice, that do not impose mandates or bans,
but that nonetheless incline people’s choices in a particular direction.”15 So then,
obviously taxes, fines, prohibitions, restrictive licensing, and of course jail terms are
not nudges since they fail to meet the liberty of choice constraint – i.e. they do not
maintain freedom of choice.
The essence then of nudging is opportunity set preservation – any
intervention that alters the context of choice without changing the options available
to the chooser. By “preservation” we mean that post-intervention, a chooser must
have fundamentally the same opportunities/options that they had pre-intervention.
Options must not be eliminated or reduced, but, as we will see, they can be
enhanced. The key question then that Part I of this thesis seeks to answer is: when
can an intervention be said to reduce a chooser’s opportunity set?
1.3. Behavioural Insights & Nudge
Much of the impetus for nudging comes from findings from behavioural
insights (behavioural economics, social psychology and cognitive psychology) – that
show that humans don’t always act rationally. However, behavioural insights and
nudging, while closely linked, are not inextricably so.
15 Sunstein, "Choosing", supra note 12 at 6.
7
The discipline of behavioural economics was founded nearly four decades
ago as a realist counterpoint to the dominant school of neoclassical economics. Lab
and field experiments repeatedly showed that humans often deviate from the
rationality predictions of neoclassical economic thought (Homo Economicus).
Behaviour that systematically deviated from that of Homo Economicus can be
considered errors, mistakes, individual failure16, shortcomings, anomalies17,
cognitive failure18, etc. Indeed, by focusing on real human behaviour, as opposed to
hypothetical Homo Economicus behaviour, the scope is arguably broadened for
government intervention.19
Behavioural insights are not just about how humans decide, but also about
how we think. The groundbreaking work of Nobel Prize winner Daniel Kahneman
shows that individual cognitive processes tend to be instinctually fast and frugal
(what he calls “System 1” thinking) and to a less frequent extent, slow and
deliberative (“System 2”).20 System 1 is instinctual, innate, and “operates
automatically and quickly, with little or no effort and no sense of voluntary
16 Philip Jones & John Cullis, “‘Individual Failure’ and the Analytics of Social Policy” (2000) 29:1 J Soc Policy 73. 17 Daniel Kahneman, Jack L Knetsch & Richard H Thaler, “Anomalies: The Endowment Effect, Loss Aversion, and Status Quo Bias” (1991) 5:1 J Econ Perspect 193. 18 Jayson L Lusk, “Are you smart enough to know what to eat? A critique of behavioural economics as justification for regulation” (2014) 41:3 Eur Rev Agric Econ 355. 19 Jones, Pykett & Whitehead, supra note 3 at 3. 20 Daniel Kahneman, Thinking, Fast and Slow (Toronto: Anchor Canada, 2013).
8
control.”21 In contrast, System 2 reflects the conscious reasoning part of the brain,
and requires greater effort. It “allocates attention to the effortful mental activities
that demand it, including complex computations.”22
Consider a classic example of how human behaviour deviates from that
predicted by neoclassical economics – the framing effect. A famous study run by
McNeil and others found that patients who were presented with survival statistics
for a procedure as opposed to mortality statistics were more likely to consent to the
treatment.23 From a logical, neoclassical standpoint, it should not matter whether
information is presented as a 90% chance of survival or a 10% chance of death, since
both are mathematically identical. But, for many people it does matter. This
deviation from logic is the evidence of a behavioural bias that we now call the
“framing effect.”
We also have all suffered at some point in our lives the action-intention gap,
where our good intentions are laid to waste by snap, tempting choices. As detached
observers, most us would agree that we need to eat healthier, make wiser financial
choices, and spend more time doing things that make us deeply happy. But when
21 Ibid at 20. 22 Ibid at 21. 23 Barbara J McNeil et al, “On the Elicitation of Preferences for Alternative Therapies” (1982) 306 N Engl J Med 1259.
9
the time to enact those intentions come to the fore, for a variety of reasons, our self-
control crumbles and we choose otherwise.
As Conly writes: “We have visions of ourselves as free, rational beings who
are totally capable of making all the decisions we need to in order to create a good
life. Give us complete liberty, and, barring natural disasters, we’ll end up where we
want to be. It’s a nice vision…[b]ut it’s false.”24
Behavioural insights have proven pivotal in informing the development of
nudges. Without insight into how humans think and choose, the small tweaks to
choice architecture that are the essence of nudging (and also what it gives it its
political popularity) could not be successful.
However, there is a misconception prevalent throughout much of the nudge
literature that to qualify as a nudge, a form of choice architecture must necessarily
play on or rely on a chooser’s behavioural biases or heuristics.25 That is, there is an
assumption that what makes a nudge qua nudge, is that it plays on or utilizes our
biased, instinctual cognition (our System 1). This conception of nudge excludes
measures such as advertising and informative warning labels that work through
24 Sarah Conly, “Three Cheers for the Nanny State”, The New York Times (24 March 2013) online: <http://www.nytimes.com/2013/03/25/opinion/three-cheers-for-the-nanny-state.html>. 25 See e.g. Bovens, supra note 6; Daniel M Hausman & Brynn Welch, “Debate: To Nudge or Not to Nudge” (2010) 18:1 J Polit Philos 123 at 127.
10
choosers’ rational decision-making (provided they are not overly graphic and thus
blurring the line between informing and persuading).
Such a conception of nudging runs counter to Sunstein and Thaler’s
definition. As Sunstein notes in his own follow-up book Why Nudge, there can be
System 2 or reason-appealing nudges.26 The definition of nudge is broad enough to
encompass information disclosure because that very much forms part of the context
of choice, and is not excluded by the liberty of choice constraint. So long as a
person’s opportunity set is preserved, nudges do not explicitly have to target
individual’s behavioural biases or heuristics.
Behavioural insights are also used in the world of nudging to identify
arguably “poor” choices that nudges can correct for (implicating welfare), and in
justifying interventions from an autonomy perspective. Regarding the former, as the
evidence from behavioural insights shows, humans make “poor” choices as a result
of not necessarily conscious choice, but by unconscious thinking. While nudges can
act by appealing to either System 1 or System 2 thinking, they are primarily targeted
at the negative welfare results of System 1 behavioural biases and heuristics.
Decisions resulting from biases and heuristics are suboptimal from a welfare
maximization perspective. This feeds into the real promise of nudges – to take
26 Cass R Sunstein, Why Nudge? The Politics of Libertarian Paternalism (New Haven: Yale University Press, 2014) [Sunstein, “Why Nudge”] at 59.
11
seemingly “bad” choices (objectively determined – e.g. failing to save for retirement,
failing to donate one’s organs, smoking cigarettes, etc.) and steer the chooser away
from them and toward “good” choices without infringing on liberty (choice
constraint). It’s seemingly the best of both worlds.
We see this impetus for nudging in later, expanded definitions. In one of his
later solo efforts, Sunstein defines a nudge as, “initiatives that maintain freedom of
choice while also steering people’s decisions in the right direction (as judged by
themselves).”27 This modified definition inputs a chooser-centric welfare direction,
but it also implicates autonomy as self-government by providing the final
directional judgment to the chooser themselves.
Decisions resulting from behavioural biases may also be considered
suboptimal if we are interested in respecting individuals’ autonomous choices over
their non-autonomous choices. Autonomy here means an alignment between a
person’s true motivations and their actions, and an overarching preference for
deliberative, unbiased decisions over automatic, biased choices. Biases such as the
status quo bias, inertia, and other forms of System 1 thinking often prevent
individuals from choosing in accordance with their own values and beliefs, reflected
on and endorsed by their System 2 cognition.
27 Ibid at 17 (emphasis added).
12
As Sunstein and Thaler write, oftentimes we make choices that we otherwise
wouldn’t have made if we, “had paid full attention and possessed complete
information, unlimited cognitive abilities, and complete self-control.”28 Therefore,
nudging individuals to make choices that their own System 2 has already judged to
be in their best interests may respect the chooser’s autonomy. We will explore this
possibility in Part II.
In sum, as we will see throughout this thesis, nudges are informed and
motivated by behavioural insights, but they are not defined by them.
1.4. Nudge and the Law
Nudging is a form of regulation, just as command and control laws are.
Nudges specifically implicate the law in two key areas. First, when the government
is the party doing the nudging (or mandating that others nudge on their behalf), the
government is using nudges as regulatory tools. Second, nudges can be used by
private actors to alter the behaviour of consumers and, given the effects that
nudging can have on autonomy, some nudges may appropriately be subject to
regulation (e.g. consumer protection legislation).
Regarding the former, at its core, law is about the regulation of human
behaviour and nudges represent a growing element in governments’ legal or public
28 Thaler & Sunstein, "Nudge", supra note 1 at 5.
13
policy toolbox. In fact, nudges have been prevalent in the legal system for some
time, particularly in the form of default rules. As Sunstein (a legal scholar) writes:
“One of the most important tasks of a legal system is to establish default rules.”29
Understanding how defaults work and which defaults are acceptable is an
important step in using defaults in the legal system (e.g. intestacy rules, default
contractual terms), particularly because defaults have been touted as powerful
nudges.30
The second way nudges implicate law is the potential for consumer
protection laws to regulate the use of nudges by private actors. Many nudges use
behavioural insights to achieve their goals, raising the spectre that human
weaknesses could be exploited for private gain. If autonomy is our concern, then
perhaps nudges that cause a decrease in individual autonomy (for example, by
causing consumers to decide with greater bias) ought to be regulated.
The normative frameworks advanced in this thesis for the identification and
evaluation of nudges can be used for both types of legal analysis. Regarding nudges
as regulatory tools, using both the liberty of choice constraint and the preservation
of autonomy restraint, ethical nudges can be identified from the broader category of
choice architecture tools. Given the special relationship between governments and
29 Sunstein, "Choosing", supra note 12 at 7. 30 See e.g. Thaler & Sunstein, "Nudge", supra note 1.
14
citizens’ autonomy and liberty, the frameworks in this thesis may prove useful for
delineating the appropriate scope of nudges that governments can draw on as
public policy tools. The political attractiveness of using nudges that respect both the
liberty of choice and autonomy of choosers may also aid governments and
policymakers in restricting their scope of action to those nudges that satisfy both
requirements.
Finally, the autonomy framework in particular can be useful for governments
seeking to regulate the deployment of nudges by private actors, and may be helpful
in underpinning future work in consumer protection legislation. For example,
governments may be interested in restricting the use of nudges (many of which
operate through an individual’s deficient decision-making process), to those that
preserve the autonomy of choosers and not make it worse, through, for example, the
introduction or expression of new behavioural biases in a chooser’s reasoning
process.
As Part II will show, restricting or regulating nudges on the basis of an
autonomy evaluation requires carefully unpacking the effects of nudges on the
autonomy of choosers. Using autonomy as a guide for consumer protection efforts is
a promising line of inquiry, and this thesis represents a solid foundation to advance
that work in future projects.
15
1.5. Two Goals & A Plan
This thesis has two goals, mirroring its original contributions to knowledge
and its organization structure. Part I will identify nudges as a subset of choice
architecture that preserves the liberty of choice of choosers. Part II will then show
that nudges should be considered ethical on autonomy grounds if they preserve the
autonomy of choosers. The overarching goal of both Parts is to provide frameworks
for identifying and ethically evaluating nudges. We will use examples throughout of
fictional and real nudges and non-nudges to emphasize distinctions, but utilizing
the two frameworks of this thesis will, in practice, be highly fact-specific. In other
words, the examples contained herein are only for illustration purposes – the world
of nudging is much bigger and varied than contemplated by this thesis.
Part I will show that we ought to adopt a liberty of choice constraint to guide
our definition of nudging. It is the liberty of choice constraint that gives nudges their
“soft” regulatory flavour – they are designed to allow individuals to maintain their
freedom of choice. The liberty of choice constraint is what makes nudges nudges, and
not “harder regulations” that are coercive such as taxes, fines, and prohibitions. The
first step in our analysis is to clarify what counts as a nudge and what does not,
using the liberty of choice constraint as our guide. To that end, this section will ask:
does the proposed intervention reduce a chooser’s opportunity set (the set of
16
options from which they can select)? If yes, the intervention is not a nudge, and
therefore outside the scope of consideration for this thesis.
Chapter 2 will identify opportunity set preservation as the tool for
distinguishing nudges from other modes of choice architecture. Chapter 2 will also
examine the easy cases for determining opportunity set preservation – elimination
(non-preservative), option substitution (non-preservative), adding options
(preservative), and altering choosers’ subjective perceptions of their opportunity sets
(preservative). Chapter 3 will then examine the “hard cases” for determining
opportunity set preservation – nudges that alter the cost of options within an
opportunity set. The ideas of opportunity costs and finite resources/budget
constraints will be central to that discussion.
Focusing solely on the liberty of choice constraint doesn’t get us very far from
an ethical perspective, since it merely tells us the basic functional hurdle that
interventions must overcome to be classified as true nudges. To evaluate nudges
ethically, we need a more robust concept.
To that end, Part II will provide a normative framework for the evaluation of
the ethics of nudging using an autonomy restraint as our guide. To be an ethical
nudge on autonomy grounds, a chooser’s overall autonomy must not be diminished
by the intervention. Note that this thesis does not purport to claim that autonomy is
the only or overriding ethical concern; it is just the one most relevant to nudging. As
17
such, when reference is made in this thesis to “ethical” or “unethical,” such
descriptors relate only to judgments about autonomy.
Part II will proceed as follows. Chapter 4 will unpack our understanding of
autonomy as authentic, participatory self-government. A person is self-governing to
the extent they are actually ruled by their authentic motivations. A nudge can affect
a person’s autonomy at any of the three stages of decision-making: motivation
formation stage (motivational autonomy), option evaluation stage (decisional
autonomy), and the resulting choices (outcome autonomy). A person’s autonomy at
the first two stages we will call “process autonomy,” while the autonomy at the final
stage we will call “outcome autonomy” (the matching of actual choices with what a
person autonomously would have chosen – their authentic motivations).
One of the key contributions of this thesis is to show that not all nudge effects
on autonomy are created equally. Chapter 5 will show that any effects of a nudge on
a chooser’s process autonomy are determinative of the autonomy ethicality of that
nudge (i.e. lexically prior). This stems from the nature of process autonomy as
reflecting the ideal of participatory self-government, whereas outcome autonomy is a
derivative, shadow concept of process autonomy. Chapter 5 will also briefly
consider the benefits of process autonomy, particularly its neutrality regarding the
content of chooser motivations.
18
Chapter 6 will show that, in light of the lexical priority of process autonomy,
for evaluating the ethical implications of nudges, our standard of evaluation is one
of process autonomy preservation. This standard applies at the individual chooser
level and should be applied independently to the two constituent parts of process
autonomy. Chapter 6 will then present a detailed framework for evaluating nudges’
effects on the process autonomy of choosers by affecting their motivation formation
stage of decision-making (motivational autonomy) and their option evaluation stage
(decisional autonomy). Regarding the latter, indicators of autonomy are the
opportunity for deliberation, the absence of cognitive biases, and informed decision-
making.
Chapters 7 and 8 will then consider nudges and outcome autonomy. Chapter
7 will show that outcome autonomy nudges (such as libertarian paternalist nudges)
function by using a hypothetical construct of an autonomous chooser to set the
direction for nudging. Chapter 8 will then briefly outline the drawbacks of outcome
autonomy as compared to process autonomy, most notably epistemic problems, and
that it requires the use of autonomy aggregation in conflict with the essence of
autonomy.
Finally, Chapter 9 will synthesize the combination of effects of nudging on
process and outcome autonomy with a final conclusion on the autonomy ethicality
of particular nudge classes. Chapter 10 will then offer concluding remarks.
19
Figure 1 below, illustrates the work of this thesis. In Part I, we narrow the
world of choice architecture (i.e. the context of choice/decision-making), to nudging
by using a liberty of choice screening framework. We are not interested in
behavioural change interventions generally, only the subset that meet the special
libertarian promise of nudging. In Part II, we then provide an evaluative framework
for identifying those nudges that can be considered ethical on autonomy grounds
because they preserve the autonomy of choosers
Figure 1 – Nudge Visualization
Figure 1 shows, as previously noted, that autonomy may not be the only
ethical principle applicable to nudging; it is simply one that is highly relevant. There
ChoiceArchitecture
Ethical
Nudges
Autonomy-repsecting
20
very well may be interventions that definitionally qualify as nudges and that respect
the autonomy of choosers, but are unethical on other grounds. For example, an
intervention that instituted paying individuals for their organs is a nudge that
respects individual autonomy but that is often considered otherwise “wrong.”31
Likewise, there may be some nudges that do not survive our autonomy
evaluation, but can on the whole still be considered ethical. For example, nudges
targeting drunk drivers may not respect the autonomy of drunk drivers, but given
the harm to others, autonomy is a subordinate ethical consideration.32 Finally, there
may be choice architectural interventions that are not nudges, but that are
nonetheless ethical on the whole or simply on autonomy grounds. In this thesis, we
are only interested in those subset of interventions that: (a) definitionally qualify as
nudges, and (b) respect the autonomy of choosers.
31 Many thanks to Colleen Flood for this suggestion. 32 Thank you to Mark White for this clarification.
21
PART I: IDENTIFYING NUDGES
Liberty Constraints
Part I will proceed as follows. Chapter 2 will detail the use of opportunity set
preservation as the most appropriate tool for identifying nudges from other forms of
choice architecture, beginning with some of the “easy” cases: option elimination,
option substitution, adding options, and altering choosers’ subjective perceptions of
their opportunity sets.
Chapter 3 will then examine the “hard cases” – how altering the cost
structure of options may reduce an opportunity set. To achieve that goal, we must
consider two additional concepts: “budget constraints” and “opportunity costs.”
And if we are to take opportunity set preservation as the elemental part of the
liberty of choice constraint, then we must also consider that a chooser’s opportunity
set might be equally reduced due to non-monetary factors.
Figure 2 below lists the classes of interventions Part I will evaluate.
22
Figure 2 – Interventions Summary Table
Option Elimination – Prohibitions
Option Substitutions
Adding Options – Enhancing the Opportunity set
Altering the Perception of Choices (Salience, Awareness, Expected Costs &
Benefits)
Costs
- Monetary costs (e.g. price)
- Increasing vs. reducing costs
- Opportunity costs vs non-opportunity costs
- Non-monetary costs (opportunity vs. non-opportunity)
- Opportunity cost shifting
- Expected costs and benefits
23
Chapter 2
The Easy Cases of Opportunity Set Preservation
2.1. Introduction
To count as a nudge, a choice architectural intervention must change choices
without changing a chooser’s objective opportunity set: “the set of options from
which he [a chooser] is free to choose.”33 This is accomplished through choice
architecture that alters the context of choice without fundamentally altering the
options themselves. A failure at this stage condemns an intervention to be labeled as
something more than a nudge.
It is possible that some opportunity set reductions may, on balance, be
beneficial from an autonomy-perspective in order to relieve problems of choice
overload (e.g. reducing an opportunity set from 100 options to a far more
manageable 20); however such interventions are not nudges as we understand them
and are thus beyond the scope of this thesis. The liberty of choice constraint is
strictly a functional one: some interventions may pass this stage yet still be
considered unethical on autonomy grounds, and vice-versa.
33 Robert Sugden, “The Opportunity Criterion: Consumer Sovereignty Without the Assumption of Coherent Preferences” (2004) 94:4 Am Econ Rev 1014.
24
This chapter will examine the (relatively) easy cases of determining
opportunity set preservation. In order to count as nudges, choice architectural
interventions cannot eliminate options or (in most cases) substitute options. On the
other hand, the introduction of new options does not affect existing options within
the opportunity set. Likewise, choice architecture that works on choosers’ subjective
perception of their options preserves the opportunity set, thereby definitionally
counting as nudging.
2.2. Comparing Opportunity Sets
Determining opportunity set preservation necessitates some form of
comparison – we have to compare what choices existed prior to an intervention and
what exists afterwards. It seems that when we are analyzing changes in opportunity
sets we have several alternatives for our comparison sets. While our second
comparison group will always be a chooser’s opportunity set with or after a
proposed nudge intervention (since that is what we are interested in evaluating), the
first or baseline comparator set is more debatable.
We have two classes of alternatives for the baseline comparator. First, is the
opportunity set a chooser faces immediately “prior” to the intervention (e.g. pre-
nudge). Such an opportunity set would include other opportunity costs and forms
of choice architecture that have already been imposed on certain choices. This is an
25
ex-ante method of comparison – meaning immediately before the event (i.e.
impugned nudge). The second option for our baseline comparator, is to attempt to
visualize an opportunity set our chooser would face without any intervention at all;
some sort of “pure” choice stripped (if theoretically possible) of all forms of choice
architecture.
Any attempt at a pure comparison will quickly run into difficulties, because
as Sunstein and Thaler repeatedly argue, choice architecture is mostly inevitable. It
is impossible to imagine choices absent any choice architecture. A seemingly simple
choice as to whether or not to purchase cigarettes in the face of graphic warning
labels is the product of a vast chain of choices made by the would-be smoker and
others. There are the obvious cases of choice architecture at play: tobacco packaging,
advertising and marketing efforts by the manufacturer, counter-advertising in the
form of public health campaigns (and its effect on the social status of smoking),
workplace rules, municipal regulations; and then the less obvious ones such as
zoning bylaws for retailers, retailers display preferences, the route the smoker takes
past the retailer, the smoker’s mental state caused by other choices, their financial
picture (which is impacted by other choices), etc.
The point is that attempting to imagine a choice inside a bubble quickly turns
into an exercise in futility. In our interconnected world there is no hope of
successfully stripping choices down to the studs without an impossibly detailed
26
historical analysis of each stage of choice architecture. All we can hope to do is to
evaluate each subsequent effort at choice architecture in its own lights – what does it
do to a chooser’s opportunity set?
This analysis is far easier since we know how choices look beforehand (as
they exist now), and we can easily theorize (or even observe) how they change in
response to a particular form of choice architecture. Thus, the ex-ante approach
must be the preferred method of evaluating opportunity set preservation.
For the purposes of determining whether an intervention preserves a
chooser’s opportunity set, we must disregard the quality of all pre-existing forms of
choice architecture and take choosers as they are immediately prior to an
intervention. This approach allows us to focus on the effects of an intervention – the
change from a certain state of affairs to another state of affairs. We are not assessing
a static state of affairs; we are simply judging the nudge. This reflects the attraction
of nudging – that they have certain effects.
But it is the character of the change that matters; it is what gives nudges their
political attraction (i.e. the preservation of chooser’s opportunity sets). How an
opportunity set came to be as it is immediately pre-intervention is irrelevant for
determining whether and how it changes post-intervention. If their opportunity set
is x, we should not inquire as to why it is x and not x+y. After all, we are only
27
interested in whether x still exists post-intervention in order to determine if the
intervention counts as a nudge.
2.3. Eliminating Options
The opportunity set preservation requirement of nudging reflects the
traditional liberal economic view that more options are better than fewer options, all
things considered. If a chooser faces a menu of food options including salad, pasta
and cheeseburger, and a health-minded intervention eliminates the cheeseburger
option, then the opportunity set has been detrimentally reduced.
This cheeseburger example can also be used to distinguish between practical
and cost-based option eliminations. If a restaurant ceases to offer a cheeseburger,
that is a practical option elimination – there simply is no way for a patron of that
restaurant to purchase a cheeseburger, regardless of cost. However, let’s imagine
instead that, in an effort to improve the health of its citizens, the government
institutes a massive tax on certain fatty foods, like a cheeseburger that effectively
triple its price. That is a cost-based intervention, that, as we will see in Chapter 3,
has the effect of eliminating the cheeseburger from many individuals’ opportunity
sets. Though not technically eliminated – people can still buy the cheeseburger – it is
no longer an option which allows the same access to other options in the previous
opportunity set.
28
2.4. Option Substitution
The substitution of any existing option for one or more new options does not
meet the choice preservation criterion of the liberty of choice constraint since such
substitution operates by eliminating a pre-existing option, just as prohibitory
interventions do. However, it is important to distinguish between eliminatory
substitutions and cost-based substitutions.
Eliminatory substitutions are cases of choice architecture that clearly eliminate
a pre-existing option. If a chooser’s opportunity set pre-intervention is A or B, and
post-intervention the opportunity set becomes A or C, this is a case of option
substitution, where we have substituted C for B.
The other class of substitutions are those where the pre-intervention option
still exists after the intervention, but its character has been altered in some way,
almost always through the addition of costs. If a chooser’s opportunity set is A or B
pre-intervention, and then post-intervention it is A or B+x (where x are new costs),
we will consider this type of intervention to be altering options (as opposed to option
substitution). While we could say that B+x has been substituted for simply B, it is
more appropriate to deal with this kind of choice architecture in subsequent sections
on costs, since it is not a pure case of substitution as option B still exists.
29
Imagine Sunstein and Thaler’s paradigmatic nudge example of a cafeteria
servery with a variety of food options arranged in a certain order.34 An intervention
designed to encourage healthy eating may focus on altering the arrangement of the
food so that patrons encounter the healthier items first and fill up their tray with
those selections. An intervention in the choice architecture could also mandate a
prohibition on items like cookies or potato chips that fail to meet certain health-
related criteria. The prohibited items would then no longer be offered beside the
healthy items. Finally, a choice architectural intervention could substitute healthier
options for unhealthy options – banana chips for potato chips; diet soda for full
calorie soda; etc. As you can imagine, such a measure would have the same effect as
a prohibition – the unhealthy options would be eliminated from diners’ opportunity
set. The fact that new choices are correspondingly added does not change the fact
that some pre-existing options have been removed from the opportunity set.
In the case of, for example, substituting Diet Coke for regular Coke, an
argument could be made that the opportunity set is “almost the same” pre- and
post-intervention. After all, the option of whether to drink some sort of soda product
or some other beverage remains unaltered. However, as the next section will show,
a large part of the work is done on how we conceive of the opportunity set – we
should conceive of it as narrowly as possible. The choice is not between some soda
34 Thaler & Sunstein, "Nudge", supra note 1 at 1.
30
and other beverage, but between all the varieties of sub-beverages as well. Thus,
eliminating the regular Coke option does eliminate pre-existing options in a manner
inconsistent with the opportunity set preservation requirement of the liberty of
choice constraint.
We encounter a more difficult case with interventions that affect not the
existence of particular options (e.g. regular Coke) but costs associated with those
options. After all, in a way, every case of imposing new costs is a case of substitution
– the option that existed pre-intervention ceases to exist exactly as it was because the
cost associated with that option has changed.
Take the example of graphic warning labels on cigarette packages. If we take
smoking/buying cigarettes as an option that exists immediately prior to an
intervention (i.e. with simple text warning labels and more manufacturer branding
on the packages), and then compare it with the option that exists post-intervention
(graphic warning labels), one could argue that we have substituted the option to
smoke (S) with a new choice S+c (smoking, plus new costs of disgust, increased
social opprobrium, etc.). You could make the argument that we have effectively
eliminated an option – just S – since a smoker can no longer choose to smoke as they
once did without having to be exposed to graphic warning labels.
This is a strong argument, but it’s not terribly helpful for conceiving of
nudges, since any imposition of costs could then be considered choice elimination
31
and this would end up collapsing the nudge category entirely since, as we will see,
the category of “costs” is fairly broad. Instead, we should acknowledge that choice
architecture involving costs are a different category of nudges than cases of pure
option substitution, and best dealt with separately in Chapter 3. The distinction is
important because some interventions that involve costs may be considered nudges,
whereas, as we have discussed, pure substitutions are not nudges.
2.5. Adding Options – Enhancing the Opportunity Set
So far we have seen that interventions can have two categories of effects on
options: options may be eliminated (prohibition) or substituted (which in the case of
pure, non-cost substitutions, amounts to elimination). Both of those types of
interventions fail to preserve a chooser’s opportunity set. Note that our requirement
for opportunity set preservation, does not mean that opportunity sets must be
identical pre- and post-intervention. It simply means that, at a minimum, options
that existed pre-intervention must still exist post-intervention. As this section will
show, our preservation criterion does not preclude the addition of new options.
The addition of options appears to meet the liberty of choice constraint of
nudging since all pre-existing options are preserved. True option additions do not
directly (and objectively) affect either the existence or character (cost) of pre-existing
options. While it may seem that adding options does not accord with a strict
32
definition of “preserved,” vis-à-vis a chooser’s opportunity set, recall that we are
interested in the state of pre-existing options. It is what happens to those pre-
existing options that guides our analysis and conclusions on the liberty of choice
constraint. In that sense then, the opportunity set is preserved since all pre-existing
options remain untouched.
We need to distinguish between the pure addition of options and additions
that also cause option substitution. Sometimes adding a new option will negatively
affect existing options such that we can say that a pre-existing option no longer
exists post-intervention. Dworkin cites the example of the rise of the automobile.35
The development of cars increased the options available to people – they could
either get around by mass transit, walking, or now, by their car. However, Dworkin
notes that as more and more people took advantage of the new mode of
transportation, funding and maintenance were diverted from mass transit such that
the option of mass transit significantly declined, though it continued to exist.
We should find the automobile addition as opportunity set preserving
because we are comparing the effects of interventions on opportunity sets
immediately before and after the introduction. We are not interested in the natural
evolution of opportunity sets over time. Thus, immediately post-automobile
35 Gerald Dworkin, The Theory and Practice of Autonomy (Cambridge: Cambridge University Press, 1988) at 72.
33
introduction, all pre-existing options continued on as they were pre-intervention.
The long-term effects of the addition can also not necessarily be traced to the
introduction of the car as opposed to other actions by other actors. That is not to say
that adding an option cannot have immediate substitution-like effects on existing
options which we would have to evaluate independently, but Dworkin’s automobile
example is not such a case.
As well, adding options may satisfy a libertarian constraint but may still have
implications for choosers’ autonomy. A liberty-minded person would generally
prefer more options to less. Indeed, this is how most neoclassical economists and
liberals think: even if unwanted, more options can just be disregarded by the
chooser if they wish. However, as behavioural insights research has shown, more
options have a greater impact on choosing than simply background noise to be
tuned-out. Choice overload, ego depletion and decision paralysis are very real risks
posed by having too many choices.36 While the pre-existing options a chooser faces
may be generally unaltered by the addition of more options, the burden of choosing
36 Ego depletion is the theory that decision-making willpower is a diminishing resource, such that, for example, later in the day individuals have less willpower because of the cumulative decisions made earlier. See e.g. Roy F Baumeister, “Ego Depletion and Self-Regulation Failure: A Resource Model of Self-Control” (2003) 27:2 Alcohol Clin Exp Res 281. Choice overload occurs when there are so many options that individuals cannot cognitively process those options well so they resort to biased or heuristic-based decision-making or choice avoidance. See e.g. Liat Hadar & Sanjay Sood, “When Knowledge is Demotivating: Subjective Knowledge and Choice Overload” (2014) 25:9 Psychol Sci 1739.
34
those same options has increased. This may have autonomy implications and we
will explore this possibility in more detail in subsequent chapters.
2.6. Perception of Options
Thus far we have eliminated option eliminations and substitutions from the
realm of nudging as they fail to preserve choosers’ opportunity sets. However, as we
will see, many other interventions work not by altering actual opportunity sets, but
by altering how those opportunity sets are subjectively perceived by choosers.
Interventions that alter the perception of choices will generally satisfy the liberty of
choice constraint because the constraint is focused on objective opportunity sets, not
subjective opportunity sets. Choice architecture can alter choosers’ perception of
their choices by affecting choosers’ awareness of their options, or affecting how
choosers’ comprehend or understand their options.
There are two ways of talking about opportunity sets – objectively and
subjectively. If an option is practically removed from an opportunity set (e.g. a
restaurant no longer offers a cheeseburger), then the objective opportunity set has
changed – it affects all real and potential choosers. Likewise, for increases in costs as
we will see in Chapter 3.
But we can also analyze opportunity sets subjectively – how do opportunity
sets appear to choosers. Because individuals have different cognitive abilities and
35
biases, opportunity sets can appear differently to different choosers. Some
interventions will alter how the opportunity set appears to some choosers, while not
changing it for others. The constant theme is that actual options are not affected –
they are not eliminated or substituted or subject to changing costs in any universal,
objective sense. Note that from an autonomy perspective, we may be just as
concerned about nudges that make choosers feel they can no longer select a certain
option as we are about interventions that objectively eliminate the same option, but
from the liberty of choice perspective, we must reject the distinction.
Let’s begin by revisiting a case study of an intervention labeled a classic
nudge – the arrangement of food in a cafeteria. Sunstein and Thaler use this
example as their opening case for nudging in their popular book, Nudge, since it
preserves the opportunity set and therefore satisfies the liberty of choice constraint
to be a nudge.
In this example, a cafeteria manager must decide how to arrange the order of
food products in a cafeteria buffet line. Sunstein and Thaler note that there are
several options she can adopt, one of which is to place the healthier items first in
line. The goal of such a move is to increase the amount of healthy food customers
purchase relative to the unhealthy food, which is consequently moved to the end of
the line. This reflects behavioural insights research that shows that people will fill
36
their trays with more of whatever they encounter first.37 If it’s the cinnamon buns,
they will purchase more of those relative to the bananas located at the end of the
line.
This kind of behaviour runs counter to the predictions of neoclassical
economics. Instead, we should all be fully informed of all the options (after all, they
all right in front of us), and able to rationally evaluate all the choices before
deciding, instead of choosing sequentially and then being stuck with sub-optimal
choices just because we encountered them first. But that is not how we behave in
practice. We perceive the opportunity set differently than a fully rational, Homo
Economicus actor. We are just so myopically focused on what is right in front of us
that we can’t plan ahead. So we perceive the opportunity set to be much smaller
than it really is. We believe we are choosing between apples and bananas, when we
are really choosing between apples, bananas, and cinnamon buns.
Rearranging the aisle in a healthy-first format, or any other manner, appears
to meet the liberty of choice constraint. We can clearly see that the opportunity set is
preserved – the nudge neither eliminates existing options, alters existing options
through the increase of opportunity costs (Chapter 3), nor does it even add options
to the same opportunity set. We fill up our plates early because by the time we are
37 See e.g. Richard H Thaler & Cass R Sunstein, “Libertarian Paternalism” (2003) 93:2 Am Econ Rev 175; Paul Rozin et al, “Nudge to nobesity I: Minor changes in accessibility decrease food intake” (2011) 6:4 Judgm Decis Mak 323.
37
fully aware of the extent of the opportunity set we face our trays or plates are
already full. It is simply too late – we have already decided.
Another interpretation of cafeteria patrons’ behaviour is that it is the result of
indifference.38 Individuals don’t know what they want to eat when they go to a
cafeteria, just that they want to eat. Their specific food preferences are formed only
upon seeing the items. And it so happens that whatever food they see first, they
tend to prefer. This reasoning reinforces the fact that the intervention only alters
how choosers perceive the opportunity set, not the objective opportunity set itself. If
an opportunity set is presented over time, like the cafeteria example, indifferent
choosers may prefer those options presented earlier. This has nothing to do with
eliminating options or increasing opportunity costs – it’s all about perception. The
opportunity set itself has not changed in the least.
But making choosers aware of options is not always an either-or proposition;
it is a comparative analysis. In such cases, we can talk more of interventions that
increase the salience of certain choices to choosers. There are several salience-related
effects that interventions can have. Figure 3 attempts to visualize these below.
38 Mira Fischer & Sebastian Lotz, "Is Soft Paternalism Ethically Legitimate? – The Relevance of Psychological Processes for the Assessment of Nudge-Based Policies" (2014) No 05-02, Cologne Graduate School Working Paper Series, Cologne Graduate School in Management, Economics and Social Sciences, online: <http://EconPapers.repec.org/RePEc:cgr:cgsser:05-02> at 9.
38
Figure 3 – Salience Effects on Opportunity Set Perception
# Salience Effect Perceived Opportunity set
Absent Intervention
Perceived Opportunity set with Salience Intervention
1 Absolute increase in salience/awareness
A
(B not perceived) A B
2 Absolute decrease in salience/awareness A B
A
(B not perceived)
3 Relative increase in salience A B A B
4 Relative decrease in salience A B A B
5 Relative increase & decrease
in salience A B A B
6 Increase in comprehension A B
7 Selective decrease in
comprehension A B
As Figure 3 shows, salience or awareness effects can take several forms. Note
that the figure here shows how the opportunity set is perceived, since objectively the
opportunity set is the same in each example – A and B with no new opportunity
39
costs, eliminations or option-eliminating substitutions. We will consider these
perception effects sequentially.
2.6.1. Changing Salience by Affecting Option Awareness (#1-2)
Many interventions, including a large part of basic marketing and advertising
efforts, are effective because they make choosers aware of a choice they might not
have otherwise known they face. An intervention in the first row of Figure 3 is much
like advertising a new product or service – it’s objectively part of the opportunity set
but choosers just don’t know it is part of their opportunity set.
Governmental advertisements attempt to work on this problem, for example
by making choosers aware of government-provided services that they may not
know they have. For example, not everyone is aware that they can register to be an
organ donor at beadonor.ca. By informing them of this option, we have added an
option to their subjective, perceived opportunity set, even though objectively it’s
been there all along (or at least since the site was established).
A similar effect is had by reminding choosers that they have an option.
Sometimes options are simply not visible due to the realities of human behaviour,
which at times can be pretty mindless. If you are eating a bowl of popcorn, each
handful you toss into your mouth is a choice, even though it doesn’t feel like it. An
intervention may attempt to insert a decision point by interrupting consumption at
40
certain intervals that has the effect of asking choosers if they really want to continue
to eat that much popcorn. For example, smaller-sized bags or bite-sized treats works
as an added decision point. With these kinds of choice architecture interventions, it
is important that they do not go too far and impose extra opportunity costs as well.
The flip side is also a proven strategy – making choosers unaware of options
that continue to form part of the objective opportunity set. Recently, fast food chain
McDonald’s intentionally removed sodas from menus and marketing materials for
its (child-centered) Happy Meal.39 The option technically remained – patrons could
still choose a soda with their Happy Meal (as opposed to juice, milk or water) – but
the awareness of that option was decreased by its removal from the menus. This led
to a substantial reduction of people choosing the soda option from 56% to 48% in its
first year, with further declines expected.40 This kind of move may have autonomy
implications that we will address in Chapter 6. For example, can we really say that
someone can really select an option if they are not even aware of it?
Interventions that make choosers more aware of an option they did not know
they had, or less aware of an option they previously perceived, seem to meet the
functional liberty of choice criteria for nudging. The options within the opportunity
39 Candice Choi, “Fewer McDonald’ s customers ordering soda with Happy Meals”, Consumerist.com (25 June 2015), online: <http://consumerist.com/2015/06/25/mcdonalds-fewer-happy-meals-are-being-ordered-with-sodas/>. 40 Ibid.
41
set themselves weren’t objectively changed at all – there were no added opportunity
costs (monetary or non-monetary) to choosing one particular option over another.
Nor were any options eliminated. Therefore, all pre-existing options have been
preserved.
2.6.2. Relative Changes in Salience (#3-5)
Salience is not just about informing choosers of their options, it is also about
making information about options more relevant to choosers. The manner of
presentation then becomes important. The third row in Figure 3 shows the effect of
interventions that increase the salience of some options, without targeting the
salience of other options. Likewise, the fourth row shows the effect of decreasing the
salience of some options while leaving the others the same. The fifth row shows a
combination effect – increasing the salience of option A while also decreasing the
salience of option B.
The combinatory effect is generally more successful than simply affecting one
of the options in a set, and mirrors what occurs in Sunstein and Thaler’s cafeteria
example. By moving the healthy food to the front of the line, the salience of that
food to patrons is increased relative to the salience of other foods. But since the
cafeteria organization is linear and finite, such a move also necessitates moving the
unhealthy food to the spot vacated by the healthy food at the back of the line. This
42
decreases the salience of the unhealthy food relative to the other foods as well.
Therefore, this intervention is a perfect example of one that preserves choosers’
opportunity sets, while altering their perception of the opportunity set by both
increasing and decreasing the relative salience of certain choices.
However, not all interventions will automatically have this dual effect, as it
depends on their nature. The presentation of items in the cafeteria was constrained
by its linear design and space limitations. Any change in the placement of one item
has to have effects on the placement of other items. Other salience interventions are
not so constrained and may have only the salience of some options affected. The
McDonald’s menu strategy described above is a good example of the latter – it
reduced the salience of the soda option with a Happy Meal by removing it from the
listed menu – without increasing the salience of other options – they remained as
they were on the menu.
2.6.3. Option Comprehension (#6-7)
In addition to affecting how choosers perceive the opportunity set vis-à-vis
awareness and salience, interventions can also affect how choosers understand their
options. Choosers may be fully aware of their options but unable to comprehend them,
causing them not to choose at all, or to stick with the default or recommended
option. By altering how options are presented to choosers or their relevance,
43
interventions can affect choosers’ understanding of their options and thus their
choices. Humans prefer information to be presented in graphical or visual format,
we also prefer stories, and are generally optimistic (optimism bias). Option
presentations that account for those behavioural insights will not only be successful
but also preserve choosers’ opportunity sets in line with the liberty of choice
constraint of nudging.
An excellent example of this type of perception-based intervention is the
framing of survival statistics by physicians in informed consent discussions with
patients. Part of the physician’s duty when discussing treatment options with
patients is to inform them of the risks of the procedure. Physicians can frame the
options in two ways: they can tell patients they have a 10% risk of mortality or they
can tell them they have a 90% chance of survival.41 Both presentation methods are
mathematically identical. The objective opportunity set will always be the same:
consent with the risks of treatment or not consent with the risks of not undergoing
the procedure. How the physician frames the information will not change that
reality.
But the framing method selected does change the options that patients select –
respondents consent more often if the information is framed positively as a chance
41 McNeil et al, supra note 23.
44
of survival.42 Framing works because it uses the affect heuristic – a decision-making
shortcut that takes how we emotionally feel about a choice to influence how we
weigh the costs and benefits. Presenting information about survival primes our
decision-making to be biased toward survival – it gives us a sense of hope. We have
a good feeling when thinking about survival; not so much when thinking about
death. It’s also a manifestation of our loss aversion – that humans feel losses more
greatly than corresponding gains.43 So even though the outcomes are identical, the
manner of presentation makes patients biased in favour of the survival option.
A similar intervention focuses on making information more understandable
by changing how survival statistics are presented from verbal/numerical to a
graphical representation (survival curves). The latter renders the options more
relevant and understandable to choosers, again without changing the objective
opportunity set.44 After all, patients can still choose whether to consent or not, with
the same opportunity costs associated with the exact same options. Hence the
example in the last row of Figure 3 of the translucent/shaded A and B –
interventions can affect choosers’ clarity or understanding of some options by
making them more (or less) clear to the chooser.
42 See e.g. Sammy Almashat et al, “Framing effect debiasing in medical decision making” (2008) 71:1 Patient Educ Couns 102.; Amos Tversky & Daniel Kahneman, “Rational Choice and the Framing of Decisions” (1986) 59:4 J Bus S251. 43 See e.g. Katrina Armstrong et al, “Effect of Framing as Gain versus Loss on Understanding and Hypothetical Treatment Choices: Survival and Morality Curves” (2002) 22 Med Decis Mak 76. 44 See e.g. Almashat et al, supra note 42.
45
The point of all these examples is that changing how choices are presented
can affect how choosers perceive their opportunity sets. The opportunity sets
themselves objectively remain unchanged so long as no options are eliminated or
new opportunity costs introduced. While there may be autonomy concerns with
interventions that alter the salience of opportunity sets, or increase/decrease the
understanding or relevance of choices to choosers (issues we will explore in Part II),
such interventions appear to preserve opportunity sets and therefore meet the
liberty of choice constraint of nudging.
2.7. Conclusion
In summary, we can easily determine the effects of some choice architectural
interventions on opportunity sets so as to include/exclude them from our
understanding of nudges. Interventions that eliminate pre-existing options or
substitute new options while eliminating old options fail to satisfy the liberty of
choice constraint, and thus do not count as nudges. On the other hand, adding
options or altering the subjective perception of options clearly satisfies the liberty of
choice constraint rendering the applicable interventions nudges. However, whether
those nudges are ethical is a question to be answered in Part II.
46
Chapter 3
The Hard Cases of Opportunity Set Preservation: Costs
3.1. Introduction
Categorizing choice eliminations or additions as preservative of opportunity
sets (i.e. liberty of choice constraint satisfying) was clearly an easy task as few
distinctions were required. Many nudges, however, affect opportunity sets in a
subtler manner. The next category of potential nudge candidates concerns those
interventions that alter existing choices by the introduction or alteration of costs –
increasing or reducing the costs of certain choices within the same opportunity set.
The opportunity set has the same number of choices, and the same choices
themselves, but the character of some of those same choices has been altered.
This chapter will evaluate whether interventions that increase the costs of pre-
existing choices within the same opportunity set satisfy the liberty of choice
constraint. A key determinant is whether the costs affected by an intervention can be
considered “opportunity costs” so as to implicate some finite, depletable resource of
a chooser governed by a budget constraint. An increase in opportunity costs
necessarily means that there is a corresponding reduction in opportunities
elsewhere (either within the same opportunity set or in some other opportunity set),
and thus a chooser’s liberty of choice would not be preserved by the intervention.
47
Other cost-related issues will be examined, including non-monetary (psychological)
costs, increases vs. reductions in costs, as well as implications for interventions that
shift the burden of costs, such as defaults.
3.2. Costs & Construing the Opportunity Set
The first step in deciding whether a certain change in costs preserves an
opportunity set lies in how we construe the opportunity set. Take the example of a
person deciding whether or not to smoke. If we construe the opportunity set
broadly – e.g. simply between the options to smoke or not to smoke – any choice
architecture that does not eliminate the smoking option would qualify as a nudge
under the idea of opportunity set preservation. Graphic warning labels or hidden
retail displays do not take away a person’s option of whether to smoke or not.
Likewise, tobacco sin taxes (even substantial ones as they often are) would preserve
that same opportunity set. Therefore, on a broad construal, those interventions
would qualify as nudges
Construing a smoker’s opportunity set as simply between smoking and not
smoking ensures that any non-eliminatory interventions will virtually always satisfy
the liberty of choice constraint. However, such a broad construal does not capture
the nuance of opportunity sets and how seemingly small tweaks to the character of
some choices (the modus operandi of nudging) fail to preserve substantially the same
48
set of options. Indeed, the findings of behavioural insights prove that in many cases,
it is the nuanced different characteristics of what appears to be the same options that
ultimately have the largest impact on individual decision-making.
An opportunity set should be construed as narrowly as possible to crystallize
the real options that individuals face. To do so, we must factor in all of the costs
associated with each option. We cannot accurately say whether an intervention
reduces or preserves a chooser’s opportunity set without a full appreciation of what
options a chooser faces, and that necessarily involves factoring in costs. Returning to
the smoker’s opportunity set, a narrower construal would require comparing the
costs of smoking pre-intervention with the costs of smoking post-intervention to
determine if the opportunity set has been reduced.
When we use the term “costs” here we mean all costs associated with an
option; price is only one component of those costs. Such non-price costs can include
increased monetary costs through higher taxes, transaction costs and potentially
non-monetary costs such inertia-related costs through restricting the sale of tobacco
to fewer outlets, discovery costs through restricting advertising and marketing,
social stigma costs, and finally psychological/emotional costs such as disgust
generated through mandatory graphic packaging. While such costs might be
belittled as trivial, if behavioural insights have taught us anything, it’s that it is often
the so-called trivial costs and factors that get washed out of neoclassical economic
49
predictions that are the source of our behavioural deviations from those very
predictions.
We will consider the implications of changing monetary costs for opportunity
set preservation before moving on to an examination of the myriad non-monetary
costs that can be found within opportunity sets.
3.3. Monetary Costs
Sunstein and Thaler exclude significant economic measures such as taxes
from their definition of nudges because they recognize that the costs of options are
an integral shaper of opportunity sets. Having the same two options but increasing
the price of one option while leaving the other unchanged does not preserve the
opportunity set.
There is a rationale for this linked with autonomy and liberty. If we are to be
liberty-minded persons, a choosers’ liberty is enhanced by having more options
available to them. Likewise, a chooser’s liberty is decreased by closing off options.
As we will see, because of opportunity costs and budget constraints, certain changes
in the cost of some choices leave choosers with fewer practical options available to
them in their opportunity sets.
However, not all cost increases have this effect, particularly when it comes to
non-monetary costs. The difference comes down to a distinction between costs
50
generally and opportunity costs.45 The latter are those monetary or non-monetary
costs that deplete a finite resource such that less of the resource is available for
“spending” on other options or opportunity sets. Not all costs are opportunity costs
– some draw on an infinite resource such that the availability of that resource for
other choices is unaffected.
There are two reasons to begin our analysis with monetary costs before
shifting to non-monetary costs. First, even though Sunstein and Thaler explicitly
disavow material economic changes from their nudge definition, many others
include economic interventions as examples of nudges.46 In fact, Sunstein and Thaler
counteract their own definition by providing examples of nudges that rely on
changes to monetary costs of options. For example, they propose carbon taxes
associated with cap and trade emission programs and justify it as a nudge on the
grounds that, “this basic approach is compatible with libertarian paternalism [a
form of nudging] because people can avoid paying the tax by not creating the
pollution.”47 Second, analyzing monetary costs will help to illuminate the key
concepts that we will then apply to the more difficult cases of non-monetary cost
nudges.
45 Mark White does an excellent job of summarizing this point: White, supra note 6 at 4-6. 46 See e.g. J S Blumenthal-Barby & Hadley Burroughs, “Seeking Better Health Care Outcomes: The Ethics of Using the ‘Nudge’” (2012) 12:2 Am J Bioeth 1; Yashar Saghai, “Salvaging the concept of nudge” (2013) 39:8 J Med Ethics 487. 47 Thaler & Sunstein, "Nudge", supra note 1 at 188.
51
3.3.1. Local Budget Constraints
This and the next subsection will show that there are two relevant reasons
why increasing the monetary cost of an option fails to preserve a chooser’s
opportunity set. The first goes to a chooser’s local (opportunity set-specific) budget
constraint. A chooser may simply not be able to allocate more money to the same
option (because they don’t have any more money practically available), thus
effectively removing that option from their opportunity set. The second reason goes
to global budget constraints and opportunity costs – that even if a chooser were
willing or able to spend more money on the same option, the increased cost
effectively reduces the resources (e.g. money) available for application in other
opportunity sets because the chooser is under a global budget constraint flowing
from the finite nature of the resource.
We can begin to explore this idea of a “budget constraint” using the example
of Mayor Bloomberg’s so-called “Big Gulp Ban” in New York City – an intervention
that is claimed both as a nudge,48 and not a nudge.49 The move, touted as a public
48 Conly, supra note 24. 49 Cass R Sunstein, “Was Mayor Bloomberg a Nanny?” (2014) 1:May Harvard Public Heal Rev.
52
health anti-obesity effort, prohibited certain establishments (such as movie theatres)
from selling “unhealthy” beverages in containers larger than sixteen ounces.50
Let’s imagine a hypothetical movie theatre patron before and after the
regulation. Prior to the regulation, they would purchase their giant 32-ounce
beverage for, let’s say for ease of analysis, $4.00. After the regulation, they can no
longer purchase a single 32-ounce beverage. But they can purchase the same amount
of liquid, just in two smaller containers. Now, let’s say that our hypothetical patron
is quite thirsty and still desires to purchase 32 ounces of soda. Let’s assume that the
smaller 16-ounce containers cost $2.50 each, so suddenly to satisfy his preference for
32-ounces of soda, our movie buff must now pay $5.00. On Sarah Conly’s view, the
opportunity set has not changed since there is no strict, practical prohibition that
prevents our patron from purchasing whatever amount of soda he wants (leaving
aside the non-monetary costs of lugging around two containers of soda).51 But if we
consider that our cinephile may be under a fixed budget constraint, Conly’s
seemingly simple conclusion fails.
Let’s assume for the sake of simplicity that our moviegoer only had four
dollars in his pocket to spend on snacks. That four dollars constitutes his budget
50 Micheal M Grynbaum, “Health Panel Approves Restriction on Sale of Large Sugary Drinks”, The New York Times (13 September 2012) online: <http://www.nytimes.com/2012/09/14/nyregion/health-board-approves-bloombergs-soda-ban.html?_r=0>. 51 Conly, supra note 24.
53
constraint. Prior to the Big Gulp Ban, he could satisfy his beverage preference, but
just, with that $4.00. After the ban, he could no longer satisfy his preference for 32
ounces of beverage with the same money he had on hand. He would be short by
$1.00. Therefore, for him, his opportunity set has changed as a result of the
intervention – he can no longer give effect to his preference for 32 ounces of soda
under his chooser-specific budget constraint.
The option choosers face is not simply whether to drink as much soda as
possible. It is to drink as much soda as they desire within their budget constraint.
Eliminating the volume-discounted, 32-ounce size effectively raises the price for the
same volume. While the option of drinking 32 ounces superficially remains, because
we have to account for a chooser’s budget constraint (our patron only has $4.00 in
his pocket), that option (the 32 ounces of total soda) is effectively eliminated.
Introducing the concept of a budget constraint allows us to conceptualize an
opportunity set in a far more nuanced, and realistic, manner. Budget constraints add
much-needed context to opportunity set analysis. Viewing an opportunity set
without consideration of a chooser’s budget constraint removes much of the
usefulness of the analysis. Everyone in the world is theoretically “free” to purchase
mega-yachts or private jets, but those options aren’t really in our opportunity set if
you don’t have the net worth to purchase them. Most of our opportunity sets are far
more modest, but only because we are subject to the reality of budget constraints
54
(usually set by a combination of our incomes, savings, and credit access). This idea
of budget constraint gives life to our use of the term “option” to describe the
contents of opportunity sets. It is simply not an option for you to purchase a yacht if
you do not have enough money.
The use of the phrase “budget constraints” here does not necessarily mean
that a given chooser has set themselves an actual, accounting budget to guide their
choices. Most people’s budgets are far more informal. We are willing to purchase
certain products or services at certain monetary costs, and if those costs increase
enough, that willingness disappears. The reason we have local budget constraints is
because money is finite – spending more of it on one product or service means there
is less of it available for other goods or services. Thus, we can think of monetary
costs as always being opportunity costs – changes to those costs affect other
opportunity sets. Any increase in the cost of one product, even one small enough to
have no impact on demand for that product, must, due to the scarcity of money,
impact other choices. Therefore, we must also consider a chooser’s budget constraint
from a global perspective.
3.3.2. Global Budget Constraints – Opportunity Costs
A person’s global budget is far more rigid than their option-specific budget
and generally finite. We only have x dollars (generally reflective of our income and
55
access to credit) to spend on all our consumption choices. Let’s assume that a
hypothetical chooser is very frugal and he manages to survive each day on a $3.00
pizza, which to keep it simple, we’ll also take to be the maximum price that he can
allocate to food for the day out of his $10 income. Of course our chooser has more
money – $7 out of his $10 daily income -- but he currently uses it for other goods or
services (including, potentially, savings). Now, let’s assume the government, in an
effort to encourage healthful behaviour, targets unhealthy foods such as pizza with
an additional tax, bringing the cost of pizza from $3 to $4. Note that prices are just
one component of costs, including taxes (i.e. taxes don’t increase prices, they
increase costs).
Since our chooser’s income is finite, any increase in costs without a
concomitant increase in income requires that our chooser make a more complicated
choice. They have two options: they can forgo the pizza and spend the whole $10 on
other goods and services; or they can keep the pizza purchase and have only $6 (as
opposed to $7) to spend on other goods and services. The option to use the whole
$10 on other goods and services existed pre-tax intervention so that in itself does not
reduce the chooser’s opportunity set – that option is preserved.
But the intervention’s effect on the second option does alter and reduce the
opportunity set. It is no secret that more money purchases more goods/services: $7
buys more choices than $6. By increasing the price of the pizza, this causes a
56
reduction in the remaining funds available for other choices (since money is finite).
Objectors, like Conly, would be right to say that our hypothetical chooser can still
purchase the pizza if they want to, but due to the increased cost, that now requires
that they forgo some other choice. Thus, their global opportunity set is reduced (i.e.
not preserved) and we can think of the cost in question as an opportunity cost.
If you have a fixed budget (i.e. due to a finite resource), higher costs of
choosing any one option means that other options, once “choosable,” may no longer
be so to the same degree. For simplicity sake, let’s say our global choice options
consist of A, B, and C. If we have $100 of money and allocate one-third to each
option, we can purchase roughly $33 worth of option A, B, and C. If, as a result of
government intervention (e.g. a nudge), the cost for the same amount of option A
increases to $40, and assuming we still want to purchase the same amount of A, then
our consumption of B and/or C necessarily has to decrease. We can no longer
purchase the same number of B and C options. Therefore, any increase in
opportunity costs, even if there is a negligible impact on localized demand for
particular options, effectively reduces a person’s global opportunity set.
To reiterate then, increasing the costs of certain options may have no impact
on a person choosing those options. However, that cannot reasonably be the
demarcation line for satisfying the liberty of choice constraint. In this instance, we
57
must look more broadly, accounting for a chooser’s global budget constraint; we
must determine if the costs in question are opportunity costs.
Therefore, because money is a finite resource, we can think of monetary costs
as always being opportunity costs. Thus, we can exclude from nudging any increase
in the monetary cost of choosing any particular option (i.e. an increase in price, a tax
or an increase in the cost of transacting for a good at some given price) because it
fails to preserve choosers’ local or global opportunity sets.
3.4. Non-Monetary Costs
Thus far, we have only considered opportunity costs and budget constraints
as they relate to the monetary cost of options. Those ideas get murkier when we
consider all the non-monetary “costs” that accompany the options we face. When
you are deciding between apples for $1.00 and bananas for $1.10, you compare the
prices (monetary costs) of those choices. But there are other non-monetary costs you
are paying to make that choice – the effort and energy of going to the store; the time
you spend grocery shopping you cannot spend on something else, etc. These are
added costs that you must bear to express your preference for either bananas or
apples.
58
Or consider the example of graphic warning labels on cigarette packages.52
Such an intervention could be considered to impose significant non-monetary costs
on smokers, namely the psychological53 costs of fear, disgust, and social costs of
increased social opprobrium, etc. by forcing smokers to look at and display hideous
graphic images. Those costs are all non-monetary and work mostly on choosers’
emotions.
The relevant question for our nudge identification effort is: are these non-
monetary costs opportunity costs, such that any increase reduces a person’s local or
global opportunity sets? As we will see, the opportunity cost determination turns on
whether a finite resource is required to satisfy the cost (i.e. a budget constraint is
implicated).
3.4.1. Opportunity Costs & Finite Non-Monetary Resources
In order to identify interventions that affect non-monetary costs as nudges,
we must determine what non-monetary characteristics of options qualifies as
opportunity costs, and which are simply costs (i.e. non-opportunity costs). The
strongest tool to make this distinction is the idea of scarcity or finiteness – that non-
52 See e.g. Christine Jolls & Cass R Sunstein, “Debiasing Through Law” (2006) 35:1 J Legal Stud 199 at 215; Sunstein, "Why Nudge", supra note 26 at 48. 53 Sunstein and others often prefer to use the term “psychic costs.” This is misleading and confusing to the lay reader. Instead, the phrase “psychological costs” more accurately captures how these types of costs function. We will use “psychological costs” going forward.
59
monetary resources that are finite or scarce (that is, they can be depleted) are subject
to a similar budget constraint analyses as with monetary costs. If so, we can label
those costs “opportunity costs” because they alter the opportunity set that choosers
face. On the other hand, costs that are not subject to a budget constraint are not
opportunity costs.
Monetary costs are clearly opportunity costs – they affect the resources
available to choosers to spend on other options. The same goes for certain non-
monetary resources of choosers – we only have so much time, energy, etc. to expend
throughout any given day. If an intervention demands that a chooser “spend” more
time or energy, then there is less available for other activities and choices. Therefore,
using our global budget constraint concept, any increase in the costs of one option
reduces the overall opportunity set available to our chooser.
We can apply this reasoning to organ donation where time acts as an
opportunity cost. Viewing our chooser’s global budget as one-day’s worth of time
(24 hours), and assuming that they have a preference not to donate their organs,
changing the default requires extra time in order to satisfy their same preference. In
some jurisdictions, such as Ontario, a person can register their wishes online, but in
many others (and for those who do not have internet access) they must either
indicate their preference in-person when renewing their driver’s licence or they
60
have to print out a form, fill it in and mail it (e.g. Nova Scotia).54 The latter
requirements can reasonably be said to impose additional time costs of five to ten
minutes. Since not donating pre-intervention required no time expenditure, if the
jurisdictions above switched to a presumed consent system while maintaining the
opt-in/opt-out procedures, changing the default would reduce the time available for
other daily choices by five to ten minutes. A further example: in Wales, which
switched to a presumed consent (opt-out) system in December 2015, if you do not
have access to the internet, you must call to register your wishes to opt-out.55 Even
completing the online form likely takes at least five minutes – non-trivial time that
could have been spent on other choices.
As opportunity costs get closer to zero, the effect on the opportunity set
correspondingly decreases. In the organ donation case, if opportunity costs are close
to Sunstein and Thaler’s ideal opt-out of “one-click” then such costs are, reasonably
considered, trivial. Saving an extra half-second will not, in any reasonable sense,
affect a chooser’s global opportunity set. On a more general level, where
opportunity costs can be considered “minimal” will be dependent on the individual
characteristics of the chooser. However, for our purposes, we do not need to
54 Service Ontario, “Organ and tissue donor registration,” online: <https://www.ontario.ca/page/organ-and-tissue-donor-registration#section-1>; Legacy of Life, “How to Be a Donor,” online: <http://www.legacyoflife.ns.ca/how_to_be_a_donor/index.html>. 55 Organ Donation Wales, online: <http://organdonationwales.org/?lang=en>.
61
perfectly quantify the cost – to determine whether an intervention is a nudge or not,
we only need to determine whether a cost is an opportunity cost and whether it is
above some trivial level.
On the other hand, interventions that alter the character of choices in such a
way as to demand more of choosers’ infinite psychological resources (e.g. fear,
disgust), seems to preserve the opportunity set in a fashion consistent with the
liberty of choice constraint because the costs in that case are not opportunity costs.
While the intervention may be said to require “more” of whatever resource is at
play, since the expenditure of that resource in that particular choice problem does
not affect what can be expended on other options or opportunity sets, there is no
impact on a chooser’s global opportunity set.
3.4.2. Identifying Finite vs. Infinite Non-Monetary Resources
There are several factors that can help guide us in determining whether non-
monetary costs relate to finite or infinite resources. This section will explore those
factors using a series of case studies from difficult borderline cases of potential
nudge interventions.
First, we can turn to tobacco-related regulations. Schnellenbach has argued
that banning indoor smoking is, in addition to a protection of harm measure for
non-smokers, a way of imposing a cost on smokers – they have to go to the trouble
62
of going outside to smoke that extra cigarette.56 Going to extra trouble is really about
inconvenience, but inconvenience itself is not a cost. Instead, it reflects other costs, in
this case spending extra time and energy on entering/exiting buildings, and
depending on the weather, it may impose discomfort on having to be outside as
opposed to indoors. There are others at play as well, but let’s focus on just these
three – time, energy, and discomfort.
The first hint that some of these costs may affect finite or scarce resources is
our use of the term “spend.” Spending something generally requires that there is a
bank or store of a resource from which to draw down or deplete. Intuitively, when
we say spend, we generally refer to finite resources. After all, can you really spend an
infinite resource?
Similar to the term “spend,” the presence of a tradeoff, indicated by the
intuitive understanding of “instead of,” also implies a finite, depletable resource at
stake. Finite resources always involve choice tradeoffs. Since you cannot spend
unlimited amounts on every choice, spending resources on one choice depletes the
amount available for other choices.
We have already considered time. In our smoking example, if it takes our
chooser an extra five minutes each time they have to leave the office to smoke, and
56 Jan Schnellenbach, “Nudges and norms: On the political economy of soft paternalism” (2012) 28 Eur J Polit Econ 266 at 268.
63
they smoke just three times a day (a conservative assumption), that’s an extra fifteen
minutes each workday. That fifteen minutes a day our smoker has “spent,” as
compared to their pre-intervention baseline, cannot be spent on other activities,
including, potentially, smoking itself. The time is gone and our chooser cannot get
more of it. Thus, we can think of time-related costs as opportunity costs – any
increase cannot definitionally count as a nudge.57
Our effort is also finite – there is only so much effort we can expend on a
series of tasks throughout the day. The reality of status quo bias and inertia lends
further credence to this view – we instinctively prefer to stay at rest, in effort-
preservation mode.58 We want to preserve our effort in part because it is limited.
Furthermore, spending effort on any one task means we have less of it left to expend
on other tasks. An intervention that requires increased effort for the same options
may qualify as imposing a non-monetary opportunity cost. Such an intervention
would thus not count as a nudge because it fails to preserve choosers’ local or global
opportunity sets.
You may have noticed a small difference between effort and time. Time is
universal and we all share the same resource bank (24 hours in a day). Effort on the
57 This argument assumes that resources (money, time, etc.) are always employed or otherwise available to serve a purpose, even a future purpose. 58 See e.g. Brigitte C Madrian & Dennis F Shea, “The Power of Suggestion: Inertia in 401(k) Participation and Savings Behavior” (2001) 66:4 Q J Econ; Samuel Johnson, “Status Quo Bias in Decision Making” (1988) 1 J Risk Uncertain 7.
64
other hand, depends very much on the individual chooser. Some people naturally
have larger “effort stores” than others.
But, for our purposes, cost proportionality is irrelevant. We are only
interested in whether opportunity sets have been objectively preserved or not.
Therefore, the magnitude of costs does not matter. Such a determination comes
down to the nature of the resources in question – they are either finite or infinite.
The same reasoning applies for why we do not factor in the size of a chooser’s
monetary budget constraint. A price increase of even 25-cents on a given choice
equally affects the determination of opportunity set preservation of a poverty-
stricken person and a billionaire: there is 25-cents less money available for other
choices. It is not the size of the pot of resources that matters: it is the fact that there is
a finite pot, and how much the pot is reduced that matters. Of course, the billionaire
cares less, since the impact of a 25-cent reduction in their budget constraint is trivial,
while for the poor a 25-cent reduction in their budget may greatly affect their other
choices. But the disproportionate impact of added monetary costs does not concern
our conclusion on whether the cost is an opportunity cost.
65
3.4.2.1. Case Study: Graphic Tobacco Warning Labels
Returning again to the area of tobacco regulation, the example of graphic
warning labels on cigarette packages is illustrative of the difficulty of determining
whether non-monetary costs qualify as opportunity costs.
Warning labels are designed to inform smokers of the dangers of smoking.
But information provision comes in many forms. They could be lengthy written
statements (akin to the small booklet that comes with many off-the-shelf drugs) or
voice-overs during advertisements. Or they could be graphical representations of
risk, or be images similar to the graphic labels on tobacco packages.
But we must also acknowledge that due to human behavioural biases and
quirks, there may no longer be any such thing as simple information provision.
Providing choosers with only objective, outcome-neutral information will nearly
always involve some form of persuasive influencing (intentional or otherwise).
Simply providing choosers with information about the choices available to them
does not objectively alter the opportunity set. While such information provision may
alter how choosers perceive their opportunity set, choosers becoming more aware of
the content and consequences of various options does not, without more, diminish
the objective opportunity set the chooser faces.
66
Behavioural insights show that individuals grossly underestimate their risks
of smoking due, in part, to optimism bias.59 In other words, smokers simply don’t
think they will get cancer – they estimate their own health risks as less than the
average smoker.60 The warning labels are designed to counter this bias, but not by
just informing choosers, but by appealing to other biases and emotional thinking –
instigating feelings of fear and disgust. Thus, these labels appeal not just to a
smoker’s (or potential smoker’s) rational mind to inform them of the consequences
of smoking, but they also appeal to choosers’ less than rational faculties – their
emotions. And they do so by increasing the psychological costs of smoking.
A full and complete description of the choice to smoke without graphic
warning labels (our baseline for choice architecture analysis) might be that it goes
without the disgust, anxiety, or social opprobrium that might attach to the smoking
option when the packaging has extremely graphic images. So it is possible then to
conceive of these psychological and social factors as “costs” that go into a chooser’s
weighing of options.
The cost analysis for non-monetary costs is no different than for monetary
costs. Indeed, graphic warning labels increase costs the same as taxes do; just a
59 Neil D Weinstein, “Unrealistic Optimism About Susceptibility to Health Problems” (1982) 5:4 J Behav Med 441; Neil D Weinstein, “Unrealistic Optimism About Future Life Events” (1980) 39:5 J Pers Soc Psychol 806. 60 FP Mckenna, DM Warburton & M Winwood, “Exploring the limits of optimism: the case of smokers’ decision making” (1993) 84:3 Br J Psychol 389.
67
different type of costs. Instead of paying with money, smokers now have to “pay”
with negative emotions (e.g. disgust). However, if we can consider our emotional
reservoirs to be close to infinite, then the new non-monetary costs imposed by
graphic warning labels are not opportunity costs – they don’t really “pay” at all. If
that is the case, then an increase in non-opportunity costs would mean that
choosers’ opportunity sets were preserved, and the intervention would count as a
nudge.
The key issue then is whether the costs at issue relate to an infinite or finite
resource. The added costs of anxiety and disgust (or smoking in a way that still
looks cool to your friends) charged to smokers by requiring them to view the
inescapable graphic warning labels appear to relate to infinite resources. Feelings of
anxiety or disgust are (alas) seemingly limitless – there really is no budget constraint
applicable to these resources. We would not say, “if I view these labels, I’m not
going to have any disgust left for other choices.” Nor would we say we “spent”
disgust or fear. There is no finite store of disgust or fear to draw down on; they are
not opportunity costs.
Recall that for our analysis, we are only interested in costs to the extent that
they reduce a finite pool of resources – a budget constraint. When the budget
constraint is finite (e.g. money, time, energy), then the resource pool is drawn down
by any increase in opportunity costs. This causes a net global reduction in options
68
due to our global budget constraint. But there is no budget constraint for some
resources because of the nature of the resource (e.g. fear, disgust). In those cases, the
addition of costs or an increase in magnitude of those costs is irrelevant, since
infinite supply cannot, by definition, be reduced.
Therefore, while the psychological costs of choosing to smoke can still be said
to be increased by the addition of the graphic warning labels, the effect of those
added costs on other choices, and therefore a smoker’s global opportunity set, is nil
because those costs affect resources that are in infinite supply. There has been a cost-
based substitution. However, though individuals may prefer not to “pay” disgust,
doing so does not diminish their local or global opportunity sets. The smoker’s
opportunity set is preserved by the intervention – they can still choose to smoke or
not smoke – and while the psychological costs of the former are higher, such an
increase is irrelevant for our opportunity set analysis because no other opportunities
are reduced or eliminated due to the increased in costs (i.e. they are not opportunity
costs). Any increase in non-opportunity costs preserves the opportunity set, and
therefore satisfies the liberty of choice constraint. Any interventions of this nature
can be considered nudges.
69
3.4.2.2. Case Study: Defaults
Another popular nudge example that involves non-monetary costs is that of
altering defaults and opt-outs, specifically in the areas of organ donation consent
and pension plan enrolment.61 Defaults operate by having a particular “choice”
selected for choosers if they (intentionally or unintentionally) “choose” to do
nothing. In its simplest form, defaults are binary yes/no choices: “do you wish to join
the company pension plan?” or “do you consent to donate your organs after your
death?” There are two options given: to join or not to join, and to consent or
withhold consent.
Despite the binary appearance of these questions, there is actually a third
option: do nothing. Failing to answer the question either affirmatively or negatively
is its own form of response. Therefore, choice architects must decide what silence
means. The employer must decide what happens to employees that do not indicate a
choice for either – are they deemed to have joined the plan or are they left out as a
result of their inaction?
61 In general, see: Cass R Sunstein, “Switching the Default Rule” (2002) 77 New York Univ Law Rev 106. Many European countries operate opt-out organ donation consent systems of presumed consent, see: Eric J Johnson & Daniel Goldstein, “Do Defaults Save Lives?” (2003) 302:5649 Science 1338. Thaler is the leading proponent of adopting opt-out systems for employee pension enrolment, see: Richard H Thaler & Shlomo Benartzi, “Save More Tomorrow: Using Behavioral Economics to Increase Employee Saving” (2004) 112:1 J Polit Econ s164.
70
In the pension plan example, in opt-out systems, all employees are deemed to
join the plan meaning that if they do not wish to join they must actively choose not
to do so. If they want to join the plan they do not have to do anything. The same
goes for those that are undecided or simply forget to positively choose. In opt-in
systems, the reverse is true: doing nothing results in not joining the plan. If an
employee wants in, they must affirmatively express that choice through whatever
action is required to join (likely some paperwork).
In the case of organ donation consent, most sub-national jurisdictions in
Canada and the United States operate under opt-in systems.62 Citizens who want to
donate their organs post-mortem must expressly consent to do so using whatever
system their government has devised (this usually involves indicating consent when
renewing government-issued ID such as a driver’s license). If you do nothing, you
are deemed to have not consented. This lumps together those who actually have a
preference not to donate and those who may be undecided or unaware of the
options.
We are concerned with defaults in our opportunity set analysis because
interventions have been proposed that switch or flip the default, and these have
62 Alberto Abadie & Sebastien Gay, “The impact of presumed consent legislation on cadaveric organ donation: A cross-country study” (2006) 25 J Heal Ecnomics 599; Sheldon Zink, Rachel Zeehandelaar & Stacey Wertlieb, “Presumed vs Expressed Consent in the US and Internationally” (2005) 7:9 Ethics J Am Med Assoc.
71
been labeled as nudges that preserve the opportunity set.63 Opt-in systems become
opt-out. Where a chooser previously had to expressly consent to donate their
organs, under the proposed intervention, they would simply have to do nothing and
have their consent deemed or presumed. The same goes for employees that do
nothing about their pension plans: prior to the default intervention such inaction
would lead nowhere, whereas after the intervention it would lead to joining the
pension plan.
Interventions that flip the default rely on the “stickiness” of defaults. Due to
behavioural biases and quirks (status quo bias, inertia), humans predominantly stick
with defaults.64 There are several behavioural factors at play. First, defaults tend to
signal an official recommended or endorsed course of action. For example, in the
case of organ donation, by flipping the default to an opt-out system, the default
signals to the citizenry that the social norm is to donate your organs. But defaults are
also powerful tools because they usually impose costs (possibly opportunity costs)
on those choosers who want to choose contrary to the default option. If you don’t
want to go along with the default (which you can do by doing nothing), there is a
cost to doing so.
63 See e.g. Kyle Powys Whyte et al, “Nudge, Nudge or Shove, Shove—The Right Way for Nudges to Increase the Supply of Donated Cadaver Organs” (2012) 12:2 Am J Bioeth 32. 64 Omri Ben-Shahar & John A E Pottow, “On the Stickiness of Default Rules” (2006) 33 Florida State Univ Law Rev 651.
72
Sunstein and Thaler rightly acknowledge that the ideal opt-out system would
involve a one-click opt-out where the non-monetary costs (opportunity or
otherwise) are as small as possible (e.g. one click on your computer).65 However,
that is rarely, if ever, the case. In most pension opt-out systems, choosers have a
limited window in which to complete a set of somewhat complex paperwork.66
The key question we must address then is whether the non-monetary costs
involve finite or infinite resources? If the latter, then the higher costs are a moot
point – they are not opportunity costs and thus a chooser’s opportunity set is
unaffected by their increase. It seems in most cases that defaults involve imposing
non-monetary costs that involve a combination of opportunity costs and non-
opportunity costs.
If you are a chooser that wants to choose contrary to the official default
position, choosing contrary to the default may involve added psychological costs of
violating a perceived social norm. Humans by nature tend to prefer to go with the
grain: we often exhibit herd mentality. Violating a perceived social norm carries
costs. However, these costs do not clearly involve a finite resource. Instead, it is
likely that, while we prefer, all else being equal, not to violate social norms, doing so
65 Even online, opting-out is rarely just one-click. A person must first navigate to the relevant page, and, at least in the case of organ donation, input a collection of personal details (requiring retrieving a health card, etc.). 66 See e.g. Thaler & Sunstein, "Nudge", supra note 1 at 111.
73
does not deplete some store of anti-social choosing. There is no objective limit on
making such choices. We do not “spend” such psychological processes that we then
cannot spend on other choices. In other words, they are not opportunity costs. There
is no budget constraint of such resources, and therefore the opportunity set is
preserved when it comes to those specific social norm-related costs.
3.5. Reducing vs. Increasing Opportunity Costs
We must now make a distinction between adding and reducing opportunity
costs and their respective impacts on our conclusion regarding opportunity set
preservation. Here is where my definition of nudging departs from that of Sunstein
and Thaler – they state that nudges must not forbid options or significantly alter a
chooser’s monetary incentives. Others echo the sentiment: “when a policy changes
the utility a person derives from an outcome, in order for it to count as a nudge the
change must not be in monetary terms.”67 These descriptions seem to contemplate a
definition of nudging that excludes any change to opportunity costs, regardless of
whether the change is positive or negative. We have already revised our cost
concerns to opportunity costs, but we must now revise it further to say that an
intervention is not a nudge if it increases opportunity costs. However, as this section
67 See e.g. Fischer & Lotz, supra note 38 at 7.
74
will show, a reduction in opportunity costs would count as a nudge because it
satisfies our opportunity set preservation criterion.
As we have seen, increasing the opportunity costs of options has the effect of
reducing a chooser’s opportunity set through the effects of local or global budget
constraints. On the other hand, reducing the opportunity cost of options tends to
have the opposite effect – it actually enhances (and at a minimum, certainly
preserves) a chooser’s opportunity set by freeing finite resources to be applied to
other options/opportunity sets. In that sense, the opportunity set would be
preserved.
Consider if the government were to offer subsidies for healthy eating.68
Instead of paying $1.00 for a bunch of bananas at the supermarket, they now cost,
say, $0.80. Once again we need to use the budget constraint concept to appreciate
how this reduction in price for this one option impacts a chooser’s local or global
opportunity set. Saving $0.20 on bananas opens our chooser to other new options
within their local (i.e. supermarket) and global opportunity sets if we assume, for
simplicity’s sake, that they have a $1.00 budget constraint. Since our chooser has
$0.20 left over, that is money they can spend elsewhere. For example, our shopper
can buy bananas and a cheap tomato as a result of the subsidy.
68 In some jurisdictions, like Canada, this is already effectively done by exempting food from general sales taxes, whereas junk food items do not enjoy the exemption.
75
We can extend the same reasoning to global opportunity sets. Just as with the
local opportunity set, reducing the price of any option within a chooser’s entire
global opportunity set, opens up new options. The savings of $0.20 can be spent
elsewhere – either locally at the supermarket or perhaps our chooser can save the
money. It doesn’t seem like much, but $0.20 a week for a lifetime adds up, and that’s
with just one small case of a reduced cost.
The magnitude of the reduction is not what matters here. Any reduction in
opportunity costs at all will enhance a chooser’s opportunity set, because there is no
amount (at least when it comes to money or other finite resources) that can’t be
saved or used for other expenditures.
This reasoning only applies to that subset of costs considered to be
opportunity costs. As we have already seen, since a budget constraint of a finite
resource is not implicated, non-opportunity costs can be increased without affecting
a chooser’s opportunity set. For the same reason, non-opportunity costs can be
reduced without positively or negatively affecting a chooser’s opportunity set. And
because our standard of evaluation is one of bare opportunity set preservation, a
neutral effect on opportunity sets satisfied the liberty of choice constraint.
Returning to our bananas example, it is theoretically possible to argue that
because we view the opportunity set so narrowly – purchase x bananas for y dollars
– that any change in the variables of x or y results in a non-preserved opportunity
76
set. On a strict interpretation, this is correct, since it aligns with the reasoning we
applied to the Big Gulp Ban – if you increase the price (the y variable) or how much
a chooser can purchase with the same amount of money, we said that the
opportunity set was not strictly preserved.
However, despite appearances, the two cases are not the same. Unlike in the
Big Gulp Ban example, a chooser in our healthy eating scenario can still choose to
pay the higher, original price of $1.00 for bananas. There is nothing (assuming the
grocer accepts overpayment) preventing our chooser from overpaying. On the other
hand, in the Big Gulp Ban case, there is no legal way for the chooser to underpay for
the same beverage volume. Therefore, in the bananas case the original option still
exists (while also opening up other choices due to the now-excess resources), while
in the Big Gulp case, the original option truly no longer exists. Since it is the state of
pre-existing options that determines whether an opportunity set is preserved or not,
once again increasing opportunity costs of finite resources does not preserve the
opportunity set. On the other hand, decreasing opportunity costs of finite resources
enhances a chooser’s opportunity set, thereby satisfying the liberty of choice
constraint to qualify as a nudge (recall as well that a nudge does not definitionally
require the presence of any behavioural anomalies).
77
3.6. Shifting the Burden of Costs
This section will examine interventions that function neither by increasing
nor decreasing opportunity costs, but by shifting which options within an
opportunity set bear those costs. Many default-related interventions do not impose
defaults where none currently exist. Instead, in a pre-intervention state, default rules
usually exist, and any proposed intervention simply alters what result obtains if a
chooser does nothing – the default is just flipped (as opposed to being created). For
the purposes of this section we will assume that a default involves more than mere
one-click costs – that the cost implicated by opting-out are indeed opportunity costs.
Let’s consider the case of organ donation again. If we consider a chooser with
a pre-existing motivation not to donate their organs, then an intervention that flips
the default (e.g. an opt-out system) seems to increase their costs of choosing. Prior to
the default-flip intervention that chooser bore no costs to make a choice aligned with
their motivation. After the intervention, giving effect to the same motivation (i.e.
making the same choice) now carries new opportunity cost (time, energy, etc.). If an
intervention imposes opportunity costs on a chooser who already does not bear
those costs, then such an intervention certainly seems to reduce their opportunity
set, and thus not count as a nudge.
78
However, if we do the same analysis from the perspective of a chooser with a
motivation to donate their organs, flipping the default seems to reduce their
opportunity costs, or eliminate them completely, thereby enhancing their
opportunity set. Where such a chooser prior to the intervention had to bear the
opportunity costs of opting-in (e.g. completing bureaucratic steps), post-intervention
they can make their desired choice by doing absolutely nothing. The chooser now
has extra time and mental effort available for other choices. Therefore, from this
chooser’s perspective, flipping the default is a nudge since it enhances their global
opportunity set since their finite resources previously devoted to giving effect to
their motivation (opting-out) are freed to be expended elsewhere.
So, what can we conclude from an intervention that both enhances some
choosers’ opportunity sets but also decreases another group of choosers’
opportunity sets? Is there some balancing to be performed, where we can conclude
that an intervention preserves the opportunity set if it does so on some net,
aggregate basis?
An aggregation approach might suggest that when an intervention increases
the opportunity cost of one option while correspondingly reducing the cost of
another option by the same amount, the overall opportunity set is preserved.
However, this is a failed line of reasoning because the aggregate opportunity costs
required by an opportunity set are not relevant to our liberty of choice analysis.
79
Switching which group of choosers (delineated by their preferences to donate
or not) bears the costs of opting-out, does not preserve the opportunity set. No
matter if the costs are identical pre- and post-intervention, for one group of
choosers, their costs have changed. If a person who previously wanted to keep their
organs must now “pay” more in finite resources (time, energy) to do so, then their
opportunity set is not preserved, as they must pull those resources from other
choices, diminishing their global opportunity set. It does not matter if another
chooser who wanted to donate their organs also experienced a counterbalancing
enhancement in their opportunity set. The net cost of an opportunity set is irrelevant
to a chooser who is being made to pay more post-intervention for the same choice
they had pre-intervention.
An argument could be made that since we must have some default, it’s
globally “cheaper” to impose the costs of opting-out on the minority. Johnson and
Goldstein note that: “explicit consent policies impose the costs of switching on the
apparent majority.”69 For example, if we can somehow determine that the majority
of people, absent the costs, would donate their organs, then we should impose the
opt-out costs on the minority. Based on survey data, we should then have a
presumed consent, opt-out regime since most people express support for organ
69 Johnson & Goldstein, supra note 61
80
donation and a desire to donate their organs. This is an attractive line of thinking,
but unfortunately contributes little to our analysis.
Because liberty of choice is an individualistic concept, for our purposes, it is
not subject to aggregation. What matters is the liberty of each chooser – each person
has a liberty to choose, not some aggregate chooser. We cannot aggregate across
choosers since the opportunity set at stake refers to the opportunity set of a
particular chooser. For example, how does the post-intervention opportunity set
exist for a chooser who does not want to donate her organs? If flipping the default
results in that chooser facing higher opportunity costs to satisfy her preferences,
then to her, the intervention will always fail to preserve the opportunity set and fail
to satisfy the liberty of choice constraint. Therefore, for an intervention to qualify as
a nudge, the opportunity set must be preserved for all choosers, not just the
majority.
Finally, the argument could be made that defaults are permitted instances of
increased opportunity costs because the formation of the pre-intervention
opportunity set was arbitrary to begin with. This reflects Sunstein and Thaler’s idea
of the inevitability of choice architecture.70 Since someone had to decide the
substantive content of the default (e.g. donate vs. not donate) prior to the “flipping”
70 Thaler & Sunstein, "Nudge", supra note 1 at 10.
81
intervention, now switching that content to the non-default is simply responding to
an initial arbitrary instance of choice architecture.
This may very well be the case, but recall for our functional liberty of choice
constraint analysis we are interested solely in the state of opportunity sets
immediately before and after a given hypothetical intervention in order to measure
the change in opportunity sets. We are evaluating the effects of interventions for
identification as nudges and resulting ethical evaluation purposes. We are not
evaluating how the initial opportunity set came into existence because we are not
evaluating a state of affairs pre-nudge – we are evaluating the nudge and its
functions and effects.
To summarize, defaults work not by introducing new opportunity costs in
opportunity sets, but by switching pre-existing costs to different options within the
same set. Any intervention that switches the content of a default involving
opportunity costs fails to preserve the opportunity set of all choosers, thereby
disqualifying the intervention from our nudge classification (though it may possibly
be justified on other grounds).
3.7. Expected Costs & Benefits
Interventions in choice architecture can also alter how choosers perceive the
benefits or costs of their options. We have already explored how altering the
82
objective benefits or costs of options tends to fail to preserve the opportunity set. But
when we speak of anticipated benefits, we mean what choosers anticipate they will
get out of selecting a certain option.71 Likewise, with respect to expected costs, we
mean what costs choosers expect to have to pay to select a given option. The key is
that these are expected, not necessarily actual, costs or benefits.
We are no longer just interested in objective, finite measurements like time,
money or energy (i.e. opportunity costs), but all benefits that choosers may receive
from options, including unquantifiable psychological benefits and amorphous
concepts like social status or customer service. The same goes for costs – here we are
interested in what choosers think a particular option will cost. These are purely
subjective determinations – the tangibles and intangibles of selecting options. If you
choose a BMW over a Honda, you may get a better, more reliable car based on
technical specs, but you may also expect that it will give you greater satisfaction –
the feel of the premium leather interior, the status conferred from the vehicle, the
customer service from the dealer, etc. These are all the benefits that are associated
with the option, which bears an associated cost. A very real possibility is that a
choice does not deliver on the expected benefits, but it was the expectation of certain
benefits that drew choosers to select that option in the first place. In fact, humans
71 Sunstein, "Why Nudge", supra note 26 at 307.
83
tend to make affective forecasting errors where we misestimate the expected utility
of choices.72
Choosers’ expectations of an option’s costs and benefits may be misaligned
from the option’s actual costs and benefits. Interventions will generally either
amplify the gap, close the gap or leave it unchanged. For example, warning labels
may cause choosers’ expected costs to increase (by making them more salient) and
move closer to objective measures, as pre-intervention choosers may not be
appreciating the full, lifetime health consequences of smoking. The idea is to merge
expected costs with actual costs – to get choosers to fully appreciate the costs of their
choices. If choosers are habitually underestimating the costs, then the benefits of a
certain choice may seem larger than the expected costs. Information campaigns in
general are designed to move the subjective expectations closer to the objective
reality.73
As much as interventions can target expected costs, most of the promise of
nudging may instead lie with moving choosers’ expected benefits. The benefits a
chooser is going to gain from selecting a particular option are generally more
subjective than the costs of an option. Subjectivity, by definition, provides a nearly
limitless scope of opportunities for interventions to change the benefits choosers
72 Ibid at 307. 73 Fischer & Lotz, supra note 38 at 8.
84
may expect. Some choosers may respond to an intervention that makes an option
seem more luxurious, while others may respond to an intervention that makes an
option seem more sensible or frugal. That’s the beauty of human decision-making:
we all have different preferences and select options for a variety of reasons. Some
choose cinnamon buns over strawberries because they highly value the sugary taste
or the smell, or the feeling of fullness. Others may love the burst of flavor from the
strawberries, or are watching their calorie intake. Others may love neither choice,
but can’t stand the seedy taste of the berries so they go with the cinnamon buns. The
point is that because each chooser is different, with different subjective preferences,
the scope for targeting subjective evaluations of benefits is larger than the scope for
targeting the much smaller sphere of subjective costs.
As we have already seen in Chapter 2, interventions that only alter how
choosers subjectively perceive their opportunity set while leaving the objective
opportunity set alone, do preserve the opportunity set and therefore qualify as
nudges vis-à-vis the liberty of choice constraint. The objective opportunity set is
preserved because there are no new costs entered into the equation, the actual
benefits are not altered and there are no real instances of option-eliminating
substitution. All of the options as they existed pre-intervention exist post-
intervention.
85
3.8. Conclusion
We have seen in this chapter that nudges will fail to satisfy the liberty of
choice constraint by imposing new or higher monetary opportunity costs (e.g.
higher prices) by deploying the idea of budget constraints. Non-monetary
opportunity costs that affect choosers’ finite resources (e.g. time, energy) also fail for
the same reason. Defaults that shift opportunity costs to a different group of
choosers effectively increases the opportunity costs for those choosers, therefore
their opportunity sets are also not preserved.
On the other side of the ledger, interventions can satisfy the liberty of choice
constraint by reducing opportunity costs (either monetary or non-monetary). The
imposition of non-monetary costs that affect choosers’ infinite resources also count
as preserving the opportunity set.
In the totality of Part I, we have seen that for interventions to be labeled as
definitional nudges, they must satisfy a liberty of choice constraint, which means
they must preserve a chooser’s opportunity set. By comparing opportunity sets
immediately before and after (hypothetically) an intervention, we can evaluate
whether a chooser’s opportunity set is preserved. Figure 4 below revisits the
summary table from Part I’s introduction and categorizes each class of intervention
effect.
86
Figure 4 – Opportunity Set Intervention Summary Table, Revisited
Opportunity Set Preserving (Nudges) Not Opportunity Set Preserving (Not Nudges)
Objective effects on options:
- Adding options
- Increasing actual, objective
benefits/utility of options
Elimination of pre-existing options
- Prohibitions
- Pure substitutions
Costs
- Reducing opportunity costs
- Non-monetary, non-
opportunity costs
Imposition of new opportunity costs
- Increasing monetary costs
- Increasing non-monetary,
opportunity costs
Altering the perception of options –
salience, awareness & expected costs
and benefits
Opportunity cost shifting (e.g. default
flipping)
87
PART II: EVALUATING NUDGES
Autonomy Restraints
Part I showed that we can identify nudges as interventions in choice
architecture that preserve a chooser’s opportunity set. The liberty of choice
constraint that we use to identify nudges is what gives nudges their political and
policymaking promise – that they respect a chooser’s liberty.
Part II of this thesis will show that, once identified, we can ethically evaluate
nudges using a standard of autonomy preservation. That is, to be ethical on
autonomy grounds, nudges must preserve or promote the autonomy of choosers.
This Part will proceed as follows. Chapter 4 will unpack our conception of
autonomy as authentic self-government and outline how nudges can affect the
autonomy of choosers using a tripartite model of decision-making with associated
autonomy concepts: motivation formation (motivational autonomy), option
evaluation (decisional autonomy; together with motivational autonomy, “process
autonomy”), and resulting choices (outcome autonomy). Chapter 5 will show that,
because of its participatory nature, process autonomy should have lexical priority to
outcome autonomy in our evaluation effort. Chapter 6 will then outline how nudges
can affect the process autonomy of choosers at the motivational autonomy and
decisional autonomy levels.
88
Chapter 7 will explore how nudges can affect the derivative concept of
outcome autonomy, focusing primarily on the use of a hypothetical construct of an
autonomous chooser, such as that advanced by Sunstein and Thaler’s libertarian
paternalism. Chapter 8 will then draw out the shortcomings of outcome autonomy
as an evaluative concept, settling on its role as a tool of last resort in limited, highly
conditional circumstances.
Finally, Chapter 9 will examine the nine possible combinations of effects
nudges can have on choosers’ process and outcome autonomy. A conclusion as to
the autonomy ethics of each combination will be rendered.
89
Chapter 4
Mapping Autonomy
4.1. Introduction
Thus far we have focused on liberty of choice. We know that to be a nudge,
and enjoy the special political promise that comes with nudging, a choice
architectural intervention must preserve a chooser’s opportunity set; it must
preserve their liberty. But consider the case of a chooser whose opportunity set is
manipulated such that an option is still technically available to them but the person
cannot perceive that the option is there because it is hidden or shrouded from them.
That person is at liberty as their opportunity set has not been reduced and all pre-
existing options still exist post-nudge.
But would we say that the chooser in that scenario has chosen against the
shrouded option, that they must really not have wanted it? Is that kind of
perception-based nudge as morally or ethically good as nudges that inform people
of all their choices and allow them time to deliberate accordingly? Focusing solely
on a chooser’s liberty of choice doesn’t help us to answer those questions. Instead,
we must turn to the concept of autonomy. Autonomy can tell us what nudges are
ethically superior to others. When evaluating and not just identifying nudges,
autonomy, not liberty, must be our guide.
90
John Locke created a famous example that will help us distinguish liberty
from autonomy.74 Locke’s example begins with a man in a prison cell who is told all
the doors are locked and the guards even go through the motions of locking all the
doors. Unbeknownst to anybody, but especially the prisoner, one of the locks is
defective, so that the person is not really confined at all – he can leave if he so
desires – he just has to push open the door. The prisoner’s liberty has not been
constrained, even though he is not aware of it. But, as Gerald Dworkin notes, the
prisoner’s autonomy has been limited: “His view of the alternatives open to him has
been manipulated by the guards in such a fashion that he will not choose to leave.
This example shows that self-determination can be limited without limiting
liberty.”75
So liberty and autonomy are not the same thing. One can interfere with
liberty without interfering with autonomy and one can interfere with autonomy
without interfering with liberty.76 Autonomy is a much richer concept than liberty
since it captures not just the ability of a person to act without external restraint, but
the whole of what it means to be a self-determining person.
The primary goal of this chapter will be to delineate the contours of
autonomy as the capacity for, and exercise of, self-government. However, before we
74 Dworkin, supra note 35 at 105. 75 Ibid at 105. 76 Ibid at 105.
91
engage in a definitional exercise, a brief detour is needed on why we should value
autonomy as the key determinant in our ethical evaluation.
Following that discussion, this chapter will also show that nudges can affect
individuals’ autonomy in three areas, mirroring the three stages of decision-making:
higher-order motivation formation, option evaluation, and choices (outcomes). A
key distinction will be made between a chooser’s autonomy at stake in the first two
stages – motivational autonomy and decisional autonomy, together referred to as
process autonomy – and the last stage that we will call outcome autonomy.
4.2. The Value of Autonomy
There is a large body of literature on the value of autonomy and there is little
that this thesis will add to that rich discussion. However, a brief overview is
required. Autonomy has both instrumental and intrinsic value, but also political
value. Instrumentally, autonomy is valuable because it can serve as a foundation for
other values. Self-determining one’s choices in life increases the probability that
people will obtain satisfaction from those choices, as people are generally thought to
have a better appreciation of their own values and beliefs than others.77 Sunstein and
77 Ibid at 111.
92
others may question the latter epistemic part of that statement in light of
behavioural insights, but they at least agree that autonomy also has intrinsic value.78
Autonomy has intrinsic value both as a process and as a value in itself.
Regarding the former, Gerald Dworkin cautions that, “[w]e should not overlook the
extent to which the process of thinking about, reflecting upon, choosing among
preferences is a source of satisfaction to individuals.”79 While many of us may
disagree with that statement when we are faced with a situation of choice overload,
we should not discount the intrinsic utility some individuals’ can get from
undertaking the choosing process. Indeed, I derive immense pleasure from the
planning and choosing process involved in vacation planning, weighing which
options best reflect my values and goal. The process of exercising autonomy itself
delivers some value to me as a chooser.
Aside from the process element, autonomy is also valuable itself for the,
“intrinsic desirability of exercising the capacity for self-determination.”80 That is,
autonomy is valuable not because it leads to greater satisfaction through the
autonomous process, but because it is what makes us human; it is a defining feature
of what it means to be an individual and live one’s own life. Being autonomous is
itself valuable, independent of the consequences of autonomy.
78 Sunstein, "Why Nudge", supra note 26. 79 Dworkin, supra note 35 at 112. 80 Ibid at 112.
93
Most relevant to our discussion, autonomy has what we can consider political
value. Aside from its attractiveness on liberty of choice grounds, nudging is also
attractive because it purports to be about preserving, if not enhancing, individual
autonomy. This is why, as we will see in Chapter 7, that Sunstein and Thaler go to
such great lengths in developing their concept of libertarian paternalism to include
an autonomy/consent element. In liberal democracies such as our own, actions by
the state involving self-regarding choices that can enhance welfare while also
respecting autonomy (and separately, liberty) are politically attractive. On the other
hand, interventions that are seen to be traditionally paternalistic (contrary to an
individual’s autonomy in the name of the chooser’s own welfare) are stigmatized. It
is possible that the political value of autonomy is derivative of the intrinsic value of
autonomy – that because we value autonomy intrinsically, we also value it
collectively or politically.
Sunstein and Thaler predominantly value autonomy for its instrumental
contributions to individual welfare. To them, autonomy’s greatest value comes from
being an ingredient of welfare, and maximizing welfare (however they construe it)
should be the predominant guide in policymaking. Sunstein and Thaler write: “We
do not disagree with the view that autonomy has claims of its own, but we believe
that it would be fanatical, in the settings that we discuss, to treat autonomy, in the
94
form of freedom of choice, as a kind of trump not to be overridden on
consequentialist grounds.”81
Respecting the autonomy of choosers in one particular choice scenario also
makes choosers more autonomous in future choice scenarios through the discovery
of one’s true motivations and the development of individual decision-making
capacities. Sunstein also argues that choice architecture can help promote autonomy
by acting as a backstop in the event of errors or bad decisions: “autonomy is
promoted…by the existence of helpful choice architecture, ensuring that if we do
not make particular decisions, we will be just fine.”82
Finally, autonomy is valuable because it is closely linked with the idea of
responsibility. If an individual has made their own choices about their lives, and
lives by their authentic motivations, thereby exercising self-government, then that
person has in a way also claimed responsibility for their lives – an important value
in its own right. Any consequences of those decisions cannot be blamed on anyone
else if that person acted with authentic self-government; they must own those
consequences.
81 Sunstein & Thaler, supra note 13 at 1167. 82 Sunstein, "Why Nudge", supra note 26 at 137.
95
4.3. Autonomy as Self-Government
Many philosophers have debated the meaning of autonomy, and so many
treatises have been and will be written to fill the term with rich meaning that this
thesis cannot possibly do justice to the concept. As Dworkin notes, autonomy means
different things to different people (namely philosophers): “It is used sometimes as
an equivalent of liberty…sometimes as equivalent to self-rule or sovereignty,
sometimes as identical with freedom of the will.”83 Since the goal of this thesis is to
evaluate nudges ethically, we need not engage in a lengthy exploration of the
concept. However, we do need to develop a functional definition.
Most schools of thought agree on some foundational premises of autonomy
that reduce to the following basic definition: autonomy broadly means self-
government. A person is autonomous when they are governed by their own values
and when they carry those values through to their choices. All conceptions of
autonomy focus on, as Dworkin writes, “a notion of the self which is to be respected,
left un-manipulated, and which is, in certain ways, independent and self-
determining.”84
The term autonomy has its origins in ancient Greek. Autos means self and
nomos means rule or law. Autonomy, literally, means self-rule or, as we will say,
83 Dworkin, supra note 35 at 6. 84 Ibid at 11-12.
96
self-government. As Gerald Dworkin writes, persons are being autonomous when:
“their decisions and actions are their own; when they are self-determining.”85
Autonomy then, is about independent self-government. It is this understanding of
autonomy that we will use in our ethical evaluation of nudges.
As the lexicological history of the term suggests, persons are autonomous if
they govern themselves. But self-government is almost as amorphous a concept as
autonomy, so it also requires further unpacking. In the next few sections we will
explore the following key elements of our definition of autonomy as self-
determination: that the self (not others) governs a person’s motivations and actual
behaviour, and that the governing self is a person’s authentic self. Authenticity will
be the central point of our analysis, and it reflects several interrelated ideas: (a) a
person’s higher-order motivations are more authentic than their lower-order desires,
and (b) a person’s motivations are authentic in part to the extent that they are
independent of the undue external influences of others.
Thus, we will conclude that autonomy is about a person: (1) actually choosing
or deciding; according to motivations that are authentic because they are: (2) of a
higher-order nature, and (3) independent from undue external influence. We will
begin with an examination of (2) and (3) before returning to the idea of effective
autonomy.
85 Ibid at 13.
97
4.4. Authenticity: Higher-Order Motivations vs. Lower-Order Desires
Consider the following problem of conflicting motivations and desires: a
person wants to be healthy and this same person also repeatedly falls victim to
cravings and indulges in overeating on a regular basis. This person obviously
desires the food they eat, but they also desire not to overeat and be healthy. In the
abstract, away from the tempting food, when surveyed for example, they
acknowledge they should not eat as much, and indeed wish they did not. Likewise,
they, like millions of other people, engage the diet and exercise industry to lose
weight, attempts which usually end in failure.
This is a challenge for nudging and other policies from an autonomy
standpoint. If each set of motivations/desires (to overeat and to be healthy) can be
said to relate to a person’s autonomy, how can we possibly design choice
architecture that respects a person’s autonomy? To answer that question, we need to
resolve the conflict in favour of one set of “wishes” which we can deem to be those
of the authentic person whose ideas we are to respect.
The answer lies in Dworkin, Frankfurt, and Christman’s works on
autonomy and a person’s hierarchy of desires. According to their work, a person can
have a variety of different, even conflicting, desires. We can group these desires into
different classes: first-order, second-order, third-order, … etc. Broadly, first-order
98
and other lower-order desires tend to be, not surprisingly, desires, as we normally
think of them akin to urges or cravings. These desires typically take the form of: “I
want that piece of cake,” or “I have a craving for a cigarette.”
Higher-order motivations (as we will call them), on the other hand, are
judgments about lower-order desires, where one can reflectively self-evaluate on
their desires.86 These desires about desires take the form of: “I want not to desire that
piece of cake,” or “I wish I didn’t crave this cigarette.” Christman summarizes the
difference nicely as lower-order desires are desires about actions – “a desire to do X
or Y” – whereas higher-order preferences are desires about lower-order desires: “a
desire to desire to do X or Y.”87 For example, a smoker can have an urge or craving
to smoke yet at the same time regret smoking and desire not only not to smoke, but
not to have cravings to smoke.
A brief note on terminology: Dworkin and other philosophers use the term
desires, wants, preferences and motivations rather interchangeably. However, it
would be far clearer to use the term “lower-order desires” to refer to those drivers of
lower-order behaviour, such as cravings and urges, and other generally
subconscious reactions (captured by Kahneman’s idea of System 1 “thinking”). We
86 Harry G Frankfurt, “Freedom of the Will and the Concept of a Person” in John Christman, ed, The Inner Citadel: Essays on Individual Autonomy (New York: Oxford University Press, 1989) at 64. 87 John Christman, “Introduction” in John Christman, ed, The Inner Citadel: Essays on Individual Autonomy (New York: Oxford University Press, 1989) at 7 (emphasis in original).
99
will also use the term "higher-order motivations” to capture the drivers of higher-
order behaviour. This is more fitting, since these higher-order motivations are what
motivates us; what we really want ourselves to be like.
As per our definition of autonomy as self-government, a person’s most
authentic self – what truly represents that person – is the self we should be
concerned with for our ethical evaluation of nudges. A person is autonomous when
their authentic self guides their decision-making. Authenticity is found in different
elements of our autonomy analysis, but to start, it is the capacity for, and presence
of, higher-order motivations that makes a person autonomous because those
motivations are more authentically a person’s own than their lower-order desires.
Dworkin writes:
[A]utonomy is conceived of as a second-order capacity of persons to reflect critically upon their first-order preferences, desires, wishes, and so forth and the capacity to accept or attempt to change these in light of higher-order preferences and values. By exercising such a capacity, persons define their nature, give meaning and coherence to their lives, and take responsibility for the kind of person they are.88
To Dworkin, it is not enough to say that a person is autonomous if they do
not want to eat cake, they must not want to want to eat cake. The former is simply a
negative version of another lower-order desire. It is the latter – a want to not want to
eat cake or smoke – that is a higher-order judgment that reflects a person’s
autonomy.
88 Dworkin, supra note 35 at 20.
100
It is not enough to simply insist that higher-order motivations are authentic
and lower-order desires are inauthentic. It must be a relative analysis – authenticity
comes in degrees. Or, another way of putting it: lower-order desires are less
authentic than higher-order motivations, but they do still contribute a level of
authenticity to a person’s overall autonomy. This echoes Sunstein’s view that
System 1 reacting (fast, intuitive thinking akin to lower-order desires) can have
legitimate claims.89 As Sunstein notes, these kinds of baser urges are often what
gives life its flavour (at least spontaneous flavour). Urges and cravings are still part
of who we are – there is some authenticity and autonomy there. This is particularly
true if there are no applicable higher-order motivations about those lower-order
desires. A person who then follows their desires is acting as authentically as they
can. Likewise, choosers may have higher-order motivations to be governed by their
lower-order desires – they want to be that kind of person – and so a person acting in
accordance with those desires would then be acting authentically and
autonomously.
Thus in our understanding of autonomy as self-government, the self that
should govern is the highest available self: “autonomy will be thought of as the
highest-order approval and integration.”90 Some individuals will be governed more
89 Sunstein, "Why Nudge", supra note 26 at 157. 90 Dworkin, supra note 35 at 19.
101
so by higher-order motivations than others, but so long as each is being governed by
their own highest-order motivation, they are all authentically autonomous.
However, if lower-order desires (e.g. craving to smoke) are in conflict with
higher-order motivations (e.g. a desire to not crave to smoke or be addicted to
cigarettes), we must prefer the latter because that is where a chooser’s authentic self
is more likely to reside. Higher-order motivations are more authentic and are, as
such, valued more because they tell us something about the kind of person that a
chooser wants to be, what they aspire to, and who they see themselves as. In other
words, we value higher-order motivations over lower-order desires because it is in
those motivations that we find the ability of a person to express their fundamental
values and beliefs, and to manifest their own conception of the good life – in other
words, their true, authentic selves.
A person’s higher-order motivations may affirm a person’s lower-order
desires as reflective of their values and beliefs, but they may also indicate that a
person views their lower-order desires negatively – they want to be a different
person from the person they are moved to be by their lower-order instincts. As
Frankfurt notes, an autonomous person must be considered: “capable of wanting to
be different, in their preferences and purposes, from what they are.”91
91 Frankfurt, supra note 86 at 64 (emphasis in original).
102
Lower-order desires do not reflect a person’s beliefs or values because they
are reactionary, and there is no higher-order evaluation of those reactions to ensure
they align with that persons’ fundamental self. 92 Lower-order desires often
correspond with physiological needs, and such needs are often unconscious.93
Choices governed by lower-order reactions may accidentally align with a person’s
values or beliefs, but the intentionality matters, as does the certainty (i.e. more often
than not, a person’s lower-order choices may not reflect their fundamental values
and beliefs). Furthermore, as Felsen and Reiner note: “lower-order desires are
generally focused on immediate rewards, while higher-order desires are more
sensitive to long-term outcomes.”94 Thus, higher-order motivations are again likely
to be more reflective of a person’s beliefs and values, which are themselves often
longer-term in nature than transient desires.
Returning to our problem of the health-motivated overeater, which version of
the self are we to recognize as deciding autonomously? Their higher-order
motivation is to be the kind of person that can overcome their temptations and live a
healthy life. According to the hierarchy of motivations and authenticity theory, we
would locate our overeater’s authentic self in their desire to eat healthier (provided
92 See e.g. Gidon Felsen & Peter B Reiner, “How the Neuroscience of Decision Making Informs Our Conception of Autonomy” (2011) 2:3 AJOB Neurosci 3 at 4. 93 Ibid at 5. 94 Ibid at 5.
103
the independence condition of authenticity is met as well as we will see in the next
section).
4.5. Authenticity: Independence
It is not enough to say that someone is autonomous or that their choice
behaviour is autonomous simply because they have higher-order motivations that
guide their actions. There is a second element to finding that a person’s higher-order
motivations are authentic – that they are independent of undue external influence.
Here is where we encounter Dworkin’s idea of procedural independence –
that a person’s higher-order motivations should be ascribable only to themselves.
The kind of person that an individual wants to be, reflective of their fundamental
values and beliefs, should be their own and not ascribable to others.95 How one
comes to their higher-order motivations must be a method of formation of their own
choosing or construction, free of manipulation, deception, withholding of
information, etc.96 Thus, the origins and the means of the acquisition of a person’s
higher-order motivations matter. The process of obtaining higher-order motivations
should not be alien to the chooser. Rather, they should recognize how they got those
motivations as authentic to themselves.
95 Dworkin, supra note 35 at 18. 96 Gerald Dworkin, “Autonomy and Behavior Control” (1976) 6:1 Hastings Cent Rep 23 [Dworkin, “Behavior Control”] at 25.
104
Consider a person who has been manipulated or to an extreme, brainwashed,
by another to value something (x) in accordance with the manipulator’s objective.
The chooser believes as part of their fundamental values that x is part of their
conception of the good life, but they do not know how they came to that value.
Technically, they are motivated to want x as part of their higher-order motivations.
But would we say that that person’s higher-order motivation for x is authentically
their own? Of course not. How one’s motivations come about matters. It is not
enough to simply locate a motivation as higher-order and call it authentic.
Christman argues that a key concept in determining whether one’s
motivations are sufficiently independent (i.e. free of undue influence) so as to be
authentic is that of “identification.”97 A chooser should identify with a particular
motivation or desire, “given the conditions of its formation.”98 A chooser may have a
particular motivation and know that it comes from an external influence, but if that
influence is not one that the chooser identifies with as being consistent with their
values and beliefs, then it is not authentic to the chooser.
Thus, what we are interested in, when investigating whether a nudge counts
as an undue external influence, is the effect of those influences on a chooser’s
motivations. Here is where we find the idea, advanced mostly by Christman, that
97 Christman, supra note 87 at 8-9. 98 Ibid at 8.
105
autonomy in terms of the origin of desires and motivations, is a historical analysis:
“the conditions of desires and values that must be met for them to be autonomous
are properties of the formation of those desires.”99 A determination of whether a
person is autonomous turns on the history of the formation of their motivations and
desires. If a person’s motivations are formed independently, then, when those
motivations govern that person’s behaviour, that person is autonomous.
The obvious drawback to Christman’s historical approach to autonomous
motivations is a practical one: how can we possibly disentangle a lifetime’s worth of
background social or environmental influences on our motivations such that we can
say some motivations are authentically our own and others are not?
Dworkin’s response is to abandon the historical approach and simply focus
on whether a person accepts their desires or motivations as their own. Dworkin
focuses on the ability to recognize one’s motivations and to coolly judge the lower-
order desires that is the key component to acting authentically. Dworkin writes that
we must accept that, “[w]e simply find ourselves motivated in certain ways and the
notion of choosing, from ground zero, makes no sense…But we always retain the
possibility of stepping back and judging where we are and where we want to be.”100
To Dworkin then, the key is not so much in focusing on the origins of motivations
99 Ibid at 9 (emphasis in original). 100 Dworkin, "Behavior Control", supra note 96 at 25.
106
but how individuals evaluate those motivations. If they reflectively accept them as
part of their own lives and identify with them, the origin does not matter.
A combined approach is best – one that considers how a chooser’s
motivations were formed and whether the (new) motivations are reflectively
endorsed by the chooser. Consider the desire many people have to be thin, but not
just for health reasons. We have been conditioned by Western society into thinking
that part of living a good life is being thin and attractive. The “desire” to be thin is
not just a lower-order desire, it is a higher-order motivation that guides many other
lower-order choices. If a person chooses not to eat due to their motivation to be thin,
such a choice appears to be consistent with their autonomy – their higher-order
motivation is governing their lower-order responses. But it is the source of that
higher-order motivation that is proving problematic – there are possibly undue
external influences at play that need to be considered. Thus, we need to inquire into
the means and manner of motivation formation.
Christman’s historical approach may identify a motivation to be thin as
suspect (i.e. not independent), but a means of evaluating them as authentic or not is
still required. Here is where Dworkin’s method of reflective self-evaluation is useful.
In order to determine if a person’s motivations are truly their own, appeal must be
made to an even higher-order level of motivation. That is, a person must effectively
107
be asked if their motivation to be thin is consistent with their fundamental values
and beliefs.
Recall Dworkin’s authenticity argument that: “[i]t is the attitude a person
takes towards the influences motivating him which determines whether or not they
are to be considered ‘his.’”101 If a chooser is shown the history and source of their
motivation to be thin, and nonetheless has a favourable attitude, then that
motivation is no longer suspect, it is endorsed and becomes part of the chooser’s
authentic motivational structure. If, on the other hand, the chooser does not endorse
that historical sourcing of their motivation, then the influences should not be
considered authentic.
The first step then, in order for a person to approve of the means of
motivation formation requires, at a minimum, for a person to be made aware of the
existence of such influences on their motives – Christman’s history of motivations.
We cannot expect a person to take an attitude about their motivations (or even
lower-order desires) without both opportunity and information. Opportunity
usually takes the form of a chance for deliberation, so that a chooser may reflectively
self-evaluate. But having the opportunity to evaluate one’s motives is useless
without the information that their motives may not authentically be their own. A
person must be aware of the influences on their motives, so that they are, “able to
101 Ibid at 25.
108
step back and formulate an attitude towards the factors that influence his
behavior.”102
Felsen and Reiner helpfully distil this idea that a chooser must be aware of
influences on motivations in order to accept or reject them into a quick reference
determination: is the influence overt or covert?103
Overt influences on higher-order motivations are generally ethically
acceptable because they provide individuals with an opportunity to accept or reject
such influences. A person may wish to incorporate such influences into their own
higher-order motivations or they can resist the influence and maintain their current
motivation system. But the key is that in order to accept or resist the influence, the
person must be aware of the influence in the first place. The idea of resistibility also
comes into play here – individuals should be able to easily resist the overt influence,
otherwise it becomes something more than a mere influence. Akin to resistibility,
Dworkin argues that influences should also be reversible – a person should be able
to reverse the effects of an influence.104
Education is an excellent example of an overt influence on a person’s higher-
order motivations that appear to satisfy the independence requirement. Higher-
order motivations resulting from education can generally said to be authentic to the
102 Ibid at 24. 103 Felsen & Reiner, supra note 92 at 8. 104 Dworkin, "Behavior Control", supra note 96 at 28.
109
chooser (depending on the particular circumstances of the education). Well-
designed general educational efforts should be aimed at equipping individuals to
achieve a state of reflection and knowledge such that at some point they can reflect
on the educational influences themselves and decide whether or not to accept those
influences as part of their authentic motivational structure.
Overt external influences can be contrasted with covert influences, which
Felsen and Reiner note, “act unconsciously [and] potentially subvert the individual’s
higher-order desires and thus would constitute such a threat [to autonomy].”105
Recall that our definition of autonomy is one of self-governance, or self-control. Any
influences that are beyond an individual’s control are clearly questionable on
autonomy grounds.106 If a person is subconsciously manipulated into holding a
particular higher-order motivation, without appreciation that they are being
manipulated, then their actions and motivations cannot be said to be authentic
because they never had an opportunity to reflectively consider those influences for
incorporation into their motivational structure. Covertness goes to the question of
authenticity – true authenticity is simply not possible with covert influences.
Authenticity then does not require complete independence from the external
influence of others, only independence from undue external influence. Influences are
105 Felsen & Reiner, supra note 92 at 8. 106 Ibid at 4.
110
undue to the extent that a chooser is unable to accept or reject/resist the influence. If
an influence is transparent to the chooser and the chooser opts to incorporate that
influence into their own motivational structure, then the resulting motivations could
be said to still be authentic to the chooser. On the other hand, higher-order
motivations arising as a result of covert influences would not count as authentic
motivations, meaning that a chooser who acts in accordance with those motivations
is not exhibiting full autonomy.
4.6. Effective Self-Government
Note that our definition of autonomy requires rule or government by the
independent, higher-order self (i.e. in accordance with authentic motivations). It
cannot be emphasized enough that we are not just interested in the capacity for
autonomy, but the expression of that capacity through behaviour and choice.
Autonomy, in Christman’s view, is both the, “the state of being capable of,
and actually, exercising self-government.”107 Harry Frankfurt goes further than
Dworkin and Christman in arguing that a person’s will is their effective desire – that
desire which moves a person to act.108 Frankfurt’s argument suggests that, from an
autonomy perspective, we should be interested in authentic motivations that are
107 Ibid, referring to John Christman, ed, The Inner Citadel: Essays on Individual Autonomy (New York: Oxford University Press, 1989). 108 Frankfurt, supra note 86 at 65.
111
also effective motivations that actually move individuals to act. 109 Frankfurt calls
these second-order volitions (as opposed to second-order desires). 110
This contrasts with Dworkin’s sole focus on the state of autonomy within a
person – whether that person is autonomous or not is independent of whether that
person actually acts autonomously. Since we are interested in the behaviour of
individuals and how their behaviour is altered by nudge interventions, we must
take a broader view of autonomy, adopting Christman’s conception of autonomy as
the capacity for, and the exercise of, self-government.
Frankfurt’s and Christman’s approaches are also more reflective of what
autonomy as self-government means. If a person has the capacity for autonomy but
does not choose options in accordance with that capacity, then they are not governed
by their authentic selves, which means they are not truly autonomous. In other
words, a person may have authentic motivations, but if those motivations are not
expressed into their actual choice behaviour then that the person is not truly self-
governing. It is not enough to say that a smoker wants not to want to smoke; their
behaviour and choices should be guided by that higher motivation to not be a
smoker.
109 Ibid at 67. 110 Ibid at 67.
112
This reflects a general understanding of respecting autonomy as respecting a
person’s will. Dworkin, Frankfurt and Christman have shown us that a person’s will
is not located in their lower-order desires but in their higher-order motivations. A
person that acts solely on lower-order impulses and urges has not “willed”
anything; they have acted in a completely reactionary way.
In sum, a person is more autonomous when they have effective motivations,
as opposed to motivations simpliciter. After all, a person with a particular higher-
order motivation may have other motivations at a similar ordinal-ranking that
conflict with those motivations resulting in that initial motivation not moving the
person to action. Since we are interested in actual choice behaviour, we are really
interested in effective (or volitional) higher-order motivations.
4.7. Process & Outcome Autonomy
We now need to translate this theoretical discussion of autonomy into a
practical framework for evaluating the effects of nudges on autonomy in decision-
making contexts. A key distinction at the heart of ethically evaluating nudges is the
difference between, and lexical priority of, autonomy found in the first two stages of
decision-making (process autonomy) and the derivative autonomy of the last stage
(outcome autonomy).
113
Decision-making starts with an indifferent person who, over the course of
their life and in response to past choice problems, develops a set of higher-order
motivations and lower-order desires. These motivations are then used (in choice
architecture that respects autonomy) to guide a person’s selection between a set of
options (option evaluation). This stage is about how a person decides – can they
access and apply their authentic motivations or are their option selections guided by
their lower-order desires? The final selection of one or more options over other
options results in a final choice (what the person has chosen and done), including,
potentially, a choice to not choose.
Figure 5 below summarizes and simplifies the decision-making process that
we will discuss. Note that each stage in the decision-making process feeds back and
informs the other stages, both within an immediate choice problem and across
choice problems over time. A person’s experience from past choices affects their
option evaluation and may also influence their motivations (e.g. perhaps they like
the kind of person they have become by making certain sets of choices).
114
Figure 5 – Simplified Decision-Making Process
The first stage, motivation formation, is what it seems – the process through
which a person develops motivations for their actions. As we have seen, these
motivations occur on a variety of ordinal levels. A person’s lower-order desires,
while predominantly physiological and reactionary, are still, to some extent, a
product of our environments (including the prior influences of others). A person’s
authentic motivations, on the other hand, are much more learned. Through years of
education and experience a person may come to value living a healthy life, or a
financially successful life as fundamental to their existence.
Choice architecture (and changes in choice architecture via nudges) has the
potential to alter individual higher-order motivations by constructing motivations
where they are not present or by failing to recognize that individuals may have
higher-order motivations to remain undecided (i.e. people have chosen not to
MotivationFormation(motivationalautonomy)
OptionEvaluation(decisionalautonomy)
Choice(outcomeautonomy)
115
choose, as we will see in Chapter 5). Nudges can also cause changes to a person’s
motivational structure through undue external influence on existing motivations.
Both sets of effects raise the possibility that any higher-order motivations resulting
from those influences (constructing or altering) may fail to be authentic to the
chooser (i.e. lack motivational autonomy).
Nudges can also affect the second-stage of the decision-making process – the
evaluation of different options (the “deciding” or “choosing” part of decision-
making). Since we know that autonomy is found most in a person’s capacity and
possession of authentic motivations, it stands to reason that option evaluations that
call on those higher-order motivations are more autonomous (i.e. have greater
“decisional autonomy”) than other processes that appeal to a chooser’s lower-order
desires. This suggests that at a minimum, the option evaluation process should
demand some level of deliberation, allowing choosers the opportunity to evaluate
which options best align with their fundamental beliefs and values. A chooser
cannot be said to choose A over B autonomously, without knowing whether that
chooser wishes to be the kind of person that chooses A over B. For example, if a
person chooses to smoke after carefully deliberating on the options, but chooses to
do so only because they have been unduly manipulated into believing that smoking
is the key to the good life, then their choice process, while deliberative, is not
116
necessarily autonomous. Here we would say that they have decisional autonomy
but not motivational autonomy.
Ideally, a chooser should have autonomy at both the motivational level and
decisional level. This makes sense: having autonomy at the latter means that they
are able to call upon their autonomy in the former, while realizing the former
requires effective autonomy in the latter.
While the first two stages of decision-making are focused on the processes
that choosers undertake (the process of, and reasons for, deciding), the final stage of
decision-making is the outcome of those first two processes – the resulting choice
that a chooser has made. If a chooser has to decide between options A, B and C and
chooses C, then the C is our “choice outcome.” This is where most nudges are
focused – on changing the actual choices that choosers make, regardless on how the
nudges work through the other two stages (which are arguably more important on
autonomy grounds).
Choices (outcomes) carry with them “outcome autonomy” to the extent that
they align with a chooser’s authentic motivations (i.e. what they would have chosen
if they were fully autonomous at the other two stages). We will occasionally refer to
this type of autonomy as “as-if autonomy.” Outcome autonomy can be thought of as
a characteristic of choices. If a person wants to not want to smoke and one of the
options is not smoking, then that option has outcome autonomy.
117
Determining the outcome autonomy level in choices is easy if the person is
provided with an opportunity to exhibit process autonomy since with process
autonomy they are likely to select choices in line with their authentic motivations.
However, establishing outcome autonomy may also be done (and is most often
done) using a hypothetical construct of the chooser and attempting to guess what
they would have chosen if autonomous. As we will see in Chapter 8 this latter
method is fraught with difficulty and uncertainty, jeopardizing the value of
outcome autonomy in our ethical evaluation of nudges.
Our representation of decision-making also contemplates that particular
choice scenarios may not require all the stages of decision-making. Consider again
the case of defaults. Defaults are predominantly about choices (outcomes):
individuals are defaulted into choices without, necessarily, any element of choosing
or appeal to a person’s higher-order motivations or even lower-order desires. A
person may be defaulted into, for example, consenting to donate their organs,
without them even being aware of it.111 If a chooser has an authentic motivation to
111 In fairness to the many proponents of opt-out donor consent schemes, none are suggesting a deliberately stealth switchover. They all want people to be aware that they can opt-out. For example, Wales, which recently switched to an opt-out program, heavily advertised the fact that the switch was occurring. With that said, however noble the intentions may be, with defaults, there are bound to be some people who simply aren’t aware, and it’s the effect of the nudge that matters for our analysis, not the intent behind it. In the Wales case, despite the two year-long publicity campaign, 24% of the population was still unaware that the organ donation system was changing immediately prior to the change (Welsh Government, “Update to Survey of Public Attitudes to Organ Donation: Key figures from Wave 10” (2 January 2016), online: <http://gov.wales/statistics-and-research/public-attitudes-organ-donation/?lang=en>.). For a consideration of awareness and defaults generally, see: N
118
donate their organs, then in this scenario they would have outcome autonomy since
the resulting “choice” (donating) matches their authentic motivation. However, that
person would not have decisional autonomy, since they may not have decided or
deliberated on anything – a choice was made for them, they never accessed and
verified their own authentic motivations. The resulting alignment is more a happy
coincidence than the result of self-government.
Because we are concerned about a person’s effective autonomy – that they act
in accordance with their authentic motivations - our ideal decision-making situation
would involve autonomy at all three stages. This means that a chooser clearly needs
motivational autonomy in order to be said to have authentic motivations, but they
also need to put those motivations into effect; bring them to bear on choice problems
by accessing and applying them at the option evaluation stage (including choosing
not to choose). If they are able to do so, we would say they have decisional
autonomy. Finally, in a scenario with good motivational and decisional autonomy,
the resulting choice should have outcome autonomy because their choices should
align with their authentic motivations.
Craig Smith, Daniel G Goldstein & Eric J Johnson, “Choice Without Awareness: Ethical and Policy Implications of Defaults” (2013) 32:2 J Public Policy Mark 159.
119
4.8. Conclusion
This chapter has shown that autonomy is a distinct concept from liberty and
valuable for many reasons. Furthermore, we should understand autonomy as
effective self-government in accordance with a chooser’s authentic motivations.
Authentic motivations are those that are both of a higher-order nature, and also
independent from undue external influences. Higher-order motivations are those
that reflect our fundamental values and beliefs and that set overarching goals for
our lives, governing the types of person we want to be, including the types of lower-
order desires or urges we wish to have.
Higher-order motivations are inherently more authentic than lower-order
desires but they are also authentic to the extent that they are independent from
undue external influences. The latter constitute those influences that are covert or
irresistible such that a person does not have a true opportunity or ability to integrate
them into their conception of the good life.
This chapter has also outlined a key distinction that we will use in the rest of
this thesis – process autonomy and outcome autonomy. Process autonomy includes
the first two stages of decision-making – motivation formation (motivational
autonomy) and option evaluation (decisional autonomy). Finally, choices resulting
from the decision-making process can be evaluated based on their levels of outcome
120
autonomy. However, as we will see in the next few chapters, outcome autonomy is
an inferior concept.
121
Chapter 5
The Lexical Priority of Process Autonomy
5.1. Introduction
In our ethical evaluation of nudges, process autonomy and outcome
autonomy should not be valued equally. Instead, process autonomy must have a
place of primacy or lexical priority due to its participatory nature. As we will see in
subsequent chapters, when evaluating the effects of nudges on autonomy, this
means that we cannot trade off gains in outcome autonomy for smaller losses to
process autonomy. In the event that a nudge does not affect process autonomy, only
then are we to look at outcome autonomy effects.
We will begin this chapter with an examination of the key participatory
nature of process autonomy, including many of the side-benefits of participation
such as learning and long-term behaviour change. Next we will explore how a focus
on process autonomy allows choosers to have a variety of individualized
motivations; that no particular motivation is the “right one.” We can split this
“heterogeneity of motivations” into two classes – one that focuses on the content of
motivations and the other that focuses on the type or characteristics of motivations.
The former goes to process autonomy permitting individuals to hold other-
regarding motivations whereas the latter permits motivations that may be context-
122
dependent, weak or strong, an absence of motivation (i.e. undecided), or meta-
motivations about motivations.
The benefits of process autonomy outlined in this chapter also count as
drawbacks to outcome autonomy, which will be further explored in Chapter 8. All
these factors support a robust nudging agenda that facilitates opportunities for
process autonomy improvements, or, at a minimum, process autonomy
preservation.
5.2. Participation
Process autonomy has lexical priority to outcome autonomy because the
former embraces the idea of participation. Process autonomy is about actual
autonomy, not hypothetical autonomy. Indeed, the key to being self-governing is
participating in one’s life. A person that is unable to put their motivations into effect is
not self-governing. The key is that a chooser should be provided with an
opportunity to be self-governing, and this means that they must have an
opportunity to express their own authentic motivations. Our focus here on process
autonomy and participation runs counter to most of the nudge world’s tunnel-
123
vision on outcomes: “[they] focus their attention on the observable starting and end
points of a choice process, and pays little attention to what ‘goes on in between.’”112
Consider the following phrases: “Julie’s choice is B,” and “Julie chooses B.”
They refer to the same idea – that Julie’s choice outcome is B, but the difference lies in
the process. We know nothing about process from the former, but we know from the
latter that Julie has actually engaged in some choice process such that we can ascribe
her choices to her more than with the former. We can be more confident that the
selection of B reflects Julie’s authentic self, and thus her autonomy.
Even if we had perfect knowledge of Julie’s authentic motivations and what
options align with those, nudging her to those choices (outcomes) without
providing her an opportunity to exercise her autonomy, would in fact be
disrespectful of her autonomy because she would be denied the opportunity to
actually self-govern. The process and the means matter and is part of what makes a
choice autonomous to begin with. It is far superior ethically to get someone to do
what they would do autonomously (assuming that can be determined), by giving
them the opportunity to autonomously choose to do it. And if we determine
outcome autonomy based on what a person would choose if fully autonomous and
112 Riccardo Rebonato, “A Critical Assessment of Libertarian Paternalism” (2014) 37:3 J Consum Policy 357 at 377.
124
they are in fact fully autonomous, there should be no gap and no need to refer to a
hypothetical version of that chooser.
As Sandhu has noted, “[p]articipatory decision processes…are preferred not
just because they lead to preferred culmination outcomes, but partly in their own
right.”113 And Mill summed up the distinction between process/participation and
outcomes nicely: “If a person possesses any tolerable amount of common sense and
experience, his own mode of laying out his existence is best, not because it is the best
in itself, but because it is his own mode.”114
As we will note in Chapters 7 and 8, because outcome autonomy works
through a hypothetical, it is possible that a chooser could be nudged towards a more
outcome autonomous choice, without that chooser even being aware of their own
authentic motivations. Because those motivations are complete hypotheticals – what
the chooser would choose with full process autonomy – they technically exist
independent of the chooser’s awareness of their motivations. Thus, it is possible that
a nudge could enhance a chooser’s outcome autonomy without a chooser even
being aware that (a) they are nudged, or that (b) the motivations reflected in those
choice outcomes are authentic to themselves.
113 Martin E Sandbu, “Valuing Processes” (2007) 23 Econ Philos 205 at 208. 114 John Stuart Mill, On Liberty (London: Filiquarian Publishing, 2006) at 90.
125
The point of this section (and the broader thesis) is that it is not enough to say
that a person has an autonomous motivation and ends up with a choice outcome
that matches that motivation. What comes in between matters - they must choose the
choice outcome for themselves, for that is what it means to act autonomously. The
doing (choosing) matters a great deal. In other words, what someone would choose
autonomously is much different than giving them a chance to actually choose
autonomously. Note the focus on the word “chance” – an opportunity ought to be
provided, but a chooser need not take it (we will return to this idea later in this
chapter).
Robert Nozick created a fascinating thought experiment that we can use to
illustrate the value of participation via process autonomy when contrasted with
outcome autonomy. Nozick asks us to imagine an elaborate “experience machine”
that users can plug themselves into and select from an enormous menu of life
experiences.115 While in the machine, the user experiences whatever life they have
selected and is completely unaware that they are doing anything but living their
chosen life. Nozick then asks why it is that many of us would choose not to use the
machine?
We can apply the experience machine analysis to our dichotomy between
process and outcome autonomy. Outcome autonomy is akin to being in the machine
115 Robert Nozick, Anarchy, State, and Utopia (Basic Books, 1974) at 42-45.
126
– it provides the illusion of living a fully autonomous life. In fact, the experience
machine may go a little further and actually provide individuals with the illusion of
process autonomy – that they are actually deciding while in their “chosen life.” But
the machine, just like outcome autonomy, is not about true participation, it is about
the illusion of certain outcomes.
Nozick offers several compelling reasons for users not to plug into the
machine that we can marshal in favour of process autonomy.116 First, Nozick notes
that experiencing things is different from actually doing things. Process autonomy is
about experiencing decision-making, not just having a choice outcome selected for
us (even if the selection is ostensibly in the interests of our autonomy). More
importantly for our analysis, Nozick argues that another reason not to plug into the
machine is that we want to be certain types of people. We do not, according to
Nozick want to be an “indeterminate blob” floating in the experience machine.
Instead, we want to the kind of person that discovers who we are through actual
experience.
The same reasoning applies to process and outcome autonomy. The machine
is a simulation, just as outcome autonomy is. Process autonomy, like life outside the
machine, is about actually living life as an autonomous decision-maker. A
hypothetical simulation will always be of a lower-order than the real thing. As
116 Ibid at 43.
127
Nozick notes about the experience machine (and by extension outcome autonomy):
“What is most disturbing about them is their living of our lives for us.”117
An excellent example of how participation and nudging works is that of
consent to organ donation. As we have noted elsewhere, a default of presumed
consent where individuals have to opt-out is extremely effective at increasing the
apparent consent rate. For example, in opt-out Austria, the “consent” rate is 99.98%
whereas in neighboring Germany (which has an opt-in system) the consent rate is
only 12%.118 It is true that close to one hundred percent of Austrians are deemed to
consent to their organs, but we certainly cannot infer form those statistics that so
many chose or actually consented to donate their organs. At best, we can only say
that their choice is to donate, but not that they have chosen. Obtaining a result by
doing nothing, possibly not even being aware that there was a choice to be made, is
certainly not participation (nor is it participation in the current regime).119
5.3. Side-Benefits of Participatory Process Autonomy
The participatory element inherent in a process autonomy approach also has
several autonomy-related “side-benefits” that buttresses our preference for process-
based approaches. This section will explore those side-benefits.
117 Ibid at 44. 118 Johnson & Goldstein, supra note 61. 119 See e.g. Smith, Goldstein & Johnson, supra note 111; Welsh Government, supra note 111.
128
Process autonomy serves a capability enhancement function.120 A person
needs to participate in their decision-making process in order to discover and
develop their motivations. The capabilities developed in that motivation discovery
and application process can be used for future decision-making. If individuals are
freed from undue external influences or de-biased and provided with an
opportunity for deliberative reflection, their capabilities are enhanced. They are
better decision-makers as a result. As scholars have noted, “by removing the
opportunities for social learning [behavioural policies] could have significant long-
term consequences for individual’s long-term social competencies.”121 Binder and
Lades call these persistent effects “dynamic effects,” often achieved through self-
discovery and application of one’s authentic motivations.122
We can think of the process of discovering and applying one’s motivations as
a learning process, the results of which can be applied in a variety of choice
scenarios post-nudge. Binder and Lades repeatedly refer to how some biases rob
individuals of the opportunity to learn.123 For example, making healthy food more
salient than unhealthy food reduces the opportunities for individuals to learn how
to make good choices for themselves. Likewise, the use of anchors robs individuals
120 See e.g. Erin Frey & Todd Rogers, “Persistence: How Treatment Effects Persist After Interventions Stop” (2014) 1:1 Policy Insights from Behav Brain Sci 172. 121 Jones, Pykett & Whitehead, supra note 3 at 52. 122 Martin Binder & Leonhard K Lades, “Autonomy-Enhancing Paternalism” (2015) 68:1 KYKLOS 3 at 13. 123 Ibid.
129
of the ability to learn about their authentic motivations through deliberative
thought.
Sunstein and Thaler also rightly note that one of the key drivers of behaviour
change is that of feedback.124 Bad decisions can have positive consequences –
learning from our mistakes is necessary to avoid making them in the future.125 To
learn, a chooser needs to participate in their choices. If a chooser is defaulted into a
particular choice without awareness or true, participatory consent, how can they be
expected to learn from that choice to inform future choices?
Nudges that work in a manner disrespectful to choosers’ process autonomy
are unlikely to empower individuals to self-regulate, leading to questionable effects
in future choice contexts.126 For example, defaults in one context have been shown to
induce individuals to follow defaults in differing, subsequent choice contexts.127
Along the same lines, Sunstein notes that defaults do not promote learning, which
may make individuals worse at decision-making in general since they are not
getting practice at the latter.128
A focus on participation and providing opportunities for individuals to
autonomously participate in their own decision-making may also help to address
124 Thaler & Sunstein, "Nudge", supra note 1 at 77. 125 White, supra note 6 at 119. 126 Binder & Lades, supra note 122 at 23. 127 Thomas de Haan & Jona Linde, “Good Nudge Lullaby” (2012) [unpublished]. 128 Sunstein, "Choosing", supra note 12 at 17.
130
some of the systemic factors causing behavioural biases, action-intention gaps and
the like. Many of our behavioural biases can, at least in part, be traced to
circumstances and the peculiarities of one’s environment.129 For example, studies
have repeatedly shown that those with fewer resources (who suffer “scarcity”) have
different behaviour patterns and cognitive biases than those with more.130 Indeed, it
has been argued certain nudges tend to, “psycho-demographically segment certain
portions of the population as being in most need of behavioural intervention.”131
Instead of suggesting that those individuals – whose choices tend to be associated
with so-called “vices” such as obesity, addictions, gambling132 – are in need of hand-
holding to reach particular outcomes, we can focus on including them in the
decision-making process by enhancing their capabilities through providing
opportunities for participatory process autonomy.
Individuals suffering scarcity or similar background/environmental
determinants, tend to be the ones most “susceptible” to default rules and other
nudges that work through behavioural biases instead of trying to overcome them as
nudges that enhance process autonomy do. This has the end of result of those
individuals ending up with (but not necessarily choosing) choice outcomes that may
129 Rebonato, supra note 112 at 394. 130 See e.g. Anuj K Shah, Sendhil Mullainathan & Eldar Shafir, “Some Consequences of Having Too Little” (2012) 338:6107 Science 682. 131 Jones, Pykett & Whitehead, supra note 3 at 51. 132 Ibid at 51.
131
not reflect their authentic motivations. Encouraging chooser participation at the
process autonomy level will ensure that those who currently (pre-nudge) do not
participate as much in decision-making have their autonomy respected at the
higher-order process autonomy level. In other words, enhancing the capabilities of
all choosers, will be an expression of far greater respect for individual autonomy
than merely focusing on outcome autonomy.
An unscientific example of this phenomenon occurred when Washington
state added a five dollar opt-out donation to state parks as part of its driver’s license
renewal.133 Flipping the default certainly accomplished its goal – there was a
resulting massive spike in the funds received.134 We could read into this that
individuals’ truly wanted to donate (i.e. that it was their authentic motivation after
all) and the switch simply made it easy enough to put that motivation into action.
Or we could just acknowledge that defaults are sticky and we don’t know anything
except that the switch was effective.
Of note though is that local media probed the distribution of increased
donation rates and found that they came disproportionately from certain
neighbourhoods – poorer ones with high immigrant and elderly populations.135
133 KIRO Team 7, “Stealth Donation Collection Earns Parks Millions”, KIRO TV (2010) online: <http://www.kirotv.com/news/news/stealth-donation-collection-earns-parks-millions/nDRqp/>. 134 Ibid. 135 Ibid.
132
Media reports suggested that individuals from those neighbourhoods were less
likely to opt-out because they didn’t have the cognitive capacity to do so (as a result
of scarcity) or because they didn’t understand the language on the form which was
in English only.136
Whatever the true reasons, the disparity cannot be ignored. Choice architects
should be vigilant regarding the socio-economic distribution of their nudge policies.
Nudges that fail to preserve the process autonomy of choosers, such as in the parks
donation case, raise the specter of “discrimination by default.” On the other hand,
nudges that focus on participation and capability enhancement would engender no
such concerns since they are directed at enhancing the motivational and decisional
autonomy of all choosers (or at least preserving the process autonomy of all
choosers).
Finally, aside from participation being the sine qua non of actually being
autonomous (self-governing), it is also valuable instrumentally – individuals derive
pleasure from the act of participating, even if they participate in a biased or less than
fully process autonomy manner.137 There are two elements to this. The first is that
the act of participating in the choosing process may give them happiness.138 Many
individuals value making their own choices, and not simply having their choices
136 Ibid. 137 See e.g. Danny Scoccia, “Paternalism and Respect for Autonomy” (1990) 100:2 Ethics 318 at 323. 138 See e.g. Rebonato, supra note 112 at 377.
133
made for them, even if they were to end up in the same place that they would have
had they had full process autonomy. The second part is one of independence and
domination – that some people would prefer to participate in their own lives, even if
they do so on a flawed or biased basis, because they do not want to be ruled by
others.
5.4. Heterogeneity of Motivations
Another reason process autonomy is valuable (and lexically prior) is because
it is the approach that accounts for complete heterogeneity of authentic motivations,
while outcome autonomy does so only to a lesser extent. Perhaps the foremost issue
that arises from the findings of behavioural insights is that individuals do not
always have stable, coherent motivations. Sometimes individuals simply do not
have motivations at all as they are undecided. At other times individuals may have
motivations about motivations.
Autonomy as self-government is neutral to the content of a person’s higher-
order motivations, so long as they are arrived at authentically. If a person doesn’t
want to want to smoke, so long as there is no undue influence, that higher-order
motivation must be respected in order for us to respect that person as an
autonomous, self-governing individual. We do not care why they don’t want to
want to smoke (outside of independence concerns).
134
Note that when we talk of a persons’ conception of the good life and
fundamental beliefs and values, these are not objective elements subject to empirical
observation and verification. A person can have an objectively irrational conception
of the good life. They could value undertaking extremely dangerous activities just
for the sake of the thrill (as some do). A choice as a result of a completely
autonomous choosing process does not necessarily have to be one that reflects
objective values. Process autonomy is designed to be neutral on the content of a
person’s motivations. So long as they choose autonomously, a person can choose
any sort of activity.
As we will see in Chapter 8, outcome autonomy has trouble dealing with
such motivational realities because it requires a clear picture of authentic
motivations in order to steer choosers toward choice outcomes that reflect those
motivations. How is that to be done for a variety of choosers, some of whom are
simply undecided or indifferent?
Nudges justified on outcome autonomy grounds also tend to view the
selection of “bad” choice outcomes as evidence of lack of rationality or, for our
purposes, a failure of process autonomy. There is no allowance under such an
approach for a person to autonomously choose bad choices. For example,
consuming unhealthy food is often thought to be contrary to someone’s best
interests. But if that if what they really want (the kind of person they want to be),
135
then it is possible for them to deliberatively and autonomously consume, for
example, junk food. As Rebonato notes, “[e]quating choices that we perceive as
“poor” with impulsive (System-I) choices [process autonomy deficient]—with
choices that we therefore feel authorized to interfere with—is an
oversimplification.”139 Another way to look at the issue is that not all process
autonomy deficient choices result in “bad” decisions (evaluated objectively), and not
all objectively bad decisions are the result of a lack of process autonomy. People can
autonomously make “bad” decisions, and that is the price of respecting autonomy.
Process autonomy, on the other hand, is neutral or blind to the content of
motivations – it is focused on the process of how those motivations are formed and
put into effect. Thus, nudges that respect choosers’ process autonomy are more
sensitive to the possible heterogeneity of motivations than those that solely respect
outcome autonomy. If a chooser is authentically undecided or indifferent, then
process autonomy respects that and may even help a chooser discover that fact and
put that motivation into action. Likewise, for choosers whose motivations are
context-dependent. If a process autonomy approach to nudging is adopted, there is
no concern that a choice architect might get the context wrong and nudge a chooser
towards an outcome that actually has less outcome autonomy than a chooser’s pre-
nudge choice.
139 Rebonato, supra note 112 at 380.
136
We will now consider the different types of motivational heterogeneity that
process autonomy approaches permit.
5.5. Non-Chooser-Regarding Motivations
Nudging, and non-nudge interventions, can affect the well-being of
individuals or organizations or society. Interventions that affect the former we can
broadly classify as chooser-regarding nudges and the latter two, non-chooser-
regarding nudges. The latter may affect the well-being of other individuals affected
by a person’s choices, but more often will mean other groups such as private
companies that benefit by choosers’ choices, and, most importantly, society.
Tracking those welfare effects, nudges can come in roughly three forms: (1)
libertarian paternalism; (ii) libertarian social welfarism; and (iii) libertarian
beneficence. The libertarian adjective is met in each case since to be a nudge, qua
nudge, the choice architectural intervention must satisfy the liberty of choice
constraint.
The difference between the three variants turns on which party benefits from
the nudge. The first one is chooser-regarding, the latter two non-chooser-regarding
(i.e. people other than the chooser benefit from the nudge). Libertarian paternalist
nudges are supposed to enhance the well-being of the chooser themselves (for a
more in-depth discussion of libertarian paternalism, see Chapter 7). Libertarian
137
social welfarist nudges (which Korobkin calls “libertarian welfarism”) are designed
to improve, not surprisingly, social welfare by encouraging pro-social behaviour
(e.g. environmental awareness, public health measures, organ donation, etc.).140
Finally, libertarian beneficent nudges are designed to improve the interests of
someone other than the subject of the nudge (the chooser) and society. This may be
the choice architect (e.g. private marketers/companies), a third party such as a
chooser’s family member, or perhaps even the government itself as an entity (if such
can be sufficiently disentangled from the idea of social welfare).
To clarify, nudges need not be definitionally restricted to those interventions
that affect choosers themselves. There is nothing in the definition of nudging that
precludes private enterprises or marketers from nudging customers. Nor is there
anything definitionally preventing governments from using nudge techniques to
improve broader social welfare. However, nudging, as a political theory, and as it is
often used in the policy literature, is directed most at choices relating to the well-
being of the chooser themselves (chooser-regarding).141
Regarding libertarian beneficent nudges, Sunstein and Thaler label this
category libertarian benevolence.142 A brief semantic distinction: the term
140 Russell Korobkin, “Libertarian Welfarism” (2009) 97:6 Calif Law Rev 1651. 141 The public also supports self-regarding nudges more than social welfarist (“pro social”) nudges: William Hagman et al, “Public Views on Policies Involving Nudges” (2015) 6 Rev Philos Psychol 439. 142 Sunstein & Thaler, supra note 13 at 1192.
138
“beneficence” must be preferred because Sunstein and Thaler’s “benevolence” in the
colloquial, everyday sense implies some sort of kindness, or charitable giving to
others. That is not the idea that is captured when we talk of private companies
nudging consumers so that those companies can increase revenue and profits. The
chooser/consumer subject to these nudges is likely not charitably benefiting the
company.
“Benevolence” seems then to impose some sort of mindset on choosers,
where none is likely present. If a nudge is well designed, choosers are likely not
even thinking of how altering their ultimate choice will benefit others (if they are
actually aware and thinking of their choices at all). The term “beneficence” on the
other hand, connotes less charitable intent; a simple conferring of a benefit
(intentional or otherwise). We will use libertarian beneficence for the remainder of
this work to denote those nudges that improve the welfare of third parties
(excluding society and the choosers themselves).
The second category of nudges - libertarian social welfarism – captures those
nudges that improve the welfare of society. Here we are talking of collective benefits
such as a cleaner environment, and a healthier and happier population. Social
welfare goods can be either aggregate benefits or benefits in their own right. For
example, a healthier population can potentially yield aggregate health cost savings
139
or productivity improvements. A healthier population is also arguably a good in its
own right.
Interventions that enhance the welfare of the chooser may also yield social
welfare benefits. Recall our earlier example of rearranging the order of food in a
cafeteria so that the healthier food is placed first. This is clearly targeted at chooser-
regarding welfare. The health of the chooser is certainly improved by choosing and
eating healthier items. But a healthier population also yields social welfare benefits
in terms of happiness, productivity and potentially reduced health or insurance
costs.
The final type – libertarian paternalism – is the primary theory of nudging. It
is an outcome autonomy approach centered on steering choosers towards particular
choice outcomes that they would choose if they were autonomous. However, the
inclusion of the term “paternalism” may cause some confusion when it comes to the
types of motivations that a person ought to have. Paternalism is generally
considered a chooser-regarding theory: a paternalist intervener interferes with a
person’s freedom in order to prevent self-harm to the person or to confer a benefit
on the chooser. The interests of others are generally not part of the paternalist
calculus.
Even though autonomy is indisputably about self-government, a nudge need
not be chooser-regarding in order to respect the process autonomy of choosers.
140
Whether a chooser’s autonomy is respected completely turns on the content of a
person’s authentic motivations. A chooser may legitimately have an authentic
motivation to help others or to contribute to society. Thus, a process autonomy-
enhancing nudge that empowered choosers to express that motivation in their
option selection would count as an enhancement to their process autonomy because
the choosers’ actions have become more reflective of their authentic selves. For
example, if a person is authentically an environmentally conscious person (i.e. that
is their authentic motivation), then nudging them with a libertarian social welfarist
nudge to select more environmentally-friendly options would not automatically
count as contrary to their process or outcome autonomy
This may seem to run counter to our earlier assertion that process autonomy
is neutral to the contents of a particular chooser’s authentic motivations. However,
such a reading would be inaccurate because process autonomy does not require a
chooser to hold a certain type of motivation, so long as whatever motivation they
have is respected by the nudge. So libertarian social welfarist nudges would be
permissible on a process autonomy basis if the affected choosers had social-
regarding motivations (provided the nudges also otherwise function so as to
preserve a chooser’s process autonomy).
One way to look at this fact is that because our idea of autonomy resides in a
chooser’s effective and authentic motivations, apparent other-regarding motivations
141
are in fact truly chooser-regarding. If a person genuinely wants to help others or
improve social welfare, then because doing so satisfies their autonomy, they have, in
a way, derived a benefit from it as well.
Outcome autonomy also allows choosers to hold primarily non-chooser-
regarding motivations. In the case of outcome autonomy-respecting nudges, a
chooser can legitimately be nudged towards choice outcomes that reflect non-
chooser-regarding interests, so long as they reflect that chooser’s authentic
motivations. In other words, since a chooser can have authentic non-chooser-
regarding motivations, a nudge toward non-chooser-regarding choice outcomes
would be beneficial on outcome autonomy grounds. For example, a chooser can be
nudged toward donating their organs,143 which presumably they would not benefit
from since they would be deceased. Or a tax scofflaw could be nudged to pay their
owed taxes, which they very clearly do not benefit from, but society obviously
does.144 In both cases, the choice outcomes (donating and pay taxes) will have high
outcome autonomy only if those are the choices that those choosers would select if
they had full process autonomy.
But as we will see in Chapter 8, it is extremely difficult to pinpoint a chooser’s
authentic motivations, let alone an entire group of affected choosers. Thus outcome
143 Johnson & Goldstein, supra note 61. 144 Katrin Bennhold, “Britain’s Ministry of Nudges”, The New York Times (7 December 2013), online: <http://www.nytimes.com/2013/12/08/business/international/britains-ministry-of-nudges.html>.
142
autonomy-respecting nudges such as the foregoing are inferior to process autonomy
approaches that do not require an inquiry into the particular content of authentic
motivations. In a process-based account where choosers are de-biased and informed,
allowing their authentic motivations to properly flow through to their choice
selections, we need not be concerned about whether nudges are reflective of a
chooser’s motivations because that should happen as a direct consequence of the
process improvements.
By improving process autonomy, we allow the chooser’s authentic
motivations to manifest themselves in a chooser’s choices. If tax filings increased as
a result of a process-based nudge (e.g. de-biasing), then we can be reasonably
confident that paying one’s taxes reflects the authentic motivations of that chooser.
At a minimum, if the nudge respected the process autonomy of choosers, we would
not need to wonder if an apparently non-chooser-regarding result was truly
representative of a chooser’s authentic motivations.
The foregoing illustrates both the power of autonomy for overcoming
traditional ways of conceiving and categorizing nudges, and also the power of
process autonomy’s neutrality vis-à-vis the content of choosers’ motivations. A
process autonomy account permits choosers to hold a wide variety of motivations,
some of which may be chooser-regarding and others of which may be non-chooser-
regarding. In fact, categorizing nudges as non-chooser-regarding or chooser-
143
regarding misses the point – who benefits from a nudge does not matter so long as
the autonomy of choosers is respected.
5.6. Context-Dependent Motivations
Another key benefit of process autonomy is that, because it is neutral to a
chooser’s motivations (i.e. it does not look past the authenticity criterion), it can help
nudge individuals with context-dependent motivations – motivations that change or
are formed depending on the particular details of a choice scenario – without
inquiring into the specifics of those motivations. Outcome autonomy, on the other
hand, requires a deeper inquiry into whether a person’s motivations are context-
dependent and in what form.
One of the main drivers of nudging is the idea that humans suffer from a
litany of action-intention gaps. People want to do X (their intention), and yet
somehow they end up doing Y (their action). In the language of autonomy, this
means that individuals have a particular authentic motivation, yet cannot or do not
act on that motivation at the option evaluation level, such that they end up selecting
an option from the opportunity set that fails to satisfy that motivation. Sunstein and
144
others call that a market failure.145 Not only is it an economic or public policy
problem, but it’s an autonomy problem.
There are two ways that nudges can address the action-intention gap. The
first is the outcome autonomy approach of nudging individuals to particular choice
outcomes that align with their motivations, but without providing choosers with an
actual participatory opportunity to make those choices (i.e. no process autonomy
improvements). As we will see in Chapter 8, one of the primary drawbacks of
outcome autonomy is that it tends to require that choosers hold stable, well-ordered
motivations. Outcome autonomy has trouble functioning when choosers have
context-dependent motivations, which many choosers happen to possess.
Process autonomy nudges, on the other hand, can easily nudge choosers to
more autonomous choice outcomes regardless of contextual factors influencing the
chooser’s authentic motivations. There are two reasons for this. First, process
autonomy is neutral to the particular motivations of choosers – process autonomy
effectively takes choosers as they are and simply allows them to choose in
accordance with that self, regardless of whether that self is context-dependent or
not. Second, and flowing from the former, is that there since there is no need to
inquire as to the particular content of a chooser’s motivations the indeterminacy
145 Sunstein, "Why Nudge", supra note 26; Peter D Lunn, “Are Consumer Decision-Making Phenomena a Fourth Market Failure?” (2015) 38 J Consum Policy 315.
145
caused by context-dependent motivations (e.g. “are we sure we looked at the entire
context?”) is avoided.
Let’s take a closer look at how process autonomy-respecting approaches
easily accommodate any context-dependent motivations of choosers. To begin, we
will illustrate how individuals’ context-dependence may change our understanding
of their motivations.
Consider again the case of consent to organ donation. In Canada, the US and
the UK, surveys reveal that roughly 90-95% of the population supports organ
donation.146 A little over half of those that “support” organ donation are “willing to
donate” or have already decided to donate. An argument could be made that these
people (~50% of the adult population) have a motivation to donate their organs. And
since the consent rate is closer to 28-30% (Ontario and the UK respectively), then
there must be a behavioural action-intention gap.147
There are two possible ways to read this data, differing on the interpretation
of whether there is an action-intention gap. The first approach is to take the 50%
“willing to donate” rate as a reflection of those individuals’ authentic motivations to
146 See e.g. Sheldon F Kurtz & Michael J Saks, “The Transplant Paradox: Overwhelming Public Support for Organ Donation vs. Under-Supply of Organs: The Iowa Organ Procurement Study” (1996) 21 J Corp Law 767 at 782; Ipsos Reid, Views Toward Organ and Tissue Donation and Transplantation (Ottawa, 2010); Organ Donation UK, Summary of Market Research Findings, (2015), online: NHS Blood & Transplant <https://www.organdonation.nhs.uk/statistics/latest_statistics/>. 147 “Registration Stats”, online: Trillium Gift of Life Network <https://beadonor.ca/scoreboard>; NHS Blood and Transplant, “Organ Donation and Transplantation: Activity Report (2011/12)” (2012), online: <http://www.odt.nhs.uk/uk-transplant-registry/annual-activity-report/>.
146
donate their organs. Thus, since their motivations don’t equate with the choice
outcomes (only half of those actually consent to donate), then we can infer that there
are some behavioural biases or decision-making errors at play at the option
evaluation stage. The problem here would be located with a chooser’s decisional
autonomy.
The other way to look at these survey results is to argue that there is in fact
no action-intention gap. This approach focuses on the supposed motivations of
those surveyed to donate their organs. By taking a broader, contextualized look at a
chooser’s motivations, we may see that perhaps only 28% of individuals had an
authentic motivation to consent to donate their organs within the organ consent system
as it is currently designed. Essentially, we can qualify the motivations with contextual
factors. As we will see, some contextual factors have legitimate claims to
representing a chooser’s authentic motivations in their own right, such that there
may not even be an action-intention gap requiring intervention.
There are many reasons why individuals may abstractly support organ
donation and say they are willing to donate, but not actually donate. These reasons
can help qualify our conclusions about the content of chooser’s motivations. First,
surveying individuals regarding their motivations does not necessarily tell us their
authentic motivations. Perhaps some individuals think they abstractly want to
donate but when they actually go ahead to do so, the thought stops them and, upon
147
further reflection and deliberation, they discover that perhaps their authentic
motivation is actually not to donate and that they were initially mistaken.
Second, we must draw a distinction from abstract motivations and concrete,
contextual motivations. Choice does not occur in the abstract. Abstraction is choice
without context, yet we know from behavioural insights that context matters a great
deal for our choices. More often than not, there are existing costs to options (not
necessarily monetary costs). There are also tradeoffs that must be made regarding
costs and benefits of particular options and between options within the same
opportunity set. And sometimes there are competing motivations on similarly-
ranked ordinal levels.
Contextual factors are often about costs, but sometimes they are also about
the presence of other options that end up amounting to opportunity costs. If you
choose a salad at a restaurant, then you are probably forgoing some other food
instead. There are even opportunity costs when deciding whether to consent to
donate one’s organs. If you spend the time and energy deliberating and putting
your choice into action, those are fewer finite resources available for other choices.
Finally, even if a resource is seemingly infinite, such as disgust, a chooser may still
not want to pay it given the benefits of a particular choice.
So, perhaps what we can say from the consent survey data then is that 28% of
choosers have a context-dependent motivation to donate (with the current
148
contextual factors), whereas 50% of people have an abstract motivation to donate. A
person may wish to consent to donate their organs, but if doing so involves the costs
of the current system (the time and effort spent to register, the mental anguish of
thinking about one’s own death) then their motivation is actually not to consent.
One way to phrase the motivations of the 50% of respondents is: “I have a
motivation to donate my organs if the cost (monetary and non-monetary) of
registering my consent is zero.” Since half of the people that have such a motivation
do not consent to donate and the costs of donation are above zero, we can distill that
statement further into three different possibilities:
(i) “I have a motivation to consent to donate at any reasonable cost, even
if larger than x.”
(ii) “I have a motivation to consent to donate if the cost is no greater than
x.”
(iii) “I have a motivation to not consent to donate regardless of the cost.”
All three of these groups are likely represented in the organ donation case.
There are people who will consent to donate even if the costs are large; their
motivations are not context-dependent in the current system. They are accounted for
in the 28% consent rate. Then there are those who do not want to donate regardless
of context (they would refuse to consent even if the cost were zero). They are not
accounted for in the 28% consent rate.
149
Then there are two possible groups arising from statement two. The first are
those who find the costs imposed by Canada’s current consent system as acceptably
minimal. These people therefore are accounted for the in the remainder of the 28%
consent rate. Finally, there are those who find the costs of Canada’s system as too
great. The system, as currently designed, does not satisfy their motivation. This does
not mean that their motivation then becomes to not donate – that is a further jump
that is not logically warranted. Instead, we should only restrict ourselves to saying
that their motivation to donate qualified by cost is not satisfied. They still hold that
motivation, so they do not become undecided, but that motivation is context-
dependent, and given the current context, is not satisfied.
The reality of context-dependent motivations is that they also reflect the
interconnected nature of individual decision-making. Think back to our feedback
loops in the stages of decision-making illustrated in Chapter 4. If 50% of
respondents to the survey say they would be willing to donate their organs, then
they have an abstract motivation. Then, at the option evaluation process when it is
time to put that motivation into action (when they decide whether they are actually
going to consent to donate or not), they discover the costs of that choice, what it
really means and what it entails. So that stage feeds back into the development of
their motivations, contextualizing their motivations from the abstract. Thus, we
could say that the abstract motivations are abstract because they haven’t yet
150
discovered the contextual factors that putting their motivations into action actually
entail.
We could also say that the context-dependency of motivations means that
individuals have weak motivations if they are subject to the whims of a particular
choice scenario.148 For example, that those who wish to donate (or not) regardless of
context (within reason) have a strong motivation while those whose motivation is
context-dependent have a weak motivation. As Sunstein notes, “if people have
intense preferences, they are more likely to opt out.”149 What this effectively means
is that default rules tend to take advantage of those with weak motivations.
The reason contextualization matters for our autonomy analysis is that
contextualized motivations should be considered higher-ranked or at least on the
same motivational level as abstract motivations. What is a better expression of your
values – that you want to be a healthy person, full stop, or that you want to be a
healthy person that also enjoys some great tasting (but sometimes unhealthy) food?
Arguably the latter is a better reflection of a person’s fundamental values and
beliefs. Thus, we ought to respect a person’s context-dependent motivation and we
would find that there is no action-intention gap that needs to be addressed through
nudging.
148 Sandbu, supra note 113. 149 Sunstein, "Choosing", supra note 12 at 77.
151
The foregoing works well for situations where the contextualization adds an
element that can also be considered to be a higher-order motivation. Being a healthy
person and being a person that is not health obsessive and enjoys some good food
could both be authentic motivations. Thus we need to inquire as to whether a
particular form of context is reflective of a persons’ authentic motivations or some
other driver of behaviour, possibly a lower-order desire.
There appear to be three types of context of relevance – context from
competing authentic motivations that ought to be respected, context from
competing lower-order desires that ought not to be respected, and context from the
external choice scenario (costs). The focus of most nudging is usually on the latter
two.
A person’s authentic motivation to donate their organs may be stymied by
their fear of thinking about their own death and a superstitious belief that they
shouldn’t take any actions contemplating death. Assuming those lower-order
desires do not reflect the kind of person they truly want to be, then we do have a
legitimate action-intention problem calling for intervention to close the gap.
Sometimes costs are not reflective of other higher-order motivations and they
truly are simple obstacles at the option selection stage of decision-making. Lower-
order desires interfering with higher-order motivations is one such case. For
example, in the organ donation case, we could imagine a person that has an abstract
152
motivation to donate their organs, but in reality that is contextualized with a
reference to effort and time-related costs. In this case, it would be a stretch to say
that a person’s authentic motivations are both to be the kind of person that donates
their organs and the kind of person that is lazy or afraid of thinking about death.
The latter instead tend to be manifestations of lower-order desires. These are true
behavioural biases at the option selection stage.
The third type of context is provided by factors endemic to the choice
problem such as objective costs. If a person has to spend time and money in order to
give effect to their abstract motivation, and the quantum of costs changes a persons’
motivations, we may consider that motivation to be properly context-dependent.
We will distinguish these kinds of costs from the kind of costs akin to fear and
superstition that are drivers of lower-order behaviour that may conflict with
authentic motivations.
Both context-dependent and abstract motivations may have legitimate claims
to representing a person’s authentic motivations. In that case, what conclusions for
nudging are we to draw vis-à-vis any action-intention gap? In the pure cases of
context-dependent motivations, it is impossible to say which one has the greater
claim. Instead, both represent an authentic motivation. We can prefer the abstract
because it is clearly a representation of a person’s autonomy – the kind of person
they want to be. Likewise, we can prefer the contextualized motivation since context
153
is reality, and assuming they are properly aware of the costs, cost considerations
may also represent the kind of person they want to be. From an autonomy
perspective, there is nothing inherently wrong with having weak motivations that
are swayed by relevant costs.
Here is where the value of process autonomy as opposed to outcome
autonomy enters the picture. Since both context-dependent and abstract motivations
are still authentic motivations, any nudge respects a person’s process autonomy by
ensuring those motivations are respected and expressed. For example, in the case of
cost context, choice architects should strive to lower the costs of a particular choice
or to increase the benefits such that the abstract motivation comes closer to reality
and is not “blocked” by contextual limitations. If there were a liberty of choice-
friendly way to improve the choice scenario such that contextual qualifiers are not
needed, then that would be an acceptable option.
So then, one way of nudging in this area is to change the motivation that is
called forth in a particular choice scenario from one that is contextualized to an
abstract version by eliminating the contextual actors. Eliminating the “costs” of
making the “healthy” choice will make it more likely that those abstract motivations
will come to fruition. Since both types of motivations – contextual and abstract –
have arguably equal claims to represent a chooser’s authentic, governing self,
nudging the chooser towards either one or the other respects a chooser’s process
154
autonomy. Removing obstacles at the option evaluation level is also consistent with
empowering choosers’ decisional autonomy.
Turning to the other two types of context, process autonomy nudges are also
found to be superior to outcome autonomy nudges. Contextual factors that bring
forth lower-order desires (e.g. biases, fears, laziness etc.) to stymie a chooser’s
higher-order motivations are obviously detrimental to autonomy since they
subordinate a chooser’s authentic self to their lower-order whims. Given our
hierarchy of motivations/desires such an outcome runs contrary to promoting
autonomy as self-government. Thus, process autonomy nudges that work, for
example, by de-biasing choosers, serves to enhance choosers’ overall autonomy.
Finally, regarding contextual factors that amount to competing claims to
represent a chooser’s highest-order motivation, a process autonomy approach that
focuses on helping choosers deliberate on their motivations in an unbiased fashion
also respects a chooser’s autonomy. Since neither motivation can be said to be
preferred on autonomy grounds, a nudge that is content neutral vis-à-vis
motivations (as process autonomy nudges are) does not prefer one over the other.
We can contrast this with an outcome autonomy nudge that requires choice
architects to determine which choice outcome to steer choosers’ towards. Since the
answer to that question is unclear (and likely indeterminate – see Chapter 8), an
outcome autonomy nudge would not be able to confidently recommend any
155
particular choice outcome as having greater outcome autonomy than in a pre-nudge
state (assuming that the pre-nudge choice is reflective of the context-dependent
motivation). In other words, an outcome autonomy nudge cannot nudge a chooser
away from the cheeseburger to the salad without preferring the abstract motivation
to the contextual motivation when both could be equally ranked.
In sum, the best approach for dealing with contextual factors are nudges that
respect or enhance the process autonomy of choosers. A singular focus on process
autonomy avoids the need to attempt to determine contextualized motivations
because appeal is never made to specific authentic motivations of choosers. Instead,
choosers’ motivations are taken as they are (provided they are authentic of course).
There is simply no need to qualify motivations, because process autonomy nudges
do not work by mimicking particular, hypothetical constructs of those motivations.
Indeed, most process autonomy nudges do not even work by affecting the authentic
motivations of choosers, instead focusing on helping choosers access those
motivations (whatever they may be) at the option evaluation level (i.e. decisional
autonomy).
Let’s take a moment just to emphasize that we have identified the cause of
the action-intention gap (if there is one) as in one of two locations, both within the
domain of process autonomy. This is another excellent reason as to why we should
be focusing on the process autonomy of choosers in order to drive long-term
156
behaviour change and not outcome autonomy where the attention of most nudging
efforts are currently focused. It is the “deficiencies” within the process autonomy of
choosers that drives the “poor” choice outcomes that is motivating nudging in the
first place. Simply changing choice outcomes by leaving those deficiencies in place
or exacerbating them will not lead to behavioural changes in the long-term. Instead,
as we will see in Chapter 8, outcome autonomy nudging is required on a constant
on-going basis.
To recap, this section has shown that what we consider a person’s
motivations to be is a complicated question that is extremely difficult to determine.
And if we cannot accurately determine those motivations (e.g. we cannot decide if a
person has a qualified or unqualified desire to donate), and which among
competing motivations ought to be ranked higher, deploying outcome autonomy-
based nudges would be imprudent and potentially unethical. Instead, we should
focus on process autonomy improvements, which due to their inherent content
neutrality, will respect choosers’ motivations no matter their level or type of context-
dependence.
5.7. Undecided Motivations
Just as process autonomy nudges accommodate the variety of context-
dependent motivations, so too do they accommodate choosers who are undecided –
157
those who have no particular authentic motivations to apply to a given choice
scenario. This fact once again flows from the motivational content-neutrality of
process autonomy. People can have all sorts of authentic motivations, including, for
some choice contexts, no particular motivations at all.
We can contrast this with outcome autonomy nudges that, because they
require a choice architect to set a particular direction, require deep inquiry into
whether a chooser is truly undecided. If a chooser is indeed authentically
undecided, then, under an outcome autonomy nudge, steering them in a particular
direction will count as a decrease in outcome autonomy (a hypothetical of what they
would choose if they had full process autonomy) because their process autonomy
takes them in no particular direction at all.
Process autonomy is the superior approach because it is blind to whether a
chooser is undecided or not. By equipping the chooser with both the opportunity
and tools for deliberative, informed and unbiased choosing free from undue
external influence, a chooser’s process autonomy is enhanced regardless of whether
that makes it more likely that a chooser will put their motivation into action. In
other words, it would not be a waste in the case of a chooser with an undecided
motivation to improve their process autonomy because it offers benefits outside of a
particular choice scenario – it ensures that they are in fact undecided and it equips
them to put their motivations into practice should they become decided. In fact, the
158
chooser may be less likely to be undecided as a result of a process autonomy nudge
because they can take the time to evaluate the choice scenario, weigh the tradeoffs
involved and potentially form a motivation on their own. All this stems from the
empowering nature of process autonomy.
Let’s return to the survey regarding organ donor consent to illustrate the
undecided chooser problem. Recall that roughly 95% of respondents support organ
donation but only 50% are willing to donate. We should not conclude from this data
that the 45% of respondents who support organ donation but are apparently not
willing to donate, in fact have a motivation to not donate their organs. What is far
more likely is that while some of the 45% authentically to not want to donate, many
are likely undecided on the matter.
Within that group there will be different levels of indecision, relating to the
causes of indecision. Some may be undecided because they have not had the
opportunity to consider the matter and weigh the costs and benefits. Providing them
with an opportunity for deliberation and informing them of the nature of the choice
problem may help choosers to discover that they do in fact have authentic
motivations regarding organ donation (i.e. that they are not undecided).
Rebonato posits that the reason some people may appear to be undecided is
that they are presented with choices irrelevant to their current activity (e.g.
159
consenting to organ donation when renewing their license).150 If that is true, then a
process autonomy-enhancing nudge may help to refocus the individual by
providing them with distinct opportunities for deliberation. Thus we might even
consider removing the donation question from the driver’s license realm as
enhancement to deliberation.
De-biasing choosers’ decision-making process (e.g. overcoming their fear of
thinking about death) may also cause them to discover they have an authentic
motivation to donate when before they were completely avoiding that discovery
process due to the presence of a behavioural bias. Some choosers may also
authentically be undecided and a process autonomy nudge respects that as well.
The benefits of process autonomy nudges in this area should be clear. A
process autonomy nudge may help to clarify individual motivations and ensure that
those who are decided on the matter are the ones who are “choosing” and those
who are truly undecided have that indecision respected. A de-biasing or
deliberation-promoting nudge does not force anyone into a choice outcome that
may not reflect their authentic motivations, including those with an authentic
motivation to be undecided. Unlike with outcome autonomy nudges, there are no
assumptions about the content of motivations required, since all motivations are
respected.
150 Rebonato, supra note 112 at 382-383.
160
Furthermore, process autonomy nudges in this one area (consent to organ
donation) may have longer-term impacts such that choosers’ process autonomy in
other choice scenarios is also improved. A person who is given the opportunity to
self-discover what they value may use that information to inform other motivations
and apply it to other choice problems. Likewise, the elimination of a bias in one
particular choice area may have carry-over effects to other choice problems.
Process autonomy nudges also respect those choosers who authentically are
undecided, a fact that outcome autonomy seems to have trouble with (see Chapter
8). We have already seen that nudges can work on altering individual’s higher-order
motivations; however, doing so may be disrespectful of the motivational autonomy
of choosers if the mode of intervention counts as an undue external influence. And,
if people genuinely have no relevant higher-order motivation, then such lack of
motivation ought also to be respected. If someone does not know what kind of
person they want to be, it is appropriate to leave the motivational side at peace so
that they can discover that for themselves.
One common response to the undecided chooser problem is that of mandated
or active choice. Sunstein argues that forced active choosing is indeed a nudge.151
The idea is that by requiring a chooser to choose, they have to turn their mind to the
problem and it effectively eliminates the opportunity to be undecided as the chooser
151 Sunstein, "Why Nudge", supra note 26 at 85.
161
must decide something. There is no default option since there is no option to do
nothing – some choice must be made in the circumstances. Leaving aside the liberty
of choice constraint question, mandated active choice is problematic from a process
autonomy perspective because individuals are required to choose when they may
have an authentic motivation not to choose. In order to avoid potentially infringing
on a chooser’s autonomy, any active choosing nudge must be well-designed,
including an option akin to “I choose not to choose.”
We need to accept, that for many reasons, some people will authentically be
undecided on a particular matter. Without that added “choosing not to choose”
option, forcing them to make a choice will violate their autonomy – they are forced
to participate when they simply do not want to. And it will also lead to a result
(choice outcome) that has little outcome autonomy. If a person’s authentic
motivation is to be undecided and they are required to select a choice outcome (or
are defaulted into one particular outcome), then any such choice outcome will have
little or no outcome autonomy since what they would select if they had full process
autonomy is none of the current choices (because they are undecided). Granted,
some outcome must result even if they choose not to choose, however, we must
primarily be concerned with the process of choosing. Thus, the answer to undecided
choosers is to not view them as a problem, but as an opportunity for empowerment
162
through further process autonomy enhancements and well-designed active choosing
nudges.
In sum, there is little downside to process autonomy nudges even in cases
where there are undecided choosers. The worst-case scenario is that a chooser
reaffirms that they are truly undecided on the matter and thus, their choice
outcomes would not change as a result of the nudge. At best, a chooser is
empowered with the tools, information and opportunity to discover that they do in
fact have a relevant authentic motivation (other than indecision) which they can
apply to the particular choice problem and perhaps other choice problems as well.
5.8. Meta-Motivations
Thus far we have seen that process autonomy is particularly valuable because
it easily accommodates heterogeneous motivations, including those regarding
undecided motivations. This section will discuss a similar situation where
individuals may also have motivations about other motivations or the decision-
making process itself. We can call these motivations about how one chooses or is
motivated “meta motivations,” or “meta-preferences,”152 or, as Sunstein calls them,
“decisions about decisions.”153 These are quite clearly higher-order motivations per
152 Smith, Goldstein & Johnson, supra note 111; Cass R Sunstein, “Legal Interference with Private Preferences” (1986) 53:4 Univ Chicago Law Rev 1129; Hamish Stewart, “A Critique of Instrumental Reason in Economics” (2008) 11:01 Econ Philos 57. 153 Sunstein, "Choosing", supra note 12 at 13.
163
our hierarchy, but what makes them slightly different is that they are motivations
specifically about choosing.
As we have seen, individuals may quite plausibly choose not to choose.154 But
there are two types of choosing not to choose – one is because a chooser is
undecided and another is that they intentionally do not want to decide. So as long as
the authenticity conditions are met, we ought to respect either instance of
individuals’ meta-motivation just as we would any other authentic motivation.
Simply because the subject of a motivation is choosing or the expression of other
motivations, it does not change the fact that respecting a person’s motivations is
akin to respecting their autonomy as self-governing persons. Just as we would
respect a person’s choices, so too should we respect their choices about choosing.
We can generally consider meta-motivations or “second-order decisions”155 to
be either higher-ranked or equally-ranked when compared to those motivations that
the meta-motivations are about. So if a person forgoes participating in a particular
choice scenario – either they refuse to indicate a choice or delegate that decision to
someone else – their autonomy is still respected because their meta-motivation is
respected. Even though a person may be forgoing direct participatory autonomy in
154 Ibid. 155 Ibid at 13.
164
a particular choice, having and exercising meta-motivations is also an exercise of
participatory autonomy.
There are many reasons why someone may have a motivation to no longer
put other motivations into action. A chooser may no longer care to spend any time
or effort on a particular choice. In that scenario, a chooser would not necessarily not
have an authentic motivation for whatever particular choice is at issue. Thus, they
would not be considered to be undecided or have no relevant motivations. Instead,
they may be considered to have a superior higher-ranked, or equally-ranked
motivation to not engage in that choice scenario. In this case a meta-motivation
effectively blocks another motivation from coming into play by forgoing
participation in the choice scenario, but autonomy is still respected by respecting
that meta-motivation.
Cass Sunstein rightly notes that having and exercising meta-motivations may
lead to an overall improvement in one’s autonomy by devoting our limited
cognitive resources to choices that really matter.156 Sunstein writes: “Sometimes we
exercise our freedom, and we improve our welfare, by choosing not to choose. That
choice opens up time and space for us, enabling us to focus on our real concerns.”157
This may have the effect of strengthening our process autonomy by creating that
156 Ibid at IX. 157 Ibid at IX.
165
time for deliberative and unbiased decision-making in line with our authentic
highest-order motivations.
It may make sense for some trivial, unimportant choices to delegate them to
others or simply adopt an “I don’t care” attitude so that we can focus on the more
important matters. In particular, Sunstein argues that delegation may be particularly
valuable and reflective of common sense.158 Not only does delegation free up our
limited decision-making capacity for other opportunities, but it may also lead to
better choice outcomes since one of the problems we may encounter at the option
evaluation stage is the ability to accurately select the choice outcome that best
reflects our motivations.
For some types of choices, this requires a lot of information and specialized
skill that a chooser may simply not possess, or it may not be efficient to acquire and
exercise. For example, Sunstein suggests that it may be better for others to make
some choices for us, such as a taxi driver selecting the best route or asking the chef
to select their recommended dish.159 Both types of delegation (a type of meta-
motivation) may work if the chooser’s motivation is to get to where they are going
the fastest or to eat the chef’s best dish. In these cases, the chooser’s meta-motivation
would be to optimize their motivations and they can do so by delegating their
158 Ibid at IX. 159 Ibid at X.
166
decision-making power to others. However, other choosers may still want to select
the route – they derive pleasure from that – or they value the independence from
choosing their own dishes to eat. Neither delegation nor independence are
inherently problematic for autonomy so long as they both reflect choosers’ authentic
motivations.
Unfortunately Sunstein takes this slightly too far in arguing that, for the same
reason, we ought to have our choices limited in some circumstances.160 This
implicitly suggests that the determination of whether to not choose (i.e. have choices
limited) belongs to another as opposed to the chooser themselves. That would not
be conducive to their autonomy. Only a chooser can have and act on meta-
motivations. The decision of whether to forgo certain choices or delegate them to
others must be confirmed by the chooser on a case-by-case basis. Otherwise, we are
inferring meta-motivations where there may not be any, and thus disrespecting that
person’s autonomy.
So we know that meta-motivations can be valuable, and we know that they
ought to be respected as reflections of a person’s autonomy. The question now
becomes what role should nudges play? Nudges have limited applicability to most
cases of meta-motivations or choosing not to choose. In order to be of relevance,
there would have to be some choosers with an authentic meta-motivation that
160 Ibid at IX.
167
nudging can affect. However, whether this group of choosers not wanting to choose
is of sufficient size/incidence is debatable.161
If a person does have an authentic meta-motivation, a nudge may
legitimately target that motivation. Consider the case of the diner who asks the chef
to select their best dish – there is no nudge at play there if initiated by the diner. One
possible method to engage nudging would be to put “Chef’s Special” beside certain
dishes on the menu, so that if a chooser did not want to choose they could easily
select the recommended options. Or, the Chef’s Special could really be a “Chef’s
Surprise,” with no particular dish indicated except for whatever the chef had
decided to cook that day. But that nudge only works if the individual’s motivation is
to choose the recommended option.
It’s important to note that that kind of nudge – indicating a recommended
option or Chef’s Surprise – may actually enhance the process autonomy of the diner
because it provides them with information that may be of relevance if their
motivation were a meta-motivation. It also provides choosers with information that
may lead to their discovery that they have a meta-motivation – “I didn’t realize that
there were chef’s recommendations or a surprise, but now that I know, I’d rather
have what the chef recommends.” Thus, process autonomy nudges can help
choosers not only to discover their meta-motivations (e.g. that they want to
161 Rebonato, supra note 112 at 371.
168
delegate) but also to put those motivations into action by providing relevant
information. This shows that once again process autonomy nudges work with
heterogeneous motivations.
We need to distinguish between nudges that may cater to meta-motivations
by informing choosers of the recommended option and those that set the
recommended option as the default, where there previously was no default. The
former respects the process autonomy of choosers while the latter does not, even if
choosers are also informed that the default is the recommended option. Introducing
a default creates the expression of new biases (loss aversion, inertia) creating a
reduction in decisional autonomy.
We can contrast defaults with what Sunstein calls “simplified active choice.”
Unlike in typical default scenarios where a chooser does not have to make a choice
or even turn their mind to the issue, simplified active choice does require some form
of choice. However, the key is that the choice is a meta-choice. The chooser is,
“asked whether they want to make a choice – or instead rely on a default.”162 This is
akin to a diner being able to order anything on the menu or simply let the chef
decide (Chef’s Surprise) – both are options. This method preserves the process
autonomy of choosers because it provides an opportunity for choosers to place
either their meta-motivation or regular motivation into action.
162 Sunstein, "Choosing", supra note 12 at XIII.
169
This, and the preceding few sections, have shown that the one of the key
benefits of process autonomy is that it accommodates heterogeneity of chooser
motivations. Process autonomy nudges work with the reality that choosers may
have social-regarding motivations, context-dependent motivations, be undecided or
have motivations about motivations. All different types are more respected under
process autonomy nudges than nudges driven by outcome autonomy.
5.9. Challenges of Process Autonomy
As we have seen thus far in this chapter, process autonomy is quite beneficial,
both as a stand-alone ethical concept and in contrast to the outcome autonomy
approach of many nudges. But process autonomy is not without its challenges. This
section will explore two main challenges relating to its effectiveness as a nudging
tool and with our construction of decisional autonomy and its apparent insistence
on living a fully deliberative life.
It can be no surprise semantically that, on the whole, outcome-driven nudges
are likely to be more effective at achieving particular changes in choice outcomes. If
policymakers or other choice architects want to change individual behaviour they
can more easily do so by ignoring process autonomy concerns. Process autonomy is
generally indifferent to outcomes. People can choose whatever they want as long as
170
they authentically and independently want to want it and provided they have the
opportunity to reflect on those wants.
For example, we have already seen that defaults, which work not through
process autonomy but a lack thereof, are powerful forms of choice architecture. It is
simply easier to get individuals to choose a particular choice if it is the default (e.g. if
people don’t turn their minds to it). If a nudge were instead to inform choosers,
provide them a cooling off period or similar opportunity for deliberative reflection,
it is doubtful that the effective consent rate for organ donation would be 99.98% as it
is in Austria. So, process autonomy nudges are not terribly effective for achieving
particular outcomes. But, since our concern is the ethics of nudging and not
effectiveness of nudging, we can set those concerns aside.
A far more difficult problem for process autonomy to address is found in the
way we have constructed our understanding at the decisional autonomy level (i.e.
option evaluation stage). Our conception of process autonomy values deliberation
over impulsiveness, unbiased informed “thinking” over biased automatic
“reacting.” On the whole, such a conception is consistent with our understanding of
autonomy as self-government, but in some cases, there may be value in biased,
instinctive reacting that our definition of autonomy is short-selling.
171
Cass Sunstein calls these the “legitimate claims of System 1” – that there is
value in making automatic, instinctual decisions.163 These legitimate claims appear
to have implications for our ideal of autonomy as self-government, which we have
claimed requires participatory process autonomy. Such claims also pose a problem
regarding choosers who may be authentically motivated to be ruled by their System
1 and with the practicalities of using our process autonomy approach.
First, we must consider the legitimate claims of System 1. Cass Sunstein is
right that living life through System 1 and not completely deliberative is what gives
life its meaning in many cases.164 For example, love is a powerful emotion that we do
not normally deliberate on and yet it guides our lives in many different areas. How
can we reconcile our desire to respect the choices made in the name of love (e.g.
“falling” in love) and yet insist on process autonomy at the same time?
Likewise, because we are so burdened with daily choices,165 and have little
experience in many of them, we rely substantially on the use of heuristics or rules of
thumb to govern many of our decisions. Gerd Gigerenzer is a pioneer in this area
and he argues that sometimes heuristics yield better and faster judgments than those
163 Ibid at 154-159. 164 Ibid at 155. 165 For example, Wansink notes that we face up to 200 daily choices regarding food alone: Brian Wansink, “Mindless Eating: The 200 Daily Food Decisions We Overlook” (2007) 39:1 Environ Behav 106.
172
based on additional computation and information.166 In other words, heuristics are
not always “bad,” and they may even be sensible given their evolutionary
underpinnings.167
This also makes practical sense as it would simply be impractical to demand
that every decision we make be unbiased and deliberative. If we had to deliberate
on each one before deciding, we would be stuck in decisional paralysis, given the
hundreds, if not thousands, of choices we make each day. We have to, as a practical
matter, delegate some choices to our “automatic’ System 1 – in a way, being
mindless helps us survive.168
So a process autonomy approach that insisted on deliberation and
information may be both impractical and also ineffective. Note that the effectiveness
we are talking about here is not about nudging per se, but about the effectiveness in
our decision-making yielding the result we want it to yield (i.e. the choice outcome
that best reflects our motivations – outcome autonomy).
This qualifier may provide some insight into how our ideal of process
autonomy can work with the reality of heuristic-based decision-making. Our
valuation of process autonomy (particularly when compared with outcome
166 Gerd Gigerenzer & Thomas Sturm, “How (far) can rationality be naturalized?” (2012) 187 Synthese 243 at 243. 167 Ibid at 248. 168 Wansink, supra note 165.
173
autonomy) is predicated on the assumption that it is the best method to achieve
autonomy as self-government. Having, identifying and exercising one’s authentic
motivations is the core of self-government. Making many life decisions by reference
to heuristics does not necessarily contradict that core. If a heuristic makes it more
likely that a chooser will select a choice outcome that meets their authentic
motivations, then such a style of choice is not incompatible with process autonomy.
Our focus of process autonomy on deliberation, information and cognitive de-
biasing is such because those are the areas where individuals encounter the most
problems in their decision-making causing choice outcome selections inconsistent
with their motivations.
Process autonomy is also concerned with individuals participating in their
decisions and on this front heuristics and System 1 reacting also scores some points.
Compare the following scenarios: a chooser being defaulted into consenting to
donate their organs and a chooser who ticked the box to donate with little
deliberative thought – he saw the box and his gut told him to donate. While the
latter may not score high on the key pillars of process autonomy, it is certainly
superior to the former where there was no participation at all. With defaults, we
cannot even be sure that the chooser engaged in any choosing process (heuristics or
deliberative).
174
However, we must be mindful of the fact that when evaluating decision-
making styles for autonomy implications, we are concerned with whether a person
is both selecting an option consistent with their motivations (outcome autonomy)
and in a manner that they recognize as part of their own lives (process autonomy).
Recall that authentic motivations are often about not just want we want as choosers,
but what we want to want; the kind of person we want to be. So if we want to be the
kind of person that thinks about our decisions and chooses in accordance with our
motivations, then a nudge that encourages deliberation and unbiased decision-
making respects that sense of autonomy.
If, on the other hand, a chooser authentically wants to be ruled by their
instincts or be spontaneous, then an autonomy preservation approach must respect
that as well. It could be said that forcing those individuals to deliberate or become
informed is a “paternalistic” interference with their autonomy.169 This may be the
case, for example, if they have an authentic motivation to live impulsively. A nudge
that imposes deliberation (e.g. cooling off periods) or informs them of their biases
would seem to counter their authentic motivations to live impulsively.
So, how does a process autonomy approach address this predicament? The
simple response is that process autonomy-based nudges would not count as
interferences with such motivations. The goal behind process autonomy nudges is to
169 Dworkin, supra note 35 at 123.
175
help individuals verify their motivations and select the best choice outcome that
matches those motivations. If an individual is authentically motivated to live
impulsively, then a nudge that attempts to de-bias them, may de-bias them vis-à-vis
the particular choice scenario, but it should also have offsetting benefits to their
process autonomy in allowing them to access and verify that their motivations are to
live impulsively. This dual effect of a process autonomy nudge on the rare subset of
choosers who authentically want to live instinctively, renders the effect of the nudge
on their net process autonomy as neutral. Such a nudge effectively asks of the
chooser – “how do you know you want to live instinctively regarding this particular
choice unless you are provided with an opportunity to reflect on that and act
accordingly?”
We must also remember that we are only using process autonomy as an
ethical evaluation tool. Such a goal does not mean that we must have process
autonomy for every choice. If you are an experienced driver and let your “auto-
pilot” System 1 brain do most of the work, that will not change under our approach
because we are not using process autonomy to evaluate all behaviour, or demand
that all behaviour accord to some minimal deliberative process. This thesis simply
posits that if we are going to nudge individual choices, we must do so, at a
minimum, in a way that autonomy is not detrimentally affected. In other words, if
we are going to engage in intentional choice architecture then such choice
176
architecture (nudges) should be performed on an ethical basis and doing so
necessarily requires that we preserve the process autonomy of choosers.
5.10. Conclusion
This chapter has shown that process autonomy is lexically prior to outcome
autonomy because the former values participation. Added benefits of process
autonomy stem to a large extent from being a concept that embraces participation
and out of a respect for the heterogeneity of motivations. A nudge that preserves or
enhances the process autonomy of choosers can accommodate choosers with a wide
variety of motivations whether they are ostensibly non-chooser-regarding, context-
dependent, or even a lack of motivation as in the case of undecided choosers. A
nudge that works through the process autonomy of choosers also respects any meta-
motivations. In sum, process autonomy has many benefits with limited drawbacks.
177
Chapter 6
Evaluating Nudge Effects on Process Autonomy
6.1. Introduction
Since process autonomy is a tool for the ethical evaluation of nudges, this
chapter will examine the different ways that nudges can affect the process
autonomy of choosers. We will begin by setting the standard for evaluating those
effects – to be considered ethical, a nudge must not decrease the process autonomy
of any choosers. That is, process autonomy ought not to be subject to aggregation
calculations, where gains for the majority of choosers are traded-off for process
autonomy losses suffered by a minority of affected choosers.
Also, because process autonomy has two constituent parts, each necessary for
our understanding of autonomy as participatory self-government, our preservation
standard applies at each level. Though a single nudge can have effects at either or
both levels of process autonomy, we will consider the effects independently. First,
we will examine how a nudge may affect a chooser’s motivational autonomy. Such
nudges will generally construct motivations (where none presently exist) or will
involve the manipulation of existing motivations. External influences on
motivations can be either covert or overt, the former being unethical while the latter
are ethical.
178
Next we will consider how nudges can affect a chooser’s decisional
autonomy. We will see that most current or proposed nudges operate through
counter-biasing (reduction in decisional autonomy) or re-biasing (decisional
autonomy-preserving). Process autonomy-enhancing nudges would instead improve
the decisional autonomy of choosers by providing information (both about the
choices and the choosing process), de-biasing cognition, and/or providing an
opportunity for deliberation. Such nudges will generally give precedence to a
chooser’s System 2 thinking. A variant of this style of nudge is active choosing
which encourages participation and not passive decision-making that is associated,
for example, with defaults.
6.2. Autonomy Preservation Standard of Evaluation
Before examining in detail how nudges can affect the autonomy of choosers,
we need to know the standard on which to evaluate any such effects. Akin to our
liberty of choice analysis in Part I, the standard should be one of preservation –
autonomy preservation. Using this framework we can evaluate the autonomy ethics
of each nudge individually, a case-by-case approach endorsed by Sunstein.170 This
will allow us to find some nudges, or types of nudges, as being more ethical than
170 Cass R Sunstein, "The Ethics of Nudging" (2014) [unpublished, available online: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2526341].
179
others on autonomy grounds, depending on their impact on choosers’ process
autonomy and outcome autonomy.
Under our standard of preservation, only those nudges that preserve or
enhance a chooser’s autonomy should be considered ethical. Note that we will use
the term “preserve” to encompass both those effects that enhance autonomy and
those that leave it unchanged. The key effect is that, at a minimum, ethical nudges
should not reduce a person’s autonomy. However, because of the lexical priority of
process autonomy, in most cases we can further revise this standard to one of process
autonomy preservation.
Autonomy as we use it here is not an ideal, perfectionist standard. Nudges
need not make a chooser completely autonomous. Indeed, such a goal is likely
unattainable due to the many social influences we have on our lives. The question is
not whether a person’s decision as a result of a nudge (or even prior to the nudge) is
autonomous or non-autonomous. Indeed, such an endeavor would likely prove
fruitless as there may not be an obvious demarcation line for determining if
decisions are autonomous or non-autonomous.171
Instead, we can view autonomy or autonomous choosing as a continuum.
Some decisions can be said to be more autonomous than others. For example, those
171 Sabine Müller & Henrik Walter, “Reviewing Autonomy: Implications of the Neurosciences and the Free Will Debate for the Principle of Respect for the Patient’s Autonomy” (2010) 19 Cambridge Q Health Ethics 205.
180
choices made more deliberatively (e.g. System 2) will be said to be more
autonomous (and have more process autonomy) than those made with unconscious
thought (e.g. instinctual reflexes of System 1). Thus, when evaluating the ethical
effects of nudges, we must ask whether a nudge provides choosers with a more
autonomous choice, or a moment of greater process autonomy.
In contrast to our minimal autonomy preservation standard, a different
approach would be to adopt a higher standard, such as one of overall autonomy
maximization. Using such a high standard, nudges wouldn’t just have to not reduce
autonomy, but would have to increase it to the maximum extent possible given the
particulars of the choice scenario. In most cases this would mean that choosers
would have to be made almost fully autonomous. This would require nudges to
remove pre-existing barriers to autonomy, such as undue external influences that
existed independent of, and prior to, the nudge intervention. Likewise, no nudge
would be permitted to operate through non-deliberative, lower-order desires, even
if those desires are already what is driving chooser behaviour pre-nudge. These are
demanding requirements and would most certainly shrink the entire concept of
nudging down to but a few exceptional uses.
Furthermore, adopting an autonomy maximization standard is tantamount to
requiring nudging. This may be particularly true regarding governments who
181
arguably have a duty of respect to their citizens.172 Our autonomy preservation
restraint instead is meant to be just that – a restraint on which nudges from the
menu of available nudges an autonomy-minded choice architect should select. The
simple availability of a nudge that is evaluated as ethical on autonomy grounds
does not make the deployment of said nudge an ethical imperative – there is no
ethical obligation on choice architects to nudge. However, nudges must, at a
minimum, not make the autonomy of choosers worse off.
Finally, the standard of where to draw the line on nudging and autonomy
may change depending on the expectations of the choice architect. For example, we
may expect that if a nudge is initiated by our government they adhere to a higher
standard of autonomy ethics than, for example, private choice architects.
Policymakers and choice architects may also wish to adopt different standards
depending on the circumstances of particular choices.173 However, for both public
and private nudging, the minimal standard should still be one of process autonomy
preservation.
172 See e.g. White, supra note 6. 173 For an excellent discussion of domain-specific determinations of nudging in the context of public health see: Alexander Morgan Capron, “Are There Public Health Domains in ‘Domain-Specific’ Health Nudging?” (2015) 15:10 Am J Bioeth 47.
182
6.3. No Aggregation Among Affected Choosers
It is also possible that a nudge may affect different choosers’ process
autonomy differently. In particular, we need to consider the possibility that a nudge
may cause a decrease in the process autonomy of some choosers while enhancing
the process autonomy of other choosers. If this occurs, how are we to balance the
two effects, if at all?
The rule must be that we cannot aggregate the process autonomy effects of
nudges across affected choosers. We cannot balance positive and negative effects on
process autonomy in search of a net beneficial effect. This rule means that a nudge
that decreases the process autonomy of any affected chooser would be deemed
unethical.
Process autonomy cannot be subject to aggregation because autonomy, in our
participatory, self-governing sense, is individualistic. Subjecting it to aggregation
amounts to something like utilitarian balancing which would sacrifice the
individual to a social aggregate of utility. One person’s autonomy cannot be
sacrificed for another’s autonomy – we cannot sacrifice process autonomy losses of
the few for the process autonomy gains of the many. Doing so would run counter to
the very individualistic (self-governing) idea of autonomy.
183
Fortunately, the incidence of nudges that have diverging process autonomy
effects is likely to be extremely low. Most process autonomy-enhancing nudges take
the form of providing a chooser with an opportunity to exercise their process
autonomy either at the option evaluation stage by de-biasing, informing or
providing space and time for deliberation, or at the motivation formation stage by
blocking undue external influences. On the flip side, nudges that cause a reduction
in process autonomy do so by removing similar opportunities for process
autonomy, which will likely affect all choosers.
It is difficult to think of any situation where a nudge taking any of the forms
above would also cause a decrease in a chooser’s process autonomy. The only
possible situation would be regarding meta-motivations – choosers’ motivations
about choosing. If a chooser had an authentic motivation to be governed by their
lower-order desires, then a nudge that provided them with information (what they
would view as irrelevant information) or required deliberation may be considered
as stymying that individual’s authentic motivations. Consider for example, a
hypothetical cigarette package that informed potential smokers of how bad they are
at calculating probability and provided them with the true probability of dying from
tobacco smoke. That would count as a process autonomy enhancement for many
choosers since it neutralizes one or more biases affecting their decision-making (e.g.
optimism bias) by providing relevant information.
184
Now, what are the effects of such a nudge on a chooser with a meta-
motivation to be governed by their lower-order desires without recourse to
statistical reasoning? They simply don’t want to be the kind of chooser that lives
their life according to probabilities; they prefer to live in the moment and damn the
future. A de-biasing informational nudge as the one above would likely not have a
detrimental effect on any choosers’ process autonomy, even a chooser who didn’t
want the information. Such a result stems from the nature of process autonomy-
respecting nudges as being focused on opportunity for process autonomy. Process
autonomy-enhancing nudges do not require a chooser to exercise process autonomy;
they need only be provided the opportunity to do so, should they so desire. If a
chooser did not want to be governed by the relevant information, then they need not
consider the information and incorporate it into their decision-making process.
Thus, so long as nudges that enhance the process autonomy of choosers do so by
providing only opportunities, we need not be concerned about detrimental effects
on other choosers’ process autonomy because it’s simply not a realistic possibility.
6.4. Independent Effects on Motivational & Decisional Autonomy
As mentioned, nudges can either increase or decrease a chooser’s process
autonomy or leave it unaffected. While our standard of ethical evaluation concerns
the overall process autonomy of choosers, nudges tend to affect either motivational
185
autonomy or decisional autonomy alone. Thus, it would be prudent to evaluate the
effects of nudges on process autonomy by evaluating its individual effects on
motivational autonomy and decisional autonomy. Only by examining the
independent effects can we hope to have a sense of process autonomy as a whole.
To count as ethical, nudges must preserve both (and separately) the
motivational autonomy and decisional autonomy of choosers. A nudge does not
respect a person’s autonomy if it makes it easier for a chooser to access and live in
accordance with their higher-order motivations, but at the same time the nudge
counts as an undue external influence such that the motivations a chooser is now
putting into practice are not authentically their own. Likewise, a nudge that is an
ethically permissible overt influence on a chooser’s motivations, but that makes it
less likely that those authentic motivations are accessed, is not autonomy respecting.
How then we do address a scenario where a chooser has excellent decisional
autonomy (unbiased and informed deliberation) as a result of a nudge, but that
decisional autonomy is used to access inauthentic motivations as they existed pre-
nudge? A nudge of this nature may perversely result in a chooser behaving even less
autonomously because they are now better able to apply those motivations that are
inauthentic (but that inauthenticity is not caused by the nudge). The answer is that
the nudge still causes an increase in a chooser’s process autonomy, but it also causes
a decrease in outcome autonomy. Process autonomy is increased because a chooser
186
is provided with an opportunity to effect their motivations with greater capacity.
Outcome autonomy is decreased because the selected choice outcome is further
from reflecting a chooser’s authentic motivations.
The key is that we only use process autonomy as an ethical evaluation tool –
that is, not to create the perfect conditions for choice, but to evaluate nudges. This
means, as with all our approaches, we are evaluating the state of choosers’ process
autonomy post-nudge against their pre-nudge state. If process autonomy (however
little it may be) is preserved at both the motivational and decisional levels, then the
nudge has preserved a chooser’s process autonomy. We do not inquire into the
authenticity of pre-existing motivations since we are only evaluating the effects of
nudges.
Consider the following example. Let’s imagine that a chooser pre-nudge has
an inauthentic motivation for I. Let’s also say that the chooser (hypothetically) has an
authentic motivation for A. If our chooser had full process autonomy they would
have a motivation for A and end up selecting options that reflect A as well (high
outcome autonomy). Alas, our chooser actually has little process autonomy. Not
only is their motivation inauthentic, but they are also biased at the option evaluation
stage so they cannot accurately put their motivation (as inauthentic as it is) into
action. So our chooser with the inauthentic motivation for I ends up selecting A. So
while we can say that the chooser has little process autonomy, we can say that they
187
coincidentally ended up with high outcome autonomy since the latter reflects their
hypothetical authentic motivations (and the matching of motivations to choice
outcomes is a gauge of outcome autonomy, not process autonomy).
Despite the matching between authentic motivations and outcomes, this
scenario is clearly problematic from a process autonomy perspective. Thus, a nudge
may be implemented that enhances process autonomy, however doing so may cause
a decrease in outcome autonomy. For example, we can imagine a nudge de-biased a
chooser at the option evaluation stage. This has the effect of improving the chooser’s
decisional autonomy so that they can choose more in line with their motivations.
The nudge also does not affect the motivational autonomy of the chooser. The end
result of the nudge is that the chooser now selects choice I, matching their
inauthentic motivation for I. Thus, we have a decrease in outcome autonomy but an
increase in process autonomy.
Adopting such a nudge would be ethical since it preserves the process
autonomy of the chooser. However, it may not be desirable from the perspective of
the choice architect if they are concerned about improving welfare. Since our
standard of autonomy preservation is not an obligation – choice architects are not
required to improve process autonomy wherever possible – our choice architect may
rightly choose to leave the chooser as they are and refrain from this kind of nudging
at the decisional autonomy level.
188
Perhaps the choice architect could instead target a chooser’s motivational
autonomy. A nudge could cause a chooser to revise their motivation from
inauthentic I to authentic A (without any concomitant effects at the option
evaluation/decisional level). Such a nudge would also count as meeting the process
autonomy preservation standard. Even though a chooser remains biased for
example, and is having difficulty implementing his motivations through to option
selection, his motivational autonomy has been improved, and thus his process
autonomy.
The foregoing examples have been the easy cases – an increase in one level of
process autonomy with no effects on the other level. While most nudges may
operate in such a manner, not all do. Some nudges will affect both the motivational
and decisional levels of process autonomy, possibly in diverging directions. In those
cases, there are two options for our ethical evaluation – either try and balance the
effects (possibly giving precedence to one stage of process autonomy) or adopt the
autonomy preservation standard at each level.
The appropriate methodology is to apply our standard of autonomy
preservation independently to each level of process autonomy – decisional and
motivational. Any decrease in either level counts as an unethical decrease in process
autonomy, even if such a decrease is seemingly “offset” by an increase in the other
type of process autonomy.
189
Consider again the case of a chooser with an inauthentic motivation “I.” The
chooser is currently minimally biased in their option evaluation process so they tend
to end up with choice outcomes consistent with I. Let’s imagine that we introduce a
nudge that has dual effects – it causes a chooser to revise their motivation to one
that is more authentic to themselves (we’ll call this authentic motivation “A”). But
the nudge also causes a change at the option evaluation level by working around a
chooser’s deliberative reasoning process, such that they are less likely post-nudge to
select choices consistent with their higher-order motivations so they end up still
selecting I. So it appears we have an increase in motivational autonomy but a
decrease in decisional autonomy. How are we to determine the overall process
autonomy effect of such a nudge?
One approach would be to try to balance the effects. However, such an
approach would require some way to measure or weigh the different effects, but
such effects do not lend themselves to quantification. How are we to quantify a
decrease in decisional autonomy achieved through the expression of a new bias? Is
it equivalent to or less than an increase in motivational autonomy? Such a
determination is impossible.
Another approach would be to assign lexical priority to motivational
autonomy since having authentic motivations is the necessary starting point to
autonomous self-government. In that case, our nudge example would seem to count
190
as autonomy preserving since it improved a chooser’s motivational autonomy. But
that doesn’t seem quite right – it completely discounts the autonomy that resides in
putting motivations into action; after all we are interested in effective self-
governance, not just nominal self-governance. Nor can we place lexical priority on
decisional autonomy, since, as we have seen in the preceding examples, evaluating
and selecting options using an inauthentic motivation is also not consistent with
self-governing as we understand it.
Instead, the preferred approach must be to evaluate the effects of a nudge on
a chooser’s motivational and decisional autonomy separately – our ethical standard
of autonomy preservation must be applied independently to each stage. The failure
of a nudge to satisfy the standard at either stage renders it unethical since it fails to
respect the process autonomy of choosers. Thus, any reduction in motivational
autonomy or decisional autonomy renders a nudge unethical.
This approach best captures our idea of autonomy as effective self-
government. A chooser must have both authentic motivations and be able to put
those motivations into action to be autonomous. In other words, while both
motivational autonomy and decisional autonomy are necessary for process
autonomy, either alone is insufficient. So it is possible that a nudge may cause an
increase in motivational autonomy, but that increase (as unquantifiable as it is) may
be for naught if decisional autonomy is also reduced. Having completely authentic
191
motivations but without any way to put those motivations into action is not
autonomy as effective self-government. The component of process autonomy that is
reduced by a nudge is controlling for the purposes of process autonomy
preservation.
6.5. Nudges & Manipulation
Process autonomy is concerned with a chooser choosing or behaving in line
with their authentic motivations and the decision-making process flowing from
those motivations. Perhaps the greatest threat to such process autonomy is
manipulation. Manipulation has also been flagged as one of the most significant
ethical problems with nudging.174
Manipulation is the perversion of a chooser’s decision-making process such
that the authenticity of a chooser’s motivations and option selections are interfered
with. Handelman writes: “[m]anipulative behavior, which is an uninvited
interference in another’s decision-making, distorts the normal process of
discovering, forming, and realizing preferences and priorities.”175
174 See e.g. White, supra note 6; T M Wilkinson, “Nudging and Manipulation” (2013) 61:2 Polit Stud 341; Jason Hanna, “Libertarian Paternalism, Manipulation, and the Shaping of Preferences” (2015) 41:4 Soc Theory Pract 618; Pelle Guldborg Hansen & Andreas Maaløe Jespersen, “Nudge and the Manipulation of Choice: A Framework for the Responsible Use of the Nudge Approach” (2013) Eur J Risk Regul 3. 175 Sapir Handelman, Thought Manipulation: The Use and Abuse of Psychological Trickery (Santa Barbara: ABC-CLIO, 2009) at 38.
192
Manipulation is troubling for our participatory understanding of autonomy
as effective self-government. Manipulation throws into doubt the authenticity of a
chooser’s effective motivations and thus the authenticity of their choices if they are
choosing in line with those motivations (as they ideally should under our
understanding of process autonomy). Any choices arising from that manipulation
result in a chooser being less self-governing (i.e. less autonomous). This makes sense
since manipulation is inherently about the subordinating of one person’s will to that
of another. Pettit would say it is about power – the domination of one (the choice
architect) over another (the chooser).176
Given the negative relation between manipulation and process autonomy, it
is sensible to use manipulation as a categorical watchword for detrimental effects of
a nudge on a chooser’s process autonomy. Nudges that are inconsistent with a
chooser’s process autonomy are manipulative and therefore unethical on our
autonomy preservation standard. This determination is independent of any
effectiveness evaluation of the nudge because, as Wilkinson notes, “manipulation
would be prima facie objectionable even if it made people better off.”177
Note that the way we use manipulation here is restricted to functional effects
on autonomy. We are not ascribing any intent on the part of the choice architect,
176 Philip Pettit, “Freedom as Antipower” (1996) 106:3 Ethics 576. 177 Wilkinson, supra note 174 at 345.
193
even though Wilkinson, for example, argues that manipulation necessarily must be
intentional.178 Nudges may be intentionally manipulative or unintentionally
manipulative, but intentionality is not determinative for our purposes; autonomy
effects are. For example, if a nudge influenced a person’s motivations in a manner
inconsistent with authentic self-government, then they are manipulated regardless
of the intent of the choice architect. Any nudge that is found to be manipulative will
be deemed to be unethical because manipulation necessarily implies a reduction in
process autonomy because it is a reduction in participatory self-government.
Many nudges operate through manipulative techniques because
manipulation as a form of choice architecture tends to pass the liberty of choice
constraint we explored in Part I. Handelman rightly notes that, “manipulation
intrudes on autonomy without limiting freedom.”179 Manipulation does not target
the opportunity sets of choosers, but their decision-making process – namely the first
two stages of motivation formation and option evaluation. Not surprisingly, the
latter distinction turns on whether a chooser is manipulated into being motivated a
certain way, or manipulated into choosing in a certain way.
Manipulative nudges can change choosers’ motivations so that they are
caused to be motivated in a way that the choice architect wants them to be
178 Ibid at 347. 179 Handelman, supra note 175 at 14.
194
motivated, without the acceptance of such motivations by the chooser themselves.
That is, their higher-order motivations are products of a manipulative process. For
example, instead of having a motivation for healthy living, a chooser is manipulated
into having a motivation for frugal living. As we will in the next section, if a chooser
is aware of the influence and accepts the alterations of their motivations, then the
nudge is simply an influence and not manipulative.
Altering motivations should (in theory) filter down to the selection of options
and realization of subsequent choice outcomes. However, such nudges may be
effective only in cases where a chooser has full decisional autonomy so that their
motivations are effective at guiding their option evaluation and selection.
Otherwise, the changed motivation will end up just as many of our motivations
currently are – not actually governing our behaviour because of behavioural biases
and heuristics at the option evaluation stage of decision-making.
Manipulation as an interference with autonomy can also occur at the
decisional autonomy level. Manipulation at that level has the effect of getting
choosers to do (choose) what the choice architect wants them to do (choose), without
regard to the chooser’s own higher-order motivations (what they authentically wish
to do). No knowledge is needed on the part of the manipulative choice architect as
to the chooser’s authentic motivations, because the choice architect does not want
the chooser to select in line with that motivation. Instead, the manipulative nudge
195
may cause a chooser to select in line with a less authentic motivation (e.g. lower-
order desire or a higher-order motivation incorrectly determined to be most relevant
to the choice scenario).
On the other hand, nudges that are respectful of chooser’s process autonomy
will be found not to be manipulative, but autonomy-promoting. For example, at the
decisional autonomy stage, de-biasing, informing and providing an opportunity for
deliberation all serve to ensure that individual’s option evaluations and selection are
guided by their own authentic motivations, not the motivations of others.
The next two sections will consider in greater depth manipulation by nudge
of a chooser’s motivational autonomy and decisional autonomy. Those sections will
delineate what makes nudges autonomy-respecting at each stage of decision-
making.
6.6. Nudging Motivational Autonomy
Few nudges directly target a chooser’s motivations. However, few though
they may be, motivation-targeted nudges tend to very powerful and can also have
the longest-lasting impacts, achieving real behaviour change over time. For example,
affecting whether a person authentically wants to live a healthy life may be
attempted, say for tobacco cessation purposes, but may have far-reaching
implications for the fabric of that choosers’ life.
196
Nudges can only affect motivations at the motivation formation stage by
changing the content of motivations – either by creating motivations where none
were present, or shifting the content from one motivation to another. Since nudges
are about changing choices (outcomes), the purpose of motivation-targeted nudges
is to change the content of motivations so that choosers who choose in line with
their motivations will select different options and thus end up with different choice
outcomes.
Note that for the purpose of this section we will assume that a chooser has
full decisional autonomy (i.e. process autonomy at the option evaluation stage).
Whatever a chooser’s governing motivation is, that motivation is accessed at the
option evaluation stage and applied to resolve a particular choice problem. We are
only concerned here with the autonomy at the motivation formation stage of
decision-making, not the application stage (option evaluation).
Nudges that work at the motivational autonomy level can be either
manipulative or non-manipulative. Manipulative nudges work by lessening the
authenticity of choosers’ higher-order motivations. On the other hand, non-
manipulative nudges leave intact, and in some cases, even improve, the authenticity
of a chooser’s higher-order motivations.
Recall that in our initial discussion of autonomy as self-government (Chapter
4), authenticity had two determinants. First, authenticity went to the ordinal ranking
197
of a chooser’s motivations such that a higher-order motivation was more authentic
than a lower-order desire. The higher-order motivation had higher authenticity
because it more accurately reflected a chooser’s true self – their fundamental values
and beliefs. The second determinant of authenticity was independence. A person’s
motivations that were sourced form an undue external influence and could not be
said to belong to the chooser were inauthentic. After all, independence from others’
machinations is what will make a chooser most authentic – that their own
motivations govern their actions, not the motivations of others.
Our requirement for independence would, at first blush, appear to eliminate
any nudges that affect the content of a chooser’s higher-order motivations, for then
the motivations would not be literally independent. It would also be problematic
because realistically no one’s motivations could be claimed to be strictly
independent, since we are all subject to a variety of background and environmental
influences on our motivations.
We get around these problems by targeting the determination of
independence not on the absolute presence of external influences, but on the undue-
ness of those influences. We are only interested in independence insofar as it is in
ingredient in authenticity. Some external influences will not threaten the
authenticity of a chooser’s motivations. If influences (nudges) on a chooser’s higher-
order motivations maintain the authenticity of a chooser’s motivations, then the
198
influence is not undue. Such a nudge would thus preserve the motivational
autonomy of the chooser, even though they may not be strictly independent.
This shows that we need not achieve perfect independence, authenticity or
autonomy. Instead, we have a far more modest goal – a nudge simply cannot reduce
a chooser’s autonomy. In the context of motivational autonomy, this means that a
nudge cannot make a chooser’s motivations less authentic.
Whether or not an external influence jeopardizes the authenticity of a
chooser’s motivations (i.e. whether the nudge is manipulative) depends to a large
extent on how they function – namely, whether they alter motivations overtly or
covertly. The latter we can classify as undue external influences, or, in other words,
manipulative nudges that fail to satisfy the autonomy preservation standard. On the
other hand, nudges that operate on the content of a chooser’s higher-order
motivations in an overt manner should not be considered manipulative or, at a
minimum, considered a permissible form of manipulation that is not autonomy-
infringing.180
In fact, overt nudges will generally either preserve, or even enhance, the
motivational autonomy of choosers. Overt external influences on a chooser’s
motivations are those that are visible to the chooser, and easily resistible.181 This
180 See e.g. Wilkinson, supra note 174 at 346. 181 See e.g. Felsen & Reiner, supra note 92 at 8; Rebonato, supra note 112.
199
permits the chooser to consciously evaluate the influence and determine whether or
not to incorporate that influence into their higher-order motivational structure.
Influences that are willingly incorporated into a persons’ motivations (possibly even
causing them to change their motivations) would not negatively affect the
authenticity of those motivations.
Clearly a chooser must be aware of an influence in order to possibly
incorporate it into their motivational hierarchy. Seeing an influence and willingly
changing one’s motivations as a result amounts to effective consent to the external
influence (i.e. nudge). The influence becomes internalized as a result of a successful
overt nudge that is endorsed by the chooser.182 The chooser thus makes the new,
altered motivation authentic to themselves. But they cannot do so if they are not
aware of the influence; without awareness the authenticity condition is not met. A
chooser cannot possibly accept and incorporate an influence if they are not aware of
the influence in the first place. Awareness is thus key for our authenticity
determination, and is the primary distinction between covert nudges (unacceptable
influences) and overt nudges (acceptable).
In order for choosers to be aware of a nudge on their motivational autonomy,
the nudge must be transparent. Such a nudge should not operate through deceptive
182 Frank Mols et al, “Why a nudge is not enough: A social identity critique of governance by stealth” (2015) 54 Eur J Polit Res 81.
200
or misleading means. This requirement reflects the ideas underlying Sunstein and
Thaler’s deployment of Rawls' transparency or publicity principle.183 Note that
Sunstein and Thaler use transparency more for justifiability at a policy level than for
the individual ethical evaluation of nudges. However, the underlying principle is
the same – individuals ought to have a choice as to whether to change their
motivations, and having a choice begins with awareness. From awareness and
choice in this case, flows authenticity.
A second requirement for overt nudges, and one flowing from the first
requirement of awareness, is that they be easily resistible at the motivational level.184
A chooser should easily be able to resist the influence if they do not want their
motivations to be altered as a result of the nudge. This should be easily satisfied by
most nudges, since to definitionally qualify as a nudge in the first place a chooser’s
opportunity set needs to be preserved. Resistibility is also closely related to the idea
of reversibility or avoidability – that influences on motivations should be easily
reversible or avoidable.185 Both resistibility and reversibility/avoidability require
knowledge and awareness of the nudge in the first place (i.e.
overtness/transparency). A chooser cannot resist what they do not know is
183 Thaler & Sunstein, "Nudge", supra note 1 at 247-248. 184 See e.g. Saghai, supra note 46; Rebonato, supra note 112; Gerald Dworkin, “Lying and Nudging” (2013) 39:8 J Med Ethics 496. 185 See e.g. Rebonato, supra note 112; Cass R Sunstein, “The Storrs Lectures: Behavioral Economics and Paternalism” (2013) 122:7 Yale Law J 1826.
201
happening, nor can they reverse a change in motivations if they are unaware that
their motivations have been changed.
Consider the example of an educational nudge that promotes healthy living –
it extolls the benefits of healthy living and how it can fit into a still-enjoyable
lifestyle. A chooser, seeing this, can either incorporate the influence into their
motivational structure or not. If being a healthy person is the kind of person the
chooser wants to be, then they would internalize the new motivation, embracing it
as part of their own, new identity. The nudge in such a case may strictly count as an
external influence, but clearly it does not decrease the authenticity (and thus
autonomy) of a chooser’s motivations. In the case of educational nudges that are
overt and easily resistible, we would not say that a chooser has been manipulated
because the external influence is not undue.
It is clear then that overt nudges are ethically permissible as they do not
implicate the authenticity of a chooser’s motivations. Overt nudges, by their obvious
nature, provide choosers with an opportunity to accept or resist, affording choosers
with an opportunity to participate in and govern their own lives. If a nudge solely
had effects at the motivational level and influenced motivational autonomy in an
overt manner, such a nudge would qualify as preserving the process autonomy of
choosers, and thus pass our ethical evaluation.
202
In contrast to overt nudges affecting a chooser’s motivational autonomy,
there are covert nudges. Covert nudges typically operate by deceiving or misleading
the chooser – they succeed by somehow keeping the chooser in the dark about their
origins and authenticity of their motivations. Covert nudges also have the result of
frustrating a chooser’s authentic motivations because they are replaced with
motivations that do not share the authenticity characteristic. As Felsen and Reiner
note, covert influences, “act unconsciously [and] potentially subvert the individual’s
higher-order desires and thus would constitute such a threat [to autonomy].”186
The key with covert nudges on a chooser’s motivations is that they cause a
chooser to want something (be motivated in a certain way) without the chooser’s
participation in developing or even holding those motivations. As a result of the
covert influence, a chooser has a certain set of motivations that they have been
manipulated to have – the chooser is not aware of why they have those motivations.
As Handelman notes: “It is not possible to be a victim of manipulation and, at the
same time, to know about it.”187
Since a person is not aware of the covert influence, any revised motivations as
a result of that influence are inauthentic to the chooser.188 After all, authenticity
186 Felsen & Reiner, supra note 92 at 8. 187 Handelman, supra note 175 at 8. 188 See e.g. Per-Anders Tengland, “Behavior Change or Empowerment: On the Ethics of Health-Promotion Strategies” (2012) 5:2 Public Health Ethics 140 at 144.
203
requires that a chooser embrace the motivation as part of their identity, reflecting
their fundamental beliefs and values. A chooser subject to a covert influence has
been manipulated without a chance to accept or resist the influence because they
were not even aware it was occurring. Thus, the internalization or endorsement
necessary for a motivation to be reflective of a chooser’s authentic self never occurs –
the opportunity is simply not granted by the covert nature of the nudge. The change
in motivations occurred without the participation of the chooser which is a strong
indicator that the nudge is failing to respect a chooser’s process autonomy. As a
result, the new motivations would more properly be ascribable to the choice
architect than to the chooser’s authentic self.
Let’s consider the example of subliminal advertising that works on a
chooser’s sub-conscious motivations. For example, we can imagine a nudge that
works subliminally to make smoking appear to be a vital part of a good life. A
chooser would be motivated to be a smoker as a result of the subliminal nudge but
they may not be aware of the source of that motivation. They certainly would not be
aware of why their motivations had changed – in other words, what caused their
change in motivation (if they are even able to perceive that they had changed).
Subliminal advertising is a widely prohibited (and generally frowned upon)
commercial activity for a reason. Out of respect for a chooser’s process autonomy
(their autonomy as choosing participants in their lives), a chooser must be able to
204
resist any external influences.189 Because subliminal advertising is covert, it is
irresistible. A chooser cannot resist that which they are not aware of. But, just as
they cannot resist the motivational change, they also cannot embrace the change and
internalize it. Thus, the new motivations as a result of the motivational nudge would
not count as authentic to the chooser. Therefore, the effect of the nudge is a
reduction in motivational autonomy (and process autonomy) of the choosers,
rendering the nudge unethical.
Thus far we have seen that nudges can affect motivational autonomy either
by decreasing it (manipulative, covert nudges) or preserving it by shifting the
content of the motivation (e.g. educational nudges). Let’s return for a moment to our
educational nudge regarding the motivation to live a healthy lifestyle, as it can help
illustrate how nudges can also, if the circumstances are right, increase the
motivational autonomy of choosers.
As a result of a healthy living educational nudge, our chooser comes to
authentically value and believe in living a healthy life. Now, suppose that the
chooser pre-nudge did not hold many values – they were living life fairly aimlessly
without any motivation about the kind of person they wanted to be. A nudge then
189 The elimination of all external influences would also achieve the same goal, but is quite clearly a practical non-starter. Instead, chooser autonomy is protected by ensuring that choosers have the opportunity to resist influences that do not align with their fundamental values and beliefs.
205
effectively constructs a motivation where there previously was none.190 Or, if you
subscribe to the theory that our beliefs and values are innate, they may simply have
not discovered the kind of person that they are, or want to be, and education brings
out that those values.
Thus, motivational nudges may be helpful in aiding individuals to discover
their higher-order motivations. This kind of discovery lends authenticity to any
resulting motivation because it brings those motivations to the fore for individual
reflection, endorsement and acceptance (or resistance, in which case the remaining
motivations (or lack of motivations) are authentic to the chooser). As a result of
either type of nudge, the chooser now has, or is aware of, an authentic motivation
with which to govern many of their life choices. The authenticity of their
motivations have been increased because it was near zero pre-nudge. Thus we could
say that their motivational autonomy has been increased as a result of the nudge.
Constructing motivations is a frequent use of nudges, deployed both in the
commercial context and in the public education realm (e.g. public service
announcements). Regarding the former, Sugden notes that, “markets do not simply
respond to pre-existing preferences; instead, market interaction may be part of the
process by which preferences are formed.”191 Consider the classic quote often
190 Paul Slovic, “The Construction of Preference” (1995) 50:5 Am Psychol 364. 191 Robert Sugden, “Why incoherent preferences do not justify paternalism” (2008) 19 Const Polit Econ 226 at 230.
206
attributed to Henry Ford - “If I had asked my customers what they wanted, they
would have said a faster horse.”
The same principles of overt versus covert nudges outlined above apply in
cases of motivation construction. So long as the construction of motivations occurs
in an overt manner such that choosers can accept or resist the change, a chooser’s
motivational autonomy is preserved. If the change is accomplished without the
knowledge and participation of the chooser in accepting and internalizing the
change, then the nudge does not preserve the motivational autonomy of the chooser.
Even though we have assumed full decisional autonomy for the purposes of
exploring the underlying concepts of manipulating motivational autonomy, the
findings of behavioural insights actually show that nudges that affect choosers’
motivations may be fruitless in affecting change because of behavioural obstacles at
the option evaluation stage. As useful as it may be to change the content of choosers’
motivations to content that best reflects the desired choice outcome (e.g. healthier
choices), without addressing any behavioural issues at the option evaluation stage
those motivations will not be effective motivations. Thus, while we may be able to
say in certain cases that nudges are effective at enhancing (or preserving) the
motivational autonomy of choosers, we cannot necessarily say that doing so will in
either case be effective at changing choices. It is for this reason that most nudges
207
forgo motivational targeting and instead operate at the option evaluation stage of
decision-making, as we will see in the following section.
6.7. Nudging Decisional Autonomy
Most nudges do not work on a chooser’s motivational autonomy, instead
predominantly affecting a chooser’s decisional autonomy at the option evaluation
stage of decision-making. And just as a chooser’s motivational autonomy can be
interfered with by manipulative nudges, so too can a chooser’s decisional
autonomy. Instead of being manipulated into having a particular motivation (and
not knowing that you have the motivation, or why), at the option evaluation stage of
decision-making a chooser is manipulated into selecting a particular option. This is
accomplished by affecting how choosers select between their choice options.
Manipulative nudges fail to respect chooser’s participatory process autonomy
by subordinating them as an autonomous chooser to the machinations of the choice
architect. This is accomplished by denying the chooser the opportunity to
participate in their own decision-making by preventing them from fully accessing
and applying their own authentic motivations. Such a nudge is designed to get the
chooser to make a choice the choice architect wants them to make, by affecting the
way the chooser chooses (as opposed to affecting their motivations). Different
nudges affect a chooser’s decisional autonomy in different ways, but the principle
208
throughout is constant: an effect on a chooser’s ability to reflect on their choices and
to evaluate those options in light of their authentic motivations.
Unlike manipulation of a chooser’s motivational autonomy, manipulation of
decisional autonomy need not occur covertly. The determining factor is whether a
chooser is able to evaluate their choices using their authentic motivations. A chooser
may be aware that they are being denied that opportunity (by an overt nudge), but
cannot necessarily do anything to address that fact. As well, studies have shown
that even informing choosers that they are being intentionally defaulted into
particular choice options (i.e. that they are biased and their biases are being used),
doesn’t necessarily change behaviour.192 While informing choosers that they are
being manipulated is necessary to overcome manipulative decisional nudges, it is not
sufficient. Informing a chooser that they are biased may not, without more, overcome
the bias. Defaults are very “sticky” because of behavioural biases, not just because
consumers are unaware of the power of defaults.
Nudges can respect, manipulate (negatively affect) or positively affect a
chooser’s decisional autonomy in three areas that mirror the elements of decisional
autonomy: deliberation, information and biases. A chooser with full decisional
autonomy would have the opportunity to deliberate/reflect on their options, possess
the relevant information about the options and their own behaviour necessary to
192 George Loewenstein et al, “Warning: You are about to be nudged” (2015) 1:1 Behav Sci Policy 35.
209
conduct such deliberation and possess the cognitive capability (the right mental
mindset) to deliberate (i.e. a lack of behavioural biases).
Deliberation, and informed and unbiased choosing promote active,
participatory choosing and self-governance by a chooser’s higher-order motivations.
They are the essential elements of decisional autonomy. Nudges that preserve or
promote any of those elements (without off-setting decreases in the other elements)
will be said to preserve or enhance the decisional autonomy of choosers. Note that
those three factors work together, they are not always distinct concepts. For
example, deliberation captures to some extent the idea of unbiased thinking, and it
also requires information. Information is useless without an opportunity to reflect
on such information and vice versa.
Any nudge that causes an overall decrease in the three elements of decisional
autonomy can be labeled manipulative and an interference with a chooser’s
decisional autonomy. Such nudges would thus be considered unethical in
accordance with our process autonomy preservation standard. If a chooser lacks
those elements of decisional autonomy they are unlikely to be able to choose in
accordance with their authentic motivations. Instead, their lower-order desires,
instinctually responding to the choice scenario will govern their option selections.
And as we know, being governed by lower-order desires is not a road to
autonomous living.
210
Felsen and Reiner best sum it up when they write that the decision-making
process, “requires access to all of the relevant information and sufficient time and
neural resources to select the option associated with the best predicted outcome: the
advantages and disadvantages of each option are weighed, and the most valuable
one is selected.”193
We will assume for this section that a chooser possesses full motivational
autonomy relevant to each choice scenario, so that there is a clear motivation that
can be accessed and applied to each choice scenario. This assumption will allow us
to evaluate just the impact of a nudge on a chooser’s decisional autonomy. This
reflects the nature of most decisional nudges as being neutral as to the content of
choosers actual motivations. What choosers authentically “want” is not of concern;
only how they decide is what matters (because how they decide determines what
outcome they decide upon).
6.7.1. Deliberation
We will begin our analysis of the effects of nudges on decisional autonomy
with deliberation. In most instances of decision-making, deliberation is a necessary
component to acting with decisional autonomy. Deliberation means that a chooser
193 Felsen & Reiner, supra note 92 at 6.
211
has an opportunity (time and space) to reflect on the opportunity set before them in
order to evaluate those options in light of their authentic motivations.
Deliberative choosing is seen when choosers have time to reflect on their
authentic motivations and/or are prompted to do so. In the language of Kahneman’s
dual-process theory, deliberation would be System 2 thinking: slow but
considered.194 Notice the highlight on thinking. This can be contrasted with non-
deliberative System 1 instinctual reacting. Selecting an option by rapidly reacting to
stimuli will likely not reflect a chooser’s authentic motivations, but their lower-order
desires.
Several approaches have been recommended containing elements that
promote respect for a chooser’s deliberation. For example, the “THINK” approach
encourages deliberative deliberation in a public democratic sense.195 Similarly, the
“STEER” approach is focused on promoting introspective deliberation.196 However,
unlike nudging, these approaches operate more directly. STEER involves holding
workshops where individuals are taught about biased decision-making in the hope
of empowering choosers to consider their decisions more deliberatively. Both of
these approaches are good approaches – they focus on capability enhancement in
194 Kahneman, supra note 20. 195 Peter John, Graham Smith & Gerry Stoker, “Nudge Nudge, Think Think: Two Strategies for Changing Civic Behaviour” (2009) 80:3 Polit Q 361. 196 Jones, Pykett & Whitehead, supra note 3 at 177.
212
the area of chooser’s critical capacity by encouraging deliberation. However, these
approaches are too direct, not-scalable, and do not provide a useful framework for
evaluating many of the nudges that already exist or have recently been proposed.
In our understanding of the importance of deliberation, several themes
emerge that can be deployed in our ethical evaluation framework: choosers must be
actively aware that they are facing a choice in order to deliberate on that choice; and
the choice scenario should be designed so as to call forth their reflective,
deliberative, thinking faculties (i.e. System 2 selves) as opposed to their automatic,
reactive instincts (i.e. System 1 selves). A choice scenario that appeals to a chooser’s
lower-order desires as opposed to their higher-order motivations will not respect a
chooser as an autonomous participant in their own decision-making. This means
that nudges that bypass a chooser’s deliberative evaluation of their choices will not
be found to preserve the decisional autonomy of choosers.
6.7.1.1. Awareness & Active Choosing
In order to deliberate, choosers ought to be clearly aware that they face a
choice. This necessitates real awareness, not merely token, implied awareness. For
example, consider the organ donation default (or any other default for that matter).
In both the opt-in and opt-out regimes, there are bound to be some choosers who are
not aware that by doing nothing they are effectively making a choice. The default
213
rules which deem a choice from inaction do not treat all choosers as deliberative
participants in their lives. A chooser cannot be said to deliberate on a matter and
apply their authentic motivations if they are not aware that they are facing a choice
scenario at all.
Despite the fact that defaults often operate on a non-deliberative basis, the
nudge of flipping a default from opt-in to opt-out is itself still preserving of
choosers’ decisional autonomy. This quirk arises from our standard of autonomy
preservation. In a situation where only the substantive content of the default is being
switched (as opposed to, for example, the absolute presence of a default), choosers
who are unaware of the default after the content switch were the same ones not
aware in the first place. Thus, for those choosers who never had
deliberation/awareness to begin with, the nudge of flipping the default did not
affect their decisional autonomy. This is why in most cases of nudging by default
where the nudge affects the content of the default rule (and not its presence), the
overall process autonomy of choosers is unaffected. We will see this in greater detail
in the section below on “Re-Biasing.”
This can be contrasted with a situation where a default rule is introduced
where, prior to the nudge, there was no such default rule. Perhaps this was because
“active choosing” was required – our chooser had to make some choice. Consider an
employer who requires all new employees to indicate as a condition of their
214
employment whether they would like to join the company pension plan. A chooser
must choose either “yes” or “no.” There is no option to not choose. In that active
choosing scenario, all choosers have some level of deliberation because they must
consider the choice scenario in order to gain employment. If the employer instead
switches to a default-based system of opting-out, overall deliberation by choosers is
likely to be reduced since some choosers in a default-based system stick with the
default because they are not aware of the default. Thus, such a nudge would be
considered a reduction in decisional autonomy, failing to preserve the process
autonomy of choosers.
Active choosing should only be considered an improvement in decisional
autonomy when compared to systems that rely on passive choice (e.g. defaults).197
For example, a nudge that switched the arrangement of choices presented in an
active choosing scenario would not count as an improvement in decisional autonomy
on an deliberation basis, since the awareness of the choice is unchanged by the
nudge. Instead, such a nudge would count as having no effect on decisional
autonomy (though it may affect outcome autonomy). On the whole, active choosing
is respectful of choosers’ decisional autonomy because it provides choosers with an
opportunity to deliberate.198
197 Sunstein, "Choosing", supra note 12 at 87. 198 See e.g. Pelle Guldborg Hansen, “Should We Be ‘Nudging’ for Cadaveric Organ Donations?” (2012) 12:2 Am J Bioeth 46.
215
Let’s visualize how the form of these different nudges may appear from a
chooser’s perspective.
Figure 6 – Defaults vs. Active Choosing
Default Active Choosing
Would you like to donate your organs?
� Yes � No
(if no selection is made you are defaulted into “no”)
To receive your license, you must select one of the following:
� I want to donate my organs
� I do not want to donate my organs
Would you like to opt-out of donating your organs?
� Yes � No
(if no selection is made you are defaulted into “yes”)
As Figure 6 above shows, the choices are technically the same – to donate or
not to donate. However, there is a significant distinction that is not visible in the
above example – that choosers in the default system have an implicit choice to not
choose. Indeed, the primary critique of active choosing (right hand column) is that it
is choice-requiring, which may be considered paternalistic.199 For the subset of
choosers who would rather not choose, a switch from a default to active choosing
system could be considered opportunity set reducing (and thus not a nudge due to
199 Sunstein, "Choosing", supra note 12 at 113.
216
the liberty of choice constraint) – they no longer have the option to not choose.
Choosers may choose not to choose for a variety of reasons, but if they must make
some choice (e.g. to donate or not), then their opportunity set has been materially
altered.
There is a way around this predicament and the solution lies in how we
design the active choosing scenario. Instead of simply mandating choice between
the same two choices (e.g. to donate or not), a third choice should be added – “I
choose not to choose.”200 That option would preserve the opportunity set as it
existed pre-nudge – individuals could still choose not to choose. Adding an option
to choose not to choose keeps the intervention in the nudging realm because it
preserves the opportunity set (when going from a default-only regime to an active
choosing regime, and without a switch of the outcome that obtains from not
choosing).201 Granted, the result of choosing not to choose is the same as the “do
nothing” default option, but we are not so much concerned with results (outcomes),
but the process used to realize such outcomes. Choosing not to choose is a greater
200 This is akin to Sunstein’s “simplified active choosing” where choosers can opt out of active choosing: Ibid at 115. 201 Note that there would still be a default in such a system, since some result must obtain if the chooser chooses not to choose (i.e. do they donate their organs or not?). In order to count as opportunity set preserving, the adoption of an active choosing regime should maintain that default rule, otherwise, for example, the option to not choose and not consent is substituted for “not choose and consent.” Such a substitution eliminates a pre-existing option and is inconsistent with the liberty of choice constraint we explored in Part I.
217
exercise of process autonomy than being defaulted into the same result, possibly
because you weren’t even aware you faced a choice.
Figure 7 below illustrates that an active choosing nudge as seen in the second
column preserves the full opportunity set that existed in the default system, where
the “choosing not to choose” option was implied from passivity.
Figure 7 – Active Choosing (accounting for meta-choices)
Active Choosing (basic) Active Choosing (best/meta)
To receive your license, you must select one of the following:
� I want to donate my organs
� I do not want to donate my organs
To receive your license, you must select one of the following:
� I want to donate my organs
� I do not want to donate my organs
� I do not want to choose
A key distinction is that choosing not to choose is different from simply not
choosing. Unlike the default system, the active choice “choosing not to choose”
category does not include those choosers who are unware that they had a choice
because a lack of awareness is not possible in this active choice scenario. An active
choice system accounting for meta-choices (i.e. choosing not to choose) ensures that
all choosers’ opportunity sets are preserved and that all choosers also have some
level of deliberation because they are all, by necessity, aware of the choice they must
make. Even if the deliberation is minimal along the lines of “I will put this choice off
218
for later,” that is a far greater level of deliberation than not even considering a choice
at all because one is simply not aware of it. The nudge could go even further and
add a fourth option – “I am currently undecided.” In the context of driver’s license,
choosers could even be asked to revisit their decision upon renewal.
Active choosing also has other benefits. Because some sort of decision is
required, active choosing has the effect of overcoming the inertia that is so
problematic in default-based systems.202 Active choosing nudges also signal to
choosers that the decision involved is important, and that they should take care and
time to reach a decision.203 In this way, active choosing regimes have a signaling
effect that may encourage deliberation.
Active choosing also overcomes some the problems faced by choice architects
when selecting what option should be the default, which may require epistemic
determinations of what the majority of choosers would choose, or what welfarist
good is trying to be promoted. Those determinations may also be subject to
decision-maker biases and influences.204 Finally, Sunstein notes that active choosing
permits dynamic revisiting of choice over time whereas default rules tend to be
more static.205
202 Ibid at 96. 203 Ibid at 90. 204 Ibid at 99. 205 Ibid at 101.
219
While active choosing may be preferential on the grounds of respecting the
process autonomy of choosers, there may be other drawbacks to adopting active
choosing systems. Sunstein in particular is worried that active choosing may cause
decision fatigue – individuals cannot be expected to deliberate on every single
choice that face throughout the day.206 This may be true, but since we are interested
in how a nudge affects a chooser’s autonomy relative to its pre-nudge state, decision
fatigue is not an overarching concern.
Sunstein also notes that active choosing may cause harm if the area of choice
is highly complex or technical.207 The concern here is that a chooser may be prone to
make a “poorer” choice than an expert, and so a default system that implicitly
signals the expert’s recommendation may be preferred. Again, that may be so, but
the effectiveness of a nudge vis-à-vis some objective “harm” is not our key
evaluative tool – autonomy is. If we are to respect the autonomy of choosers, then
poor decisions may be an unfortunate consequence. And on process autonomy
grounds, active choosing is preferable to passive choosing (e.g. defaults) because of
the element of deliberation.
It’s important to note here that our focus on deliberation is outcome-
neutral.208 A process autonomy account is not concerned with whether choosers
206 Ibid at 146. 207 Ibid at 147. 208 Ibid at 93.
220
select the donating, not donating, or not choosing options. As we saw in Chapter 5,
this is one of the key benefits of process autonomy – that it accounts for, and
permits, a wide variety of authentic motivations, including undecided motivations.
Our process autonomy focus is only concerned with whether a chooser is able to
access and apply those authentic motivations. Encouraging deliberation, regardless
of the outcome of the deliberation, is a key part of respecting a chooser’s
participatory process autonomy.
6.7.1.2. System 1 & System 2
Active choosing encourages deliberation on two fronts – first, by ensuring
choosers are aware of the choice, and second, by parlaying that awareness into an
engagement with a chooser’s reflective, deliberative self – their System 2. Nudges
will generally appeal to one of two types of human cognition – System 1 or System 2
– reflecting the dual-process account of human psychology developed by Stanovich
and West,209 and subsequently popularized by Daniel Kahneman in Thinking, Fast
and Slow.210
Nobel winner Daniel Kahneman’s groundbreaking work shows that
individual cognitive processes tend to be instinctually fast and frugal (what he calls
209 Keith E Stanovich & Richard F West, “Individual differences in reasoning: implications for the rationality debate?” (2000) 23 Behav Brain Sci 645. 210 Kahneman, supra note 20.
221
“System 1” thinking) and to a less frequent extent, slow and deliberative (“System
2”).211 System 1 is instinctual, innate, and “operates automatically and quickly, with
little or no effort and no sense of voluntary control.”212 Sunstein argues that the
defining feature of System 1 is that it works automatically.213 The human brain
delegates many decisions to System 1 since System 1 is faster and less cognitively
burdensome. In other words, it is more efficient to do so. Much of the work of
behavioural insights is dedicated to discovering biases and heuristics (rules of
thumb) which tend to be manifestations of System 1 thinking.214
In contrast, System 2 reflects the conscious reasoning part of the brain and
requires greater effort. System 2, “allocates attention to the effortful mental activities
that demand it, including complex computations.”215 To further the distinction,
Sunstein and Thaler write that System 1 is: “uncontrolled, effortless, associative, fast,
unconscious, skilled,” while System 2 is: “controlled, effortful, deductive, slow, self-
aware, rule-following.”216 Another way of characterizing the difference is that
System 1 operates on a passive, reactive basis while System 2 operates via active
thinking.217
211 Ibid. 212 Ibid at 20. 213 Sunstein, "Why Nudge", supra note 26 at 31. 214 Kahneman, supra note 20. 215 Ibid at 21. 216 Thaler & Sunstein, "Nudge", supra note 1 at 20. 217 Rebonato, supra note 112 at 385.
222
More importantly for our present analysis however, is the role of deliberation
in the two systems. System 1 is not very deliberative (if at all). Automatically or
instinctually reacting to stimuli is not a marker of deliberation in any sense of the
term. System 2, on the other hand, is quintessentially about deliberation – slow,
reflective thinking about one’s choices and one’s motivations.
In most cases, we can thus view System 2 decision-making as more strongly
associated with autonomy than System 1. Recall that our understanding of
autonomy is one of participatory self-government in accordance with a chooser’s
authentic motivations. Participation and higher-order motivations are the two
elements of that conception most applicable to our cognitive deliberation analysis. A
chooser who deliberates on their choices is engaging with their motivations in a
participatory fashion. And deliberation is more likely to lead to a chooser deciding
in line with their authentic motivations because it allows choosers to consider and
evaluate the opportunity set with that goal in mind.
On the other hand, choosers who are simply reacting to choice stimuli
(System 1) are not participating in their own self-governance in any meaningful
fashion. System 1 reacting is primarily driven by a chooser’s automatic lower-order
desires, not their higher-order motivations. Because a System 1-based decision is not
deliberative, a chooser has little opportunity to access and apply their authentic
motivations, as they would if System 2 were in charge. This fact is borne out in
223
surveys – Sunstein reports that individuals are more supportive of System 2 nudges
since they believe their decisions are more authentically their own when reached
through deliberation.218
One of the side benefits of nudges that strengthen a chooser’s System 2 is that
they tend to operate overtly or transparently.219 This can be contrasted with nudges
that encourage a chooser to “decide” using their System 1, effectively
disempowering a chooser.220 System 1 decision-making tends to operate on a
subconscious level – the chooser is not fully aware that they are making a decision
or why they are making a particular decision. They are prompted to act as opposed to
deciding to act.
While we can say that a decision made by a chooser’s System 2 is more
autonomous than a decision made by a chooser’s System 1, we are not concerned on
an absolute basis with those facts. Instead, we are interested in whether a nudge
causes the chooser to deliberate more (i.e. more System 2 and/or less System 1) or less
(i.e. more System 1 and/or less System 2) than in their pre-nudge state. If a chooser
pre-nudge is choosing with little deliberation because they are ruled by their System
1 selves, then a nudge that also appeals to their System 1 won’t necessarily be found
218 Cass R Sunstein, "Do People Like Nudges?" (2016) [unpublished, available online: <http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2604084>]. 219 Binder & Lades, supra note 122 at 12. 220 Jones, Pykett & Whitehead, supra note 3 at 52.
224
to be causing a reduction in deliberation (i.e. a reduction in decisional autonomy).
Nudges will be found to preserve or enhance the decisional autonomy of choosers if
they appeal to a chooser’s System 2 more than on a pre-nudge basis (by either
strengthening System 2 or leaving it unchanged). On the other hand, nudges that
appeal to a chooser’s System 1 more than on a pre-nudge basis will be found to
reduce the decisional autonomy of choosers.221 It’s not the end point that matters, but
the effect – reducing, preserving or increasing deliberation, and therefore decisional
autonomy.
In light of the distinction between System 1 and System 2 and their respective
implications for a chooser’s autonomy, Binder and Lades propose a variant of
libertarian paternalism they call “autonomy-enhancing paternalism.”222 Their theory
favours interventions that, “improve, rather than, reduce, individuals’ ability to
make critically reflected, unbiased, autonomous decisions.”223 Their goal is similar to
the one of this thesis, though the execution is radically different. The stated purpose
of their work is: to “help individuals to become better decision-makers and thus
make better informed, less biased, and more autonomous choices over time that
may better reflect their true preferences.”224
221 See e.g. Bovens, supra note 6. 222 Binder & Lades, supra note 122. 223 Ibid at 4. 224 Ibid at 6.
225
Binder and Lades’ understanding of autonomy is one firmly and solely rooted
in a chooser’s System 2. Our theory of autonomy reflects many of the same ideas but
goes further – autonomy is found not just in deliberation, but in unbiased and
informed deliberation because doing so promotes a chooser’s participation in governing
their lives in accordance with their authentic motivations.
We have already seen one example of a nudge that works through a
chooser’s System 2 (albeit without affecting the strengths of the two systems) –
active choosing. Active choosing is a preferred system on autonomy grounds not
just because it creates awareness of choice where none is present, but also because it
is increases the likelihood that a chooser’s System 2 will be in charge of making the
relevant decision. Requiring some (even minimal) level of deliberation means that a
chooser must reflect on their choices, likely (but not assuredly) calling into action
their System 2. Since a chooser cannot reflect on their choices using their System 1
instincts, requiring reflection ensures that System 2 is at least asked “its” opinion on
the matter, even if the opinion is simply to let System 1 take over.
The active choosing example also illustrates an important caveat in nudges
that appeal to a chooser’s System 2 thinking – requiring deliberation means
respecting a chooser as a System 2 “thinker,” but this could result in a chooser’s
System 2 deciding to leave System 1 in charge. However, for that decision to be
autonomous, some deliberation is necessary. Thus, if a nudge interfered with a
226
chooser’s reflections of their lower-order desires,225 such a nudge fails to preserve
the chooser’s decisional autonomy.
Some nudges also strengthen a chooser’s more deliberative System 2 by
weakening the power of System 1, in part by reducing the power of temptation.
Handelman calls these “intellectual manipulations.”226 Responding to temptation is
generally an automatic, instinctual, System 1 response which reflects little autonomy
on the part of the chooser.227 Cooling-off periods, for example, allow individuals to
re-think their decisions when they are not subject to the whims of System 1
responses.228 Cooling off creates detachment from a choice scenario, necessarily
calling for deliberation by a chooser on a choice that they are facing (e.g. waiting
periods for gun purchases) or that they already made (e.g. door-to-door hot water
heater sales).229 Such a nudge could thus be said to preserve the decisional autonomy
of choosers.
Likewise, penalty defaults stimulate deliberation by ensuring that
individuals’ deliberative selves are called to action. And, as we will see in
225 Dworkin, "Behavior Control", supra note 96 at 26. 226 Handelman, supra note 175 at 46. 227 See e.g. Binder & Lades, supra note 122 at 5. 228 Colin Camerer et al, “Regulation for Conservatives: Behavioral Economics and the Case for ‘Asymmetric Paternalism’” (2003) 151:3 Univ PA Law Rev 1211 at 1238. 229 See e.g. Daniel B Klein, “Reply to Sunstein” (2004) 1:2 Econ J Watch 274; Jonathan Klick & Gregory Mitchell, “Government Regulation of Irrationality: Moral and Cognitive Hazards” (2006) 90 Minn Law Rev 1620; Stephen J Hoch & George F Loewenstein, “Time-Inconsistent Preferences and Consumer Self-Control” (1991) 17:4 J Consum Res 492.
227
subsequent sections, informing choosers that they suffer from behavioural biases, or
that they are being nudged also respects the decisional autonomy of choosers
because it allows choosers the opportunity to deliberate on that information and
factor it into their decision-making.
On the autonomy-reducing side of the ledger we find nudges that strengthen
System 1 at the expense of System 2, or those that weaken System 2 to the benefit of
System 1. A chooser may be denied the opportunity to appeal to their authentic
motivations through the design of the choice scenario, leaving their lower-order
desires to govern their behaviour. Again we go to Dworkin to see that these types of
influences constitute the class of methods that, “are destructive of the ability of
individuals to reflect.”230 Handelman calls these types of nudges “emotional
manipulation” that seek to have the target act impulsively.”231
One example here would be nudges that operate through subconscious cues,
such as subliminal advertising. Whatever choice arises from those cues is not one
generated by a chooser’s deliberative, System 2 self, but one generated by a
chooser’s System 1. Another example of a nudge that works through a chooser’s
System 1 non-deliberative reacting is a shift from a simple choice presentation to one
that is more complex. Behavioural insights show that if the context of choice is too
230 Dworkin, "Behavior Control", supra note 96 at 27. 231 Handelman, supra note 175 at 46.
228
complex, System 1 is activated and choosers tend to resort to heuristic-based
reacting and not System 2 deliberation.232
This also works the other way – simplifying a complex choice has the
opposite effect – the power of System 1 is lessened and individuals are more likely
to deliberate on their choices. For example, in the United States, shifting from the
traditional food pyramid nutritional guide to a simple, graphic-based MyPlate is one
such simplification nudge.233 So is the re-design of fuel economy labels that make it
easier for choosers to understand relevant information,234 thus increasing the
likelihood that they will deliberate and reflect on what product options match their
authentic motivations.
In this section we saw the variety of ways that nudges can affect choosers’
decisional autonomy by working on their level and quality of deliberation vis-à-vis
choice awareness and mode of cognition. Some nudges, such as active choosing,
promote deliberation by mandating awareness – requiring choosers to turn their
minds to a choice. Others promote deliberation by wresting control of a decision
from a chooser’s reactive and passive System 1 and shifting it to their more
reflective, active thinking System 2. Those nudges that preserve or enhance the level
232 See e.g. Cass R Sunstein, Simpler: The Future of Government (New York: Simon & Schuster, 2013). 233 Ibid at 133. 234 Sunstein, "Why Nudge", supra note 26 at 67.
229
of deliberation undertaken by choosers can be said to at least preserve the decisional
autonomy of choosers.
Deliberation is but one element of decisional autonomy. However, all three
elements – deliberation, information and biases – are interconnected. Enhanced
deliberation lessens (but does not eliminate) the likelihood that biases will factor
into a chooser’s decision-making process as biases tend to appear most frequently in
System 1 choices. Information is also an important element of deliberation – one
cannot truly deliberate without relevant information to deliberate on.
6.7.2. Biases
Aside from deliberativeness, one of the other key differences between System
1 and System 2 modes of cognition are the presence of behavioural biases or
heuristics. Choosers who decide through System 1 automatic reacting are more
likely to have their decisions tainted by biases or heuristics, while System 2
decisions are less likely to be so affected. This difference stems from deliberativeness
– if a chooser does not have the cognitive time or energy to think about their options
and how they fit into their overall life plan (i.e. match their authentic motivations),
then they are more likely to rely on rules of thumb to make decisions. Indeed, many
of the biases that humans exhibit are rooted in our fast, instinctual systems.235
235 Thaler & Sunstein, "Nudge", supra note 1; Kahneman, supra note 20.
230
A chooser’s deliberation (selection of an option from an opportunity set)
should be as optimal as possible in order for their authentic motivations to govern
their option evaluation process. Behavioural biases, and frequently, the use of
heuristics in decision-making often hinder that goal. A chooser may be able to
deliberate between options, but if the deliberation process is based on a rule of
thumb as opposed to actual information (e.g. costs and benefits of particular
options) and the accurate weighing of that information, a realization of a person’s
authentic motivations through to option selection is doubtful.
When we talk of biases, we are typically referring to the use of heuristics
(rules of thumb) or biases in reasoning. Consider the framing effect, where humans
view and weigh the same statistics differently depending on how they are
presented. We have previously seen that a chooser is more likely to consent to a
surgical operation if the odds of survival are presented as “90% chance of survival,”
versus “10% chance of death,” even though the two are mathematically identical.236
Likewise, our paradigmatic cafeteria example also indicates a bias that choosers
have towards food placed earlier in the line. Finally, biases can also mean the ruling
presence of emotions, such that System 1 takes over as opposed to the more logical
System 2 – we see this with graphic warning labels on cigarette packages.
236 See e.g. Armstrong et al, supra note 43.
231
That does not mean that System 1 thinking is necessarily “bad” because it is
more biased. Many of those biases exist for evolutionary reasons,237 and System 1
reacting helps us do many important tasks that we would not be able to do if we
constantly deliberated. For example, we can catch a fly ball not by thinking about
trajectory, but by experience.238 In fact, if we thought about it too much, we might
miss. The same goes for driving a car, which many experienced drivers are able to
safely do based on experience without active deliberation.
However, biases are problematic for our understanding of autonomy because
we are interested in choosers exhibiting authentic and effective self-government.
Biases at the decisional autonomy level decrease the likelihood that a chooser will be
able to apply their relevant authentic motivation to the choice problem for two
reasons. First, biases tend to work through a chooser’s System 1, which we know to
be less deliberative, and deliberation is linked to decisional autonomy. However
useful System 1 reacting may be for effective decision-making, when it comes to
autonomy, unbiased thinking must generally be preferred, which means that, in
many cases (but not all), System 2 is to be the preferred mode of cognition, even if
237 See e.g. Douglas T Kenrick et al, “Deep Rationality: The Evolutionary Economics of Decision Making” (2009) 27:5 Soc Cogn 764; Peter Hammerstein, “Evolutionary Adaptation and the Economic Concept of Bounded Rationality -- A Dialogue” in Gerd Gigerenzer & Reinhard Selten, eds, Bounded Rationality (Cambridge: The MIT Press, 2001); Henry Brighton & Gerd Gigerenzer, “Are rational actor models ‘rational’ outside small worlds?” in S Okasha & K Binmore, eds, Evolution and Rationality: Decisions, Cooperation, and Strategic Behaviour (Cambridge: Cambridge University Press, 2012) at 84. 238 Arie W Kruglanski & Gerd Gigerenzer, “Intuitive and Deliberate Judgments Are Based on Common Principles” (2011) 118:1 Psychol Rev 97.
232
only to confirm that System 1 should be left in charge in a particular choice
scenario.239 Second, biases themselves affect how a chooser sees a choice problem, so
that even if a chooser is engaging in some level of deliberation, a bias may prevent
that chooser from accessing and applying their most relevant authentic motivation.
It’s important to distinguish between the two effects on decisional autonomy
– deliberativeness and accurate evaluation of choice options – because a chooser can
be deciding deliberatively (e.g. System 2 is in charge), but also still be biased. For
autonomy purposes, we want a chooser to reason (i.e. deliberate), but we also want
them to reason effectively. Consider a chooser who exhibits recency bias whereby
they overestimate the statistical probabilities of an occurrence by overweighting
recent events they can remember.240 For example, there was just a terrible plane
crash, and the chooser adjusts their travel plans as a result even though statistically
flying is still much safer than driving. Our chooser has just used faulty statistical
239 Biases are an indicator for the autonomy level of decision-making, but not determinative. As noted earlier, the use of some biases or heuristics in decision-making is desirable and may in fact be an expression of one’s autonomy – a person’s System 2 autonomously wants to leave their System 1 in charge. For example, if you are going to catch a fly ball and a bystander yells instructions at you – either “trust your gut” or “calculate the trajectory” – that nudge can engage in your gaze heuristic under System 1 or your deliberative System 2, respectively. If you choose to let your let your biases work for you, you are still acting autonomously because you have chosen to do so using your System 2. In other words, your System 2 delegated decision-making to a biased System 1, and did so in an autonomous fashion. However, most nudges do not work like this – a chooser is not provided with the opportunity to autonomously choose to be biased/governed by their System 1. 240 See e.g. Gerd Gigerenzer & Henry Brighton, “Homo Heuristicus: Why Biased Minds Make Better Inferences” (2009) 1 Top Cogn Sci 107.
233
logic to select from a menu of options even though their authentic motivations (e.g.
to travel safely) are unchanged.
Consider again the arrangement of food in a cafeteria. If a chooser, regardless
of their authentic motivation, succumbed to their present bias for whatever food
was immediately in front of them (i.e. first in line),241 the chooser would not be
choosing autonomously because (a) they are not evaluating their options
deliberatively, and (b) they are not accurately evaluating their options – they are
overweighting the value of the food in front of them compared to the food at the
end of the line.
Consider another example: a chooser’s optimism bias that causes them to
underestimate their risk of negative health consequences from smoking can be
present even if a chooser is deliberatively thinking about their choices and trying to
calculate a cost-benefit tradeoff. As we saw in Part I, many nudges operate in this
manner – by changing how choosers’ perceive the costs and benefits of certain
choices without actually changing those costs and benefits. And changes in
perception are often accomplished by working with or through individual biases.
This dichotomy of how biases affect autonomy tracks the different types of
biases that exist. Some biases (e.g. emotional decision-making) speak to the mode of
241 See e.g. Andrew S Hanks, David R Just & Brian Wansink, “Smarter Lunchrooms Can Address New School Lunchroom Guidelines and Childhood Obesity” (2013) 162:4 J Pediatr 867.
234
cognition that drives a choice (System 1 vs. System 2), while other biases (e.g.
framing effect, optimism bias) go to how choosers evaluate their options. We will for
the most part be concerned with the latter since we have already addressed the
former – System 2 deliberation is preferred as opposed to System 1 instinctual or
emotional decision-making because a chooser is more likely to be effectively
governed by their authentic motivations.
We now know how biases affect a chooser’s autonomy, so we need to turn our
attention to examining how nudges can affect a chooser’s biases. Nudges can either:
make a chooser more biased (counter-biasing), which would count as a reduction in
decisional autonomy; make a chooser less biased (de-biasing), which would count as
an increase in decisional autonomy; or they can not directly affect the biases of
choosers (re-biasing), thereby simply preserving the chooser’s decisional autonomy
as it existed pre-nudge. Counter-biasing and re-biasing nudges are bias-utilizing in
that they leave a chooser’s bias in place (i.e. as it is pre-nudge) and either redirect it
(re-biasing) or attempt to counter it with the expression of a new bias (counter-
biasing). No attempt is made with these nudges to improve the process autonomy of
choosers by removing biases that we now know to be generally incompatible with
decisional autonomy. On the other hand, de-biasing nudges attempt to accomplish
such a goal.
235
6.7.2.1. Re-Biasing & Counter-Biasing
Re-biasing nudges are those that attempt to use a chooser’s existing biases
(that are leading them to arguably select outcome autonomy deficient choices) to
steer the chooser towards selecting different options. In other words, the same bias
is used again, but simply redirected toward a different outcome. The “re” in re-
biasing does double duty here – both to indicate that the same bias is used again
(repetition), and to signal that it is redirected to the selection of a different option.
Our paradigmatic case of the ordering of food in a cafeteria is an excellent
example of a nudge that operates through re-biasing. Since choosers are
behaviourally biased to choose more of the first foods presented (as a result of
salience and present bias), a nudge that only switches the types of foods found at the
beginning of the line uses the same existing bias for a different outcome (healthier
food choices). The fundamental part of re-biasing nudges is that no new or
additional biases are utilized by the nudge.
Re-biasing nudges seem to have no negative effect on a chooser’s decisional
autonomy since the chooser is not more or less biased than they were in a pre-nudge
state. The result of the same bias is simply redirected towards a different choice
outcome. Since their level of biased decision-making is held constant by the nudge,
their decisional autonomy is also held constant. While it may not be ideal that a
236
chooser is selecting in a decisional autonomy-deficient state because of the presence
of a bias, since we are evaluating the ethical effects of a particular nudge (i.e.
compared to a pre-nudge state), we must disregard that unfortunate reality.
On the other hand, counter-biasing nudges utilize one or more additional or
new biases that are not already being exhibited by a chooser in order to
counteract/neutralize and potentially overwhelm the effects of other biases that are
exhibited pre-nudge. For example, due to optimism bias, individual smokers vastly
underestimate their personal health risks from smoking. Graphic warning labels
attempt to correct for or neutralize the result of this miscalculation by engaging
individuals’ affective bias and recency bias. A similar use is found in risk
narratives.242 These newly introduced biases cause individuals to overestimate
statistical probabilities, but since they are underestimating them to begin with, the
effect is meant to be a wash. As Binder and Lades write: “These risk narratives aim
to counteract individuals’ optimism bias (which causes too much risk taking) by
harnessing individuals’ availability bias.”243
It’s important to remember here that our behavioural biases are ingrained
and exist independent of any particular choice scenario, though they may be
brought out or expressed in certain scenarios. So it is not as if a counter-biasing
242 See e.g. Jolls & Sunstein, supra note 52. 243 Binder & Lades, supra note 122 at 14.
237
nudge causes a new bias, so much as it simply causes the expression of one or more
particular biases. At the same time as causing the expression of a new bias, a
counter-biasing nudge may also decrease the expression of another bias. Counter-
biasing nudges do not improve the state of biases, they simply lessen the effect of
pre-nudge biases, by combining them with the effects of the newly introduced
biases.
At first blush, it would seem that accounting for counter-biasing nudges
would contradict our stated purpose of evaluating only the effects of nudges and
not considering the level of autonomy that choosers had pre-nudge. If a chooser is
misevaluating their options pre-nudge, but that is corrected for using a new bias
post-nudge, then arguably their decisional autonomy has not been reduced.
Reaching that conclusion inevitably requires a consideration of the chooser’s pre-
nudge bias expression (and therefore decisional autonomy as well), which would be
contrary to our evaluative standard of taking choosers as they are immediately pre-
nudge.
Note that it is not necessarily the quantity of expressed biases that matter, so
much as it is their effect. A counter-biasing nudge introduces a new nudge into a
chooser’s option evaluation stage of decision-making, but that does not mean that a
chooser is automatically to be considered less decisionally autonomous. Nor would a
nudge that removed the expression of one of two offsetting biases necessarily cause
238
a choose to become more decisionally autonomous. Instead, it is the effect of the
biases that matter for our process autonomy analysis – are choosers better able to
evaluate their options in light of their authentic motivations?
Counter-biasing nudges must be considered unethical vis-à-vis decisional
autonomy because they invariably have symmetrical effects on asymmetrically
biased choosers. Nudges are policies of general application – they are not tailored to
individual choosers. In our evaluation we are not to consider the state of a chooser’s
autonomy pre-nudge because accounting for the varying states and strengths of
biases for every chooser is an impossible task. We are not all equally biased – some
choosers will not be biased pre-nudge, others will be strongly biased, and others
will be less biased.244 Thus, if a nudge caused all choosers to become more biased in
one particular area, but they didn’t all start from the same point, then the nudge
could possibly cause a reduction in process autonomy for those who weren’t
exhibiting the bias pre-nudge.
In particular, a counter-biasing nudge would reduce the decisional autonomy
of those choosers who are unbiased pre-nudge. Even if that group of affected
choosers is a small minority, the nudge would violate our principle of process
autonomy preservation, rendering it unethical. Since process autonomy is the heart
244 Gregory Mitchell, “Why Law and Economics’ Perfect Rationality Should Not Be Traded for Behavioral Law and Economics' Equal Incompetence” (2002) 91:1 Georgetown Law J 67.
239
of our conception of participatory self-government, we cannot aggregate process
autonomy, which adopting a counter-biasing calculus would require (more on this,
and how we may be able to aggregate outcome autonomy in Chapter 8).
Let’s consider again the example of graphic warning labels on cigarette
packages. While many choosers underestimate the health risks of smoking because
of optimism bias, some do not (i.e. they accurately estimate the risks). As we’ve
already noted, graphic warning labels work in part by causing choosers to
overestimate their personal risks of smoking. So for those choosers who
underestimated their risks by an equivalent amount pre-nudge, post-nudge they
may now accurately assess their risks. For them, decisional autonomy has arguably
been increased since they are now able to accurately assess their options. However,
for those choosers who pre-nudge accurately assessed their risks, the graphic labels
nudge may cause them to overestimate the risks post-nudge. This causes a reduction
in those choosers’ decisional autonomy because they are now biased when they
were not pre-nudge. So, even though the same nudge may cause an improvement in
the decisional (process) autonomy of those choosers who pre-nudge underestimated
their risks, we cannot trade off their decisional autonomy gains for decisional
240
autonomy losses in the minority.245 Doing so would be contrary to the very nature of
process autonomy.
The disqualification of counter-biasing nudges from the class of ethical
autonomy-preserving nudges also ensures that nudges cannot be initially labeled
ethical and then become unethical due to factors external to the nudge design. In
particular, this ensures that nudges that persist in a particular choice environment
will not have perverse effects if, through other means, the pre-existing bias that the
nudge was designed to neutralize changes or disappears.246 For example, perhaps
through changing social norms or educational efforts, smokers’ optimism bias
regarding the risks of smoking is lessened. If the graphic warning label nudge
persisted in light of those changes, smokers would then mostly overestimate their
risks because the need for specific neutralization has been reduced. Such a result
would clearly be problematic from an autonomy perspective, further buttressing the
view that counter-biasing nudges must be found, as a class, to be process autonomy
reducing.
There is also some evidence that counter-biasing can have unintended effects.
For example, one study showed that attempts to manipulate choosers in the
245 For a consideration of other examples of using recency bias to increase individual perceptions of risk and their potential negative effects see e.g. Nick Gill & Matthew Gill, “The limits to libertarian paternalism: two new critiques and seven best-practice imperatives” (2012) 30 Environ Plan C Gov Policy 924 at 931. 246 See e.g. Binder & Lades, supra note 122 at 23.
241
opposite direction of a particular bias (optimism bias) actually exacerbated the
bias.247 Furthermore, if only some biases are known (a real possibility since the
science is constantly evolving), offsetting biases may have indeterminate effects.248
6.7.2.2. De-Biasing
In addition to counter-biasing and re-biasing, nudges can also affect the
decisional autonomy of choosers by de-biasing them. De-biasing can occur in two
ways. First, biases can be overcome by shifting the locus of decision-making from a
chooser’s System 1 to their System 2, as we saw earlier in this chapter. This can be
accomplished, for example, by prompting choice in advance so as to encourage
deliberation, or by simplifying choices. Since many biases are expressed
predominantly as a result of System 1 “reacting,” a shift to greater System 2 control
should result in a reduction in biased behaviour. This is borne out in the evidence –
when choosers are encouraged to think analytically, the power of the framing effect
is reduced.249
247 Neil D Weinstein & William M Klein, “Resistance of Personal Risk Perceptions to Debiasing Interventions” (1995) 14:2 Heal Pschology 132. 248 Gregory Besharov, “Second-Best Considerations in Correcting Cognitive Biases” (2004) 71:1 South Econ J 12. 249 See e.g. Ayanna K Thomas & Peter R Millar, “Reducing the Framing Effect in Older and Younger Adults by Encouraging Analytic Processing” (2012) 67:2 Journals Gerontol Ser B Psychol Sci Soc Sci 139; Dilip Soman & Maggie Wenjing Liu, “Debiasing or rebiasing? Moderating the illusion of delayed incentives” (2011) 32 J Econ Psychol 307.
242
The second mode of de-biasing is more direct – a nudge achieves its goal of
altering choice outcomes by directly de-biasing choosers. De-biasing in this sense is
the process of making people aware of their behavioural biases so that the biases are
consciously and temporarily or permanently removed (overcome) from a person’s
decision-making process. Since biased behaviour is generally anathema to decisional
autonomy, the reduction or removal of biases will have the effect of enhancing the
decisional autonomy of choosers.
This mode of de-biasing also achieves its purpose by activating a chooser’s
System 2, but it is done so specifically to encourage a chooser to reflect on their
biases, whereas the first method of de-biasing activated System 2 to reflect on their
options.
De-biasing of this nature begins with informing choosers of their biases
which are causing them to choose in a certain fashion. As Dworkin writes, “[if] the
agent does not know the real reasons for his actions, he cannot reflect on such
reasons and make a favorable or adverse judgment concerning them.”250 De-biasing
in this fashion also has “domain general” effects – it applies across choice scenarios
and has lasting effects.251 In other words, de-biasing leads to better decision-making
and not just for a particular choice problem
250 Dworkin, "Behavior Control", supra note 96 at 26. 251 Carey K Morewedge et al, “Debiasing Decisions: Improved Decision Making With a Single Training Intervention” (2015) 2:1 Policy Insights from Behav Brain Sci 129.
243
Druckman has argued that presenting choosers with certain types of
information can have a de-biasing effect.252 For example, in the context of the
framing effect, providing choosers with credible, expert advice regarding the
recommended choice appears to overcome the logical inconsistency associated with
framing.253 Druckman concludes that people follow the advice regardless of the
frame.254 However, what Druckman suggests may not actually be an instance of de-
biasing, but counter-biasing. The question is whether the advice is information in
the sense that it activates a chooser’s deliberative System 2 or whether the advice
represents some sort of new bias, driving the chooser to select in line with a
particular norm.
True de-biasing attempts to eliminate the influence of a bias on a particular
choice by informing choosers of their biases so that they may factor that fact into
their decision-making. Few nudges actually do this. This means that nearly all
nudges are re-biasing or counter-biasing. They skip any attempt to correct cognitive
deficiencies and instead utilize those deficiencies to shape choices away from
welfare-harming towards welfare-promoting goals (as the choice architect sees it).
252 James N Druckman, “Using Credible Advice to Overcome Framing Effects” (2001) 17:1 J Law, Econ Organ 62. 253 The “credibility” of some of the advice provided in this study is highly questionable. For instance, indicating which choice option reflects which American political party’s position should not be considered “credible expert advice.” 254 Ibid at 76.
244
Jolls and Sunstein are particularly guilty of labeling the latter kind of intervention –
counter-biasing – as de-biasing, when no such actual de-biasing takes place.255
The focus of nudging on re-biasing and counter-biasing is understandable
when we remember that the goal of nudging is to change choice outcomes, and not
to respect the autonomy of choosers, which is, at best, a secondary concern. De-
biasing attempts are simply not particularly effective, possibly because they are
predicated on the assumption that informing choosers of their biases will eliminate
them. However, that assumption means that biases first function because they are
covert to the chooser. That is not necessarily borne out in the evidence, which
suggests that even if choosers are aware of their biases, overcoming those biases to
act with greater decisional autonomy does not flow automatically from awareness.
As mentioned earlier, Loewenstein has noted that informing people they are being
defaulted doesn’t really change behaviour.256
Instead, more successful de-biasing nudges will be those that attempt to
combine the two modes of de-biasing – activating/strengthening System 2 and
informing choosers of their biases. As noted, activating System 2 should have the
effect of encouraging a chooser to choose on a less-biased basis. Informing a chooser
of their biases will then allow them to deliberatively plan their lives to avoid
255 Jolls & Sunstein, supra note 52. 256 Loewenstein et al, supra note 192.
245
impulses or adopt commitment devices necessary to overcome sticky biases (if
informing choosers of their biases is itself not enough to eliminate the effects of
those biases).
In conclusion, we have seen the variety of ways that nudges can affect the
decisional autonomy of choosers by working on their behavioural biases. A nudge
that encourages or strengthens the power of System 2 at the expense of System 1
promotes the decisional autonomy of choosers. Nudges that qualify as re-biasing
utilize existing (pre-nudge) biases to redirect choosers to different choice selections.
In doing so, they preserve the decisional autonomy of choosers. On the other hand,
counter-biasing nudges will generally fail to preserve the decisional autonomy of
choosers because of the asymmetry of choosers’ pre-nudge biases, causing some
choosers to suffer a reduction in autonomy that cannot be offset by decisional
autonomy gains due to the very nature of process autonomy. Finally, the small
subset of nudges that de-bias choosers will enhance their process autonomy and are
ethically preferred.
6.7.3. Information
We have already seen that nudges can affect choosers’ decisional autonomy
by affecting how they decide – System 1 and System 2 – and how they evaluate their
options – biased or unbiased. The third way that nudges can affect the decisional
246
autonomy of choosers is by altering what information choosers use to evaluate their
options. Many nudges function by providing information to choosers or, to a lesser
extent, by hiding information from choosers.
Information is vital for a chooser to act with decisional autonomy. Recall that
we are concerned with choosers choosing from an opportunity set that option which
best matches their motivational autonomy. Doing so is an impossible task without
information about those options. Even if choosers are provided with an opportunity
to fully deliberate on their options, we cannot expect that they will be able to put
their authentic motivations into action without some understanding of what options
they face and the characteristics of those options so that they can be appropriately
matched with authentic motivations. Imagine telling a chooser: “you have all the
time in the world to think about this choice, and I really want you to think about it,
so please choose whatever option you think best matches your life plan between Box
A and Box B, but I’m not going to tell you anything about what’s in the boxes.” To
exhibit decisional autonomy then, choosers must be provided with information
about the opportunity set and the characteristics of the options within the set.
Information is particularly important because oftentimes our authentic
motivations tend to be more abstract and it is at the option evaluation stage that we
need to “discover” what options within an opportunity set best reflect those abstract
motivations. For example, if a chooser has an authentic motivation to live a healthy
247
life, then it stands to reason that when they are choosing from a menu of food
options they need to know calorie counts in order to choose the healthiest option (or
at least to balance health and taste considerations). In fact, without sufficient
relevant information, a chooser is more likely to select from an opportunity set using
heuristics (rules of thumb). A health-minded chooser might think, “a salad, that’s
probably the healthiest option,” when sometimes salads are deceptively unhealthy.
Thus, a chooser must have sufficient information in order to determine which
choices best reflect their higher-order motivations.
Information is also important in order to empower choosers to self-nudge if
they so desire. A chooser may want to change their behaviour, but without relevant
information they may be unable to do so. As Jones, Pykett and Whitehead write: “At
their best, behaviour-changing policies seek to empower the individual by
providing easily accessible and salient information, which can be processed in such
a way that enhanced forms of decision making can be made.”257
Nudges can affect the information a chooser utilizes at the option evaluation
stage of decision-making either by providing less relevant information or more
relevant information. Relevance is an important qualifier. Simply providing a
greater quantity of information will not help a chooser match choice options to their
authentic motivations. Providing more relevant information can be accomplished
257 Jones, Pykett & Whitehead, supra note 3 at 52.
248
either by decreasing the quantity of information but increasing the overall relevance
of information, or by providing an additional quantity of relevant information.
Generally, providing relevant information will not cause a loss to a chooser’s process
autonomy. The opposite would be true for nudges that negatively affect information
– the quantity of relevant information could be reduced (e.g. shrouded) or the
relevance of existing information altered in a reductive manner – both have the
same negative effect on decisional autonomy.
One key qualifier is necessary – relevance is a matter of subjectivity.
Information will enhance a chooser’s decisional autonomy if it is information that
chooser would factor into their option evaluation process. Information need not be
objectively or rationally relevant. If the information is relevant to the chooser, then it
is relevant to determining the effects of the nudge on that chooser’s decisional
autonomy. For example, if a chooser would factor into their decision-making
information about their neighbours’ electricity consumption, even though doing so
may be objectively irrational, that information is relevant to the chooser. A nudge
that provides that information to the chooser would be positively affecting the
provision of relevant information and thus decisional autonomy.
Unfortunately, because the relevance of information is subjective, it will be
difficult to definitively determine the effects of informational nudges on decisional
autonomy. We cannot know whether information provided or removed by a nudge
249
is relevant until we have seen whether it is factored into a chooser’s post-nudge
decision-making, which obviously happens after the nudge. This results in
circularity: the information necessary for a chooser to exercise decisional autonomy
(i.e. relevant information), is whatever information the user would actually use in
exercising their decisional autonomy.
Notwithstanding this apparent indeterminate circularity, we can attempt to
make some generalizations about nudges that affect information relevance. In
particular, we can determine what information could possibly be relevant such that
any reduction in the provision of that information might count as failing to preserve
the decisional autonomy of choosers. Nudges that increase the provision of possibly
relevant information will also be likely to preserve the decisional autonomy of
choosers.
Once again, we must note that we are interested in the effects of nudges on
decisional autonomy vis-à-vis relevant information. This means we are not
interested in the starting point of a chooser’s information. If a choice architect has
information that may be relevant to a chooser but does not provide that information
in their nudge design, then, so long as the information was not being provided on a
pre-nudge basis, then the withholding of that information in the nudge does not
render the nudge anathema to decisional autonomy (though doing so may be
unethical on other grounds).
250
There are three classes of information that nudges can convey: information
about the chooser; information about the opportunity set; and information about the
options within the set. The first class goes to information about the chooser, as we
just saw with de-biasing nudges that inform choosers that they are being nudged
and/or suffer from a particular behavioural bias. For example, individuals can be
nudged via a reminder later in the day that they are more likely to be suffering from
decision fatigue (ego depletion).258 Information about a chooser’s biases will likely be
relevant to the affected choosers, so providing that information will be, at a
minimum, preservative of decisional autonomy and likely even enhancing of
decisional autonomy.
The second and third classes of information are what we are concerned most
about in this section. These classes of information go to the content of the choices –
what the options are and their associated characteristics (i.e. costs and benefits).
Nudges can affect the extent to which choosers are aware or unaware of their
opportunity set. As we saw in Part I, altering the perception of options does not
technically eliminate a form of choice architecture from counting as a nudge, as a
chooser’s liberty of choice is objectively retained. However, negatively affecting how
258 Roy F Baumeister, “Yielding to Temptation: Self-Control Failure, Impulsive Purchasing, and Consumer Behavior” (2002) 28:4 J Consum Res 670; Martin S Hagger et al, “Ego depletion and the Strength Model of Self-Control: A Meta-Analysis” (2010) 136:4 Psychol Bull 495; Baumeister, supra note 36.
251
a chooser perceives an opportunity set by, for example, shrouding some options,
does not respect the autonomy of choosers. A nudge that effectively hides, shrouds or
reduces the salience of an option that existed pre-nudge cannot be said to preserve
the decisional autonomy of choosers. A chooser who may have selected an option
pre-nudge because they believed it best reflected their authentic motivations no
longer has that option, from their perspective. Again, relevance is an important
guide – if any chooser could view the affected (shrouded) option as relevant to their
decision-making process, then the nudge negatively affects their autonomy on
information grounds.
For example, consider an opportunity set that consists of twenty different
investments that a person can invest in within a self-directed pension plan. This
information is listed all on one page. Now imagine a nudge designed to encourage
the selection of specific funds that have been shown to yield better performance over
the long-term. So, the nudge removes five underperforming options (though still
frequently selected for non-performance-related reasons) from the main page listing
and lists them on another page with a link that says “More Here.” The options
technically remain and therefore the choice architecture counts as a nudge (after all,
it’s only one click away).
But choosers would perceive their opportunity set to be reduced – they do
not have full information about the extent of their options since the information
252
appears to cover only fifteen options when in fact there is twenty. If the nudge were
to occur in reverse – previously shrouded options were brought to the fore –
choosers would have more relevant information with which to make their choices.
So, a nudge that reduces the information a chooser has about the constituent options
that make up an opportunity set would fail to preserve the decisional autonomy of
choosers. Handelman calls these interventions “limiting manipulations,” as they are,
“intended to limit a target by maneuvering the target toward one specific option or
reducing the number of options that he considers while making a decision.”259
We see a similar situation even in active choosing scenarios, because they
may be poorly designed. In 2011, on the advice of their Behavioural Insights Team,
the UK organ transplant authority switched from an opt-in default system to what
they call a “prompted choice” scheme for organ donor consent when signing up for
a driver’s license.260 In order to complete their application, applicants must tick one
of three boxes:
(A) Yes, I would like to register;
(B) I do not wish to answer this question now; or
259 Handelman, supra note 175 at 45-46. 260 Department of Health, “Driving up organ donations”, (2011), online: GOV UK News Stories <https://www.gov.uk/government/news/driving-up-organ-donations>; Muireann Quigley, “Nudging for Health: On Public Policy and Designing Choice Architecture” (2013) 21:4 Med Law Rev 588 at 595. Characterizing this type of intervention as “prompted choice” is correct, but slightly misleading. Choosers are indeed prompted to think about the issue, but they are also required to make a choice – mandated or active choosing may be a better term.
253
(C) I am already registered on the NHS Organ Donor Register.
Note that there is one glaring omission – there is no option applicable for a
chooser who does not wish to be a donor. Prior to the adoption of the active choice
system, such a chooser could choose not to donate by doing nothing. In this
scenario, they would have to select the delayed choice option, but that doesn’t really
seem to fit – they want to answer the question with “no,” they just can’t. Instead,
their “no” option has been lumped in with the “not yes” of option B.261 We could
argue that such a switch amounts to a reduction in the opportunity set, but that’s
not quite right – pre-nudge their choice was always “not yes” because the system
was one of opt-in. The active choosing scenario preserves that option; it just does it
poorly by hiding the “no” option with the “not yes” wording.
Instead, it best to think of this problem as one of shrouded options. Such
shrouding is problematic for decisional autonomy, because even assuming a chooser
deliberates on the matter, how are they to match their motivation to the options,
when there is no good option that matches their motivations? At least prior to the
intervention, they could give effect to their motivation by doing nothing (in an
explicit consent, opt-in system).
261 In opt-in regimes, organ donation agencies generally do not track those who do not wish to donate. Instead they have two sets of data – those who consent to donate (yes) and those who have not yet consented to donate (not yes). No distinction is made in the latter group between those that have expressed a wish not to donate and those that simply have not decided. In this example, option (B) is recorded as “not yes.”
254
The third, and likely largest, class of relevant information that nudges can
affect are the characteristics of options within the opportunity set. Nudges of this
nature would leave the perception of the constituency of the opportunity set
unaltered, but change the information that choosers receive about the cost and
benefits of those same options. We are really interested here in perceived costs and
benefits (a prime area for nudging identified in Part I). If the information about the
costs and benefits is information that a chooser would believe is relevant (i.e.
factored in their decision-making), then a nudge that reduced the provision of such
information would not be found to be decisional autonomy preserving, and vice
versa.
An example of a nudge that may be beneficial to a chooser’s decisional
autonomy would be one that, as alluded to above, provides choosers with
information about social norms. For example, nudges have been developed that
inform choosers of their standing vis-à-vis their neighbours when it comes to
domestic energy usage.262 Such information is relevant to many choosers and thus
would form a part of many choosers’ decision-making. These choosers are driven by
an authentic motivation to be part of the in-group and if a nudge provides choosers
with such peer information, then the nudge is assisting those choosers to discover
262 See e.g. Kevin Charles Redmon, “Who’s Saving Electricity in Your Neighborhood?”, Pacific Stand (March 2012), online: <http://www.psmag.com/nature-and-technology/whos-saving-electricity-in-your-neighborhood-39932>.
255
what choice option matches their authentic motivations – a net positive for
decisional autonomy. On the effectiveness side, care must be taken when designing
nudges, as backfiring has reportedly occurred in some instances, one theory being
that peer information may be demotivating to some, causing them to effectively
“give up.”263
Relevancy here plays the role of an important caveat with nudges that
provide this kind of peer information (and all informational nudges generally).
Relevancy requires both that the information must be particular to the chooser and
the information must be true. Regarding peer information, the peer group must
actually be subjectively relevant to the chooser – i.e. consist of the actual people
whose energy usage a chooser is interested in. If the information is said to be about a
chooser’s neighbourhood, but is actually about a different neighborhood, then such
information is not relevant to the chooser (so no increase in decisional autonomy).
Relevancy and truth go hand-in-hand. If, for example, a chooser is led to
believe that a certain option will yield x benefits (e.g. social standing), but the option
does not in fact lead to such benefits, then the information is not actually relevant to
the chooser. And because the chooser has been led to select a particular option based
on that information, it is likely that the selected option may not align with their
263 John Beshears et al, “The Effect of Providing Peer Information on Retirement Savings Decisions” (2015) 70:3 J Finance 1161.
256
authentic motivations. In other words, by providing choosers with false information,
the nudge has robbed choosers of the opportunity to put their authentic motivations
into action at the option evaluation level of process autonomy.
However, false informational nudges are also powerful because the
information has the appearance of relevance, when there is no such thing. For
example, telling choosers there will be a high turnout in an election (regardless of
the truth of that statement) has been shown to actually increase voter turnout.264 The
turnout information during an election is relevant to choosers because they want to
be in the in-group. But providing them with false information will not help them
match their desire to be in the in-group with the appropriate option selection and
may in fact lead them in the opposite direction of their authentic selves.
Relevancy can also be enhanced by nudges that simplify information about
options or opportunity sets, without objectively affecting the opportunity set in a
non-libertarian manner. For example, the US efforts mentioned earlier at revising
the nutritional guide to MyPlate, simplifies the same nutritional information in a
manner that is more relevant to choosers. A food pyramid is far less relevant than a
plate which people actually use. The nutritional information is more accessible, and
thus arguably less shrouded. Such nudges that simplify the presentation of relevant
264 Alan S Gerber & Todd Rogers, “Descriptive Social Norms and Motivation to Vote: Everybody’s Voting and so Should You” (2009) 71:1 J Polit 178.
257
information can make that information even more relevant to choosers, thereby
enhancing their decisional autonomy.
One further qualifier – we must draw the line between objective and
subjective information (not relevancy). The foregoing ethical analysis applies mainly
to the provision of objective information. Providing choosers with subjective
information is a different type of nudging that does not implicate decisional
autonomy. Consider that branding and other marketing nudges may provide the
illusion of certain costs and benefits of particular products. In fact, all marketing
efforts are nudges designed to “creat[e] the illusion that the product can satisfy
desires that it probably cannot.”265 Note the last part of that quote is a subjective
determination.
For example, Apple has built through its marketing efforts the illusion that its
products are “cool.” For some choosers, purchasing Apple products will in fact
make them feel more “cool.” For others, there will be disappointment, but the
“information” provided (i.e. the perception created) never factually promised
anything. And because that information was subjective, it cannot readily be subject
to any false/true distinction.
To summarize this section, nudges can affect information about a chooser’s
biases, the opportunity sets they face and the objective characteristics of options
265 Handelman, supra note 175 at 55.
258
within those sets. Nudges that provide choosers with more relevant, objective
information will be found to enhance the decisional autonomy of choosers. Nudges
that reduce the relevant information provided will have the opposite effect on
process autonomy and must be considered unethical. Finally, those nudges that
merely work on subjective perceptions and do not provide information per se, will,
because we cannot determine the truth of those efforts, be found to simply preserve
the decisional autonomy of choosers.
6.8. Conclusion
This chapter has shown that nudges can affect a chooser’s process autonomy
in many ways. Any negative effects on the two constituent parts of process
autonomy – motivational autonomy and decisional autonomy – will independently
render any such nudge unethical for failing to preserve the process autonomy of
choosers. As well, any process autonomy effects of nudges must be evaluated on a
relative basis (pre-nudge vs. post-nudge) and should not be subject to
aggregation/maximization.
On the whole, to say a chooser’s choices reflect their autonomous self-
government, that chooser must have had an actual opportunity to make their own
decisions by reflecting and deliberating on their authentic motivations and selecting
the options that they think best match those motivations. It is not the same thing to
259
nudge them towards particular options just because those options may be what the
chooser would have selected if they had full process autonomy (assuming we can
even make such a determination). The goal should not be just good decisions
(outcomes), but autonomous decision-making. Whether or not one participates in their
lives should not be dependent on the goodness of the outcomes, but on a respect for
their participatory autonomy.
At the level of motivational autonomy, this chapter has shown that nudges
can work by altering existing motivations or constructing motivations where none
presently exist. Empowering nudges are those that are overt – transparent to the
chooser – so that a chooser can accept or reject any such influence as authentically
altering their motivations. Unethical nudges (i.e. failing to preserve motivational
autonomy) are those nudges that work on altering choosers’ motivations in a covert
manner, preventing choosers from endorsing the new motivations as authentically
their own.
Regarding decisional autonomy, we have seen that nudges can affect
choosers’ deliberation, biases and information. Those nudges that promote or
preserve the opportunity for deliberative, unbiased and informed thinking will be
found to preserve decisional autonomy. Those nudges that instead promote System
1 instinctual responses (which tend to be uninformed and/or biased) must be
260
considered unethical because they take choosers further from a place where their
authentic motivations can effectively govern their behaviour.
261
Chapter 7
Outcome Autonomy
7.1. Introduction
Thus far we have seen how nudges can affect a chooser’s process autonomy,
but they can also affect outcome autonomy. Indeed, many nudges are designed
specifically to enhance a chooser’s outcome autonomy. Since nudges affect which
options are selected by choosers, any change in choices (outcomes) will likely also
lead to a change in outcome autonomy. Recall that outcome autonomy is a
hypothetical construct of process autonomy – an option and resulting choice has
high outcome autonomy if it is what that chooser would have selected as if they had
process autonomy. This and the following chapter will show that outcome
autonomy is a problematic concept, but that there is still a (limited) place for it in
our ethical evaluation of nudges.
Outcome autonomy is the main justification for nudging offered by nudge
proponents, most notably in the form of Sunstein and Thaler’s libertarian
paternalism. However, unlike this thesis, those proponents make outcome
autonomy, not process autonomy, of lexical priority such that the preservation of a
chooser’s outcome autonomy is made to be the primary determinant of ethical
nudging. However, outcome autonomy is in fact derivative of, and dependent on,
262
process autonomy and as such is of lower lexical priority. Thus, any effects of
nudging on the outcome autonomy of choosers can be determinative for our ethical
analysis only when process autonomy is unaffected by a nudge.
This chapter will begin by exploring the connection in outcome autonomy
between motivations and choices. Next, a variety of theoretical approaches to
outcome autonomy will be analyzed, focusing on Sunstein and Thaler’s libertarian
paternalism. Finally, we will explore the limited role of outcome autonomy in our
ethical analysis which restricts the use of outcome autonomy nudges to exceptional
choice scenarios.
7.2. Outcome Autonomy: Alignment Between Choices & Motivations
Despite its many shortcomings, outcome autonomy is an important element
of living autonomously. All else being equal, a person can be considered to have
greater autonomy if they end up with choices (outcomes) that reflect or align with,
their authentic motivations as opposed to choices that do not match their
motivations. This section will show how outcome autonomy is used to set the
direction of nudging.
At its core, nudging is about and motivated by choice outcomes. The whole
point of nudging is to get individuals to make a different choice than they are
currently making (without the nudge). Nudging needs a direction; people need to
263
be nudged towards something, some particular choice/option. Indeed, the premise
of nudging is that people should be nudged, “in directions that will improve their
lives.”266
This outcome-based motivating force is predicated on the assumption that
humans frequently make welfare-reducing errors, often as a result of faulty
reasoning, behavioural biases or the use of heuristics, that may not necessarily
reflect their authentic motivations. As Sunstein and Thaler write: “With respect to
diet, smoking, and drinking, people’s current choices cannot reasonably be claimed
to be the best means of promoting their well-being.”267
So we know that nudges are designed to improve the well-being of choosers.
However, well-being can have several meanings. In the objective sense, it can be
equated with traditional markers of welfare such as health, longevity, maybe even
happiness, as the subtitle to Nudge would suggest.268 Nudges that are motivated by
achieving these goals will generally fall into the category of traditional paternalism.
Here we usually see specific measurements of well-being: “[I]t is legitimate for
choice architects to try to influence people’s behavior in order to make their lives
longer, healthier, and better.”269 There is no concern for process or outcome autonomy.
266 Thaler & Sunstein, "Nudge", supra note 1 at 255. 267 Ibid at 7. 268 Ibid. 269 Ibid at 5 (emphasis added).
264
Traditionally paternalistic nudges could very well cause a reduction in process and
outcome autonomy.
However, we can also conceive of well-being in a different, more subjective
sense consistent with self-government or autonomy. This leaves the determination
of what is in a person’s best interests or their well-being to the chooser themselves –
a person knows best what advances their own well-being since they are the one that
has authentic motivations guiding those judgments. Sunstein ably boils this down
to: “whatever choosers think would make their lives go well.”270
Under this conception, a chooser’s well-being is enhanced if their choices
have outcome autonomy – ending up with choices that align with their authentic
motivations. If a chooser has an authentic motivation to maximize their lifespan,
then resulting choices (i.e. selected options) that reflect that motivation have high
outcome autonomy. If, on the other hand, the chooser’s choice conflicts with that
motivation, we would say that the choice has low, or no, outcome autonomy.
Many of our current choices could be said to lack outcome autonomy as
evidenced by the action-intention gap (or in the language of this thesis: outcome-
motivation gap). This gap arises when individuals have a motivation to do
something (their autonomous will) and then fail to follow through on that
motivation, or they attempt to follow through but make a mistake in their option
270 Sunstein, "Choosing", supra note 12 at 73.
265
selection (as expressed by their actual choices). If a person has a motivation that
translates into doing A and yet they do B, then that is not a reflection of their
autonomy, since their choice does not align with their autonomous motivations. A
person’s outcome autonomy then would be enhanced if an intervention closed the
action-intention gap and aligned motivations (intentions) with choices (actions). The
UK’s Behavioural Insights Team explicitly points to closing the action-intention gap
as a key driver in designing effective nudge interventions.271
7.3. Theoretical Outcome Autonomy Approaches
The extent to which a chooser’s choices align with their motivations is the
sole determination of outcome autonomy. Thus, the motivation side of the equation
is key – we (and choice architects) must know the content of a chooser’s authentic
motivations in order to steer them towards choices that reflect those motivations.
Choice architects can adopt a variety of approaches to solving this problem, varying
with what source they use for determining a chooser’s motivations. The dominant
approach – libertarian paternalism – requires choice architects to appeal to a
hypothetical construct of a chooser with complete decisional autonomy. Other
271 Behavioural Insights Team, Applying Behavioural Insights to Organ Donation: preliminary results from a randomised controlled trial (December 2013), online: <https://www.gov.uk/government/publications?departments%5B%5D=behavioural-insights-team>.
266
approaches require varying degrees of (but not full) decisional autonomy. The next
two subsections will explore these approaches.
7.3.1. Libertarian Paternalism
We will begin with a brief detour into the lexicology of libertarian
paternalism, since the term seems oxymoronic, and has been claimed as such.272
However, that conclusion is based on a conflation of liberty and autonomy in the
way that Thaler and Sunstein deploy the term “paternalism.” The first part of
libertarian paternalism – the libertarian part – refers to the means of nudging, akin
to our liberty of choice constraint. It is the second part of libertarian paternalism –
the paternalistic part – that refers to outcome autonomy. Sunstein and Thaler’s use
of “paternalism” refers to the fact that the actual will of choosers (their choices) are
being softly interfered with by nudges, but are being done so with some sort of
autonomous consent – outcome autonomy.
Libertarian paternalist theory requires that choice architects appeal to an
autonomous hypothetical construct of the chooser to determine what choice that
construct would make, and then the choice architect is to nudge choosers toward
that choice outcome. Sunstein and Thaler’s theory of libertarian paternalism clearly
272 See e.g. Gregory Mitchell, “Libertarian Paternalism Is An Oxymoron” (2005) 99:3 Northwest Univ Law Rev 1245; Vipin P Veetil, “Libertarian paternalism is an oxymoron: an essay in defence of liberty” (2011) 31 Eur J Law Econ 321. For Sunstein and Thaler’s responses see: Sunstein & Thaler, supra note 13.
267
endorses outcome autonomy by providing that nudges should, “influence choices in
a way that will make choosers better off, as judged by themselves.”273
However, this definition of libertarian paternalism doesn’t get us very far.
Rather, it would seem to take us back to the neoclassical economic account – a
person’s judgment about their well-being is revealed by the choices they actually
make. Recognizing this problem, Sunstein and Thaler provide a further qualifier to
their libertarian paternalism definition – a nudge is libertarian paternalistic if it
steers choosers towards a choice that they would make as-if they “had paid full
attention and possessed complete information, unlimited cognitive abilities, and
complete self-control.”274 We have seen most of those elements before: they very
closely reflect the three parts of decisional autonomy – information, unbiased
cognition and deliberation. Note that deliberation is the only element missing from
libertarian paternalism, replaced by self-control.
Thaler argues that their conception of libertarian paternalism amounts to
“libertarian best guess.”275 This means that the choice architect must make their best
guess as to what the chooser would choose for themselves if they had the time and
273 Thaler & Sunstein, "Nudge", supra note 1 at 5 (emphasis in original). 274 Ibid at 5. 275 Richard Thaler, “The Argument Clinic”, (2010), online: Cato Unbound <http://www.cato-unbound.org/2010/04/16/richard-thaler/argument-clinic>.
268
expertise to make an informed choice.”276 Thaler’s definition thus seems to
contemplate some element of deliberation in providing for the “time” requirement.
Notice however that libertarian paternalism, as with all outcome autonomy
approaches, does not actually require the chooser to possess any process autonomy
or to participate in any meaningful way in their own decision-making. The idea is to
match motivations to choices by eliminating many of the behavioural gaps that arise
at the choice evaluation stage of decision-making. That is a sensible goal to
maximize outcome autonomy; however, it ignores the essence of autonomy as
participatory self-government. Autonomy is not just about realizing motivations; it is
about discovering and exercising motivations.
The one interesting omission from the libertarian paternalist framework is
any requirement that the hypothetical construct possess motivational autonomy.
When the determination is made about the direction of nudging, it is done by
reference to a hypothetical construct of the chooser – one that is imagined to have
many of the elements of decisional autonomy, but not necessarily authentic
motivations. This is a glaring omission, as steering a chooser towards a choice that
reflects their inauthentic motivations is not respectful of either process or outcome
autonomy.
276 Ibid.
269
Sunstein and Thaler are interested in nudging people towards choices that
they would make in line with their own (possibly inauthentic) motivations if they
were able to fully access and give effect to those motivations at the option
evaluation stage of decision-making. Thus, we can say that libertarian paternalism
refers to a hypothetical construct of a chooser that possess decisional autonomy, but
not full process autonomy since no inquiry is made into the authenticity of the
chooser’s motivations (e.g. the motivations may be tainted by undue external
influences). While libertarian paternalism does not go far enough into inquiring as
to the authenticity of a person’s motivations, it does ensure (if functioning properly)
that a chooser will end up with a choice outcome that would reflect what they
would have chosen from a menu of options using a deliberative, unbiased decision-
making process.
An example may help to clarify how libertarian paternalism functions and
also show how convoluted relying on outcome autonomy can be. Let’s imagine a
chooser who is renewing their driver’s license and is faced with the task of deciding
whether or not to donate their organs. In an opt-in system (one-click) they must
choose to donate, but our chooser does not (as most do not – recall that only 28% of
Ontarians are registered donors). The low consent rate is problematic from a social
welfare perspective and also because we have reason to suspect that because of the
biases that go into the stickiness of defaults, the “decision” not to donate may not be
270
what individuals would choose if they had full decisional autonomy. In other
words, we have reason to be believe that choosers’ authentic motivations may in fact
be to consent to donate their organs.
Enter the proposed (and in some countries, adopted) nudge of flipping the
default and adopting an opt-out system where individuals are presumed to consent
to donate their organs unless they expressly act otherwise. Such a nudge is then
justified using the idea of outcome autonomy as per libertarian paternalism. The
assumption is that if people had full information, were not lazy (inertia), and were
not biased into avoiding thinking about death, they would probably consent to
donate their organs. In other words, most people have an authentic motivation to
donate their organs, but there are decisional problems at the option evaluation level
that are interfering with giving effect to those motivations.
Outcome autonomy would thus be enhanced by such a nudge – choosers end
up with a choice (consenting to donate) that appears to reflect their authentic
motivations. But there is no process autonomy in those choices; the nudge here does
not actually give choosers any improvements to their process autonomy either in the
form of de-biasing assistance, help to overcome their inertia or an active choosing
frame. Instead, it is imagined what choice would result if those process
improvement were made, and then the chooser is steered in that direction, but with
the same biases and inertia still intact. We cannot actually ascribe the choice to
271
donate to the choosers in a manner that we can say that the choice is a result of a
process of self-governance. Thus we end up with a choice that has high outcome
autonomy, but little true autonomy because process autonomy was bypassed.
Yet, nudge proponents often conflate the two types of autonomy, thinking
that outcome autonomy is indeed autonomy. Consider the following statement from
the bureaucrat in charge of a new default savings program in the UK: “We're seeing
over 90 percent of people sticking with it…That's 9 out of 10 people are [sic]
deciding that this is the right thing for them.”277 We simply cannot make any
inferences of that nature from the data – very few choosers may have actually
decided anything. It is possible that some of the choosers have simply not
considered the matter at all, if they were even aware of it.
Furthermore, the opt-out data post-nudge (e.g. ~90%+ consent rate) is used to
inform the hypothetical appeal even though choosers are still as decisional
autonomy-deficient as they were pre-nudge (where only ~30% consented). In
essence, pre-nudge we say: “people are lazy and for a host of reasons we don’t
believe that their failure to opt-in is a genuine expression of their autonomy.” Post-
nudge we turn around and then say: “look at how many people failed to opt-out,
they must really want to donate.” In other words, on a process level, we have no
277 Chris Arnold, “Why Is It So Hard to Save? U.K. Shows It Doesn’t Have To Be”, NPR (23 October 2015) online: <http://www.npr.org/2015/10/23/445337261/why-is-it-so-hard-to-save-u-k-shows-it-doesnt-have-to-be?>.
272
reason to believe that “choices” post-nudge are any more reflective of a person’s
process autonomy than they were pre-nudge.
The organ donation example also illustrates a key theme underlying
libertarian paternalism: that the kinds of choices individuals would make if they had
decisional autonomy tend to be those that promote their objective welfare: saving
for retirement, healthy eating, safe driving, etc. After all, these are the types of
choice outcomes that Sunstein and Thaler are really interested in. Note their
statement that: “[I]t is legitimate for choice architects to try to influence people’s
behavior in order to make their lives longer, healthier, and better.”278 What remains
to be seen is what would happen if appeal is made to a person’s hypothetical
decisional-autonomous self and that self directs that choices be made that appear to
run counter to a person’s welfare (determined objectively or in the eyes of the choice
architect). The fact that there are no nudges currently of this nature is curious.
7.3.2. Other Outcome Autonomy Approaches
Libertarian paternalism is the dominant theory of nudging, but there are
other similar approaches that have been proposed (some long before the idea of
nudging even came into existence). All are of the outcome autonomy variety. The
common strand running through these other outcome autonomy theories is an
278 Thaler & Sunstein, "Nudge", supra note 1 at 5.
273
appeal to a more decisionally autonomous version of the chooser to guide the
direction of nudging. This section will briefly consider two such theories: Cass
Sunstein’s “informed chooser” approach, and John Hodson’s “unencumbered
chooser” approach.
Sunstein’s “informed chooser” approach is, unsurprisingly, very close in
principle to libertarian paternalism. However, there are a few key differences. In his
approach, Sunstein suggests that choice architects are asked to steer choosers
towards choice outcomes that they would select if fully informed or to a lesser
extent, “adequately informed” (however that is determined).279 For example,
regarding defaults, Sunstein proposes that choice architects should “[s]elect the
default rule that reflects what most people would choose if they were adequately
informed.”280
This is a slightly lower standard than the full decisional autonomy called on
by libertarian paternalism and actually takes us further from the ideal of full
outcome autonomy based on full process autonomy. Notice that Sunstein’s concern
is only information, not how the information is processed (i.e. with full cognition).
Nor is he concerned that a chooser possesses willpower or time for deliberation.
Likewise, regarding information, the requirement is one of “adequate information”
279 Sunstein, "Choosing", supra note 12 at 73. 280 Ibid at 73.
274
not “complete relevant information.” These qualifications mean that the
hypothetical constructs of choosers are not taken to have full process autonomy
even at the option evaluation stage of decision-making (i.e. decisional autonomy).
Thus, this approach is less likely to result in nudges that enhance even the outcome
autonomy of choosers.
On the flip side, one benefit of Sunstein’s “informed chooser” approach is
that it is more determinative because it is less hypothetical. There are simply fewer
variables that a choice architect would need to account for in their hypothetical
construct and thus the epistemic difficulties encountered by most outcome
autonomy approaches may be lessened.
Just as Sunstein focuses on information, John Hodson focuses on the the non-
informative factors – cognition and willpower – in his argument for intervention in
support of a chooser’s “unencumbered will.” 281 To Hodson, a person’s autonomous
authority (to set the direction of nudging) is found in those decisions a person
makes when unencumbered. Encumbered decisions are those that, “are made in
circumstances which are known to affect decision-making in such a way that the
person making the decisions sometimes come to believe that the decisions were
mistaken or unfortunate.”282 Thus, a chooser would come to regret the decision.
281 John D Hodson, “The Principle of Paternalism” (1977) 14:1 Am Philos Q 61. 282 Ibid at 66.
275
Without engaging in a lengthy discussion of the role of regret, it appears that
encumbrances would thus cover cognitive biases and likely System 1 emotional
reactions that individuals would not wish to have guide their actions. For example,
an encumbered chooser may be one that decides based on their automatic, intuitive
System 1 and this decision conflicts with a person’s authentic motivations. The goal
here then is, just like with libertarian paternalism, to close the action-intention gap
by matching outcomes to motivations. Note that there is no discussion of
information in Hodson’s approach.
The real problem with all of these approaches – libertarian paternalism,
informed choosers, unencumbered choosers – is that they are solely outcome
autonomy-based. None ask choice architects to actually provide choosers with
process autonomy. Instead, they all require appeals to specific hypothetical
constructs of choosers. This places both enormous pressure on getting the evidence
of higher-order motivations right, but it also misses the point of autonomy as being
actually self-governing, not just ending up in the same outcome one would reach if
they were self-governing.
With that said, between all the foregoing approaches, it does appear that
libertarian paternalism may offer the best hope at preserving or enhancing the
outcome autonomy of choosers. Libertarian paternalism accomplishes this by
appealing to a hypothetical construct that possesses all of the elements of decisional
276
autonomy (reading in Thaler’s amendment regarding time for deliberation).
Libertarian paternalism calls on a choice architect to identify a chooser’s higher-
order motivations (not necessarily authentic though) and then nudge the chooser
towards choice outcomes that reflect those motivations, since there should be no
problems at the option evaluation stage that would prevent the hypothetical
construct from accessing and acting on those motivations.
However, note that libertarian paternalism still falls short of perfect outcome
autonomy because it does not inquire into, nor require, that choosers be steered in
directions consistent with authentic motivations. Rather, while a chooser may be
nudged towards decisions consistent with their higher-order motivations, those
motivations may fail to be authentic. Thus it is entirely possible that libertarian
paternalism may end up causing an overall decrease in outcome autonomy if
choosers are nudged to choice outcomes that reflect their higher-order motivations,
but those outcomes are different from those that reflect their authentic motivations.
7.4. Outcome Autonomy as a Tool of Last Resort
Outcome autonomy is an important element of ethical nudging; however, for
the reasons outlined thus far, and those in the next chapter, it should not have the
lexical priority that is ascribed to it by nudge proponents. Instead, it must be
subordinated to the far more superior concept of process autonomy. Any effect of a
277
nudge on process autonomy will generally be determinative for the ethical
evaluation of that nudge.
As well, process autonomy is the best way to close the action-intention gap
and improve outcome autonomy. Providing choosers with an opportunity to
exercise self-government through process autonomy that accesses and enables
authentic motivations is the best method to ensure choices match those authentic
motivations. Outcome autonomy should flow from a good exercise of process
autonomy. There will be no reference to a hypothetical construct required – what
the chooser would select if autonomous is what the chooser does select when they
are autonomous. Thus, the best chance we have for positively affecting the outcome
autonomy of choosers resides with the promise of process autonomy.
However, not all nudges will, or can (due to the design of the choice
scenario), affect process autonomy. Thus, to evaluate those nudges on autonomy
grounds, we must turn to outcome autonomy. All else being equal, and assuming
there is no effect of a nudge on process autonomy, a nudge that moves choosers
towards a choice that they would have selected if they had more process autonomy
(i.e. consistent with authentic motivations (a part of process autonomy)), must be
considered ethically superior to moving them to a choice with less outcome
autonomy (i.e. traditional paternalism that does not take into account chooser
motivations).
278
Outcome autonomy still has autonomy value; it is not a concept with zero
utility. The exercise of good, even perfect process autonomy, may not necessarily be
enough to live a completely autonomous life. One must still end up with outcomes
that reflect their authentic motivations. Consider again organ donation. During his
life a chooser might exercise very good process autonomy in deciding he would like
to donate his organs, and he may indicate as much on the registry. However, when
the chooser dies, a physician might approach the chooser’s family to inquire about
organ donation. If the family were to disrespect the chooser’s wish to donate and
refuse consent to a donation, that chooser’s autonomy is disrespected, but at the
outcome autonomy level.
This example shows that outcome autonomy does not always automatically
flow from process autonomy. Sometimes one does not achieve the full realization of
process autonomy until some future time. However, in order to give effect to a
chooser’s process autonomy, the chooser ought to end up with outcomes that reflect
that process.
The organ donation example also shows that outcome autonomy derives its
value from process autonomy, and is not an independent concept. Where there is
high process autonomy (i.e. strong, authentic motivations), there will be outcomes
with high outcome autonomy. However, in cases where there is zero process
279
autonomy, there is zero outcome autonomy, since there are no authentic
motivations to which final outcomes can be matched as a matter of autonomy.
Thus, if the only effect of a nudge is to increase outcome autonomy, the
nudge must be ethical because some sense of autonomy of the chooser is improved.
On the flip side, if the only effect is to reduce outcome autonomy, then the nudge is
unethical because the chooser’s overall autonomy has not been preserved by the
nudge. Thus, our autonomy restraint would not be satisfied.
7.5. Conclusion
In this chapter we have seen that outcome autonomy is a characteristic of
choices (i.e. selected options), and thus all nudges that are effective will likely affect
outcome autonomy. Those choices that reflect what a hypothetically more
autonomous version of a chooser would have selected (i.e. better reflect that
chooser’s higher-order motivations (preferably authentic)), have greater outcome
autonomy than those choices that do not.
Despite its use in the design of many nudges, outcome autonomy is lexically
inferior to process autonomy. Thus, libertarian paternalism and other outcome
autonomy approaches should only be used as tools of last resort. That is, they have a
limited role, restricted to choice scenarios where a nudge has no effect on process
autonomy. Only then should choice architects adopt nudges that improve the
280
outcome autonomy of choosers. Among outcome autonomy approaches, libertarian
paternalism is the best option for achieving greater outcome autonomy.
281
Chapter 8
The Challenges of Outcome Autonomy
8.1. Introduction
This chapter, along with its earlier counterpart on the benefits of process
autonomy, show why outcome autonomy is a problematic concept that should only
be used in exceptional cases as an ethical tool of last resort. The directions of nudges
should be determined by the chooser themselves, not on a hypothetical basis (using
the construct of outcome autonomy), but by actually possessing and exercising
process autonomy. This means having authentic motivations and accessing those
motivations at the option evaluation stage of decision-making. Nudges should
facilitate this.
There are two types of problems with outcome autonomy, what we will call
theoretical problems and practical problems. The theoretical problems have already
been addressed in the preceding chapters on process autonomy. Outcome autonomy
is derivative of, and not a substitution for, process autonomy (and lexically ranked
accordingly). The theoretical problems with outcome autonomy are the mirror
opposite of the benefits that give process autonomy its precedence. Namely, being
autonomous means experiencing and acting autonomously (i.e. living and choosing
282
autonomously), not just ending up with a result that one may have hypothetically
chosen if they had varying degrees of process autonomy.
This chapter is instead focused mainly on the practical problems of outcome
autonomy. This chapter will explore the following practical problems: (1) the
epistemic problem inherent in the hypothetical nature of outcome autonomy
approaches; (2) related to the former, an inability for outcome autonomy to address
undecided choosers; (3) the difficulty with determining the outcome autonomy
effects of nudging, particularly in scenarios of truly inevitable choice architecture;
(4) the problem of aggregating outcome autonomy; (5) short-term behaviour change
vs. long-term motivational change; and (6) an inapplicability to all choice scenarios.
Notwithstanding these practical limitations, outcome autonomy does have
value as a concept. And in the rare cases where the practical limitations are
overcome, outcome autonomy can serve a legitimate role as our guide for evaluating
the autonomy effects of nudging. But those circumstances should be reserved
exclusively for situations where process autonomy has a diminished role due to the
nature of the particular choice problem.
8.2. The Epistemic Problem
As we saw in the previous chapter, outcome autonomy nudges (e.g.
libertarian paternalist nudges) are to steer choosers towards options that reflect their
283
motivations. If choice architects are to be successful at designing and implementing
such nudges, they clearly need to know the content of choosers’ motivations (ideally
authentic motivations). This creates an epistemic problem (i.e. information
constraint)283 – whether choice architects have, or could have, the requisite
knowledge to nudge individuals toward choices that they hypothetically would
have chosen if they had exercised greater process autonomy (or in the case of
libertarian paternalist nudges – decisional autonomy).284
However, the hypothetical nature of this exercise creates uncertainty as to
whatever conclusions a choice architect may reach about a person’s outcome
autonomy. As Sugden asks: “How, without making normative judgments, do we
determine what counts as complete information, unlimited cognition, or complete
willpower?”285
The knowledge requirement of outcome autonomy does not imply that
choosers are unaware of their own motivations. Outcome autonomy theory and
nudges appear to make no such claims. Outcome autonomy carries no explicit or
implicit comment about a chooser’s level of awareness of their own motivations.
Indeed, it allows for a chooser to be fully aware of their authentic motivations. But,
283 See e.g. William Glod, “How Nudges Often Fail to Treat People According to Their Own Preferences” (2015) 41:4 Soc Theory Pract 599. 284 On this point, Mark White, among others, is firmly in the “no” camp, given the “counterfactual” nature of the exercise: White, supra note 6. 285 Sugden, supra note 191 at 232.
284
as we have learned from behavioural insights (e.g. the action-intention gap),
awareness of one’s motivations does not guarantee that one selects options in line
with those motivations. Choice architects may help in that regard by steering
choosers toward choices that reflect choosers’ authentic motivations. Thus, choice
architects don’t need better knowledge of a chooser’s motivations than the choosers
themselves; they just need to be better than choosers at matching choices with those
motivations.
One cannot say that a chooser’s pre-nudge choices are not reflective of a
person’s outcome autonomy, without an understanding of what their outcome
autonomy is. There are three kinds of knowledge that are needed for choice
architects to successfully nudge choosers toward choices that respect their outcome
autonomy: (1) knowledge of chooser’s motivations; (2) knowledge of the
authenticity of those motivations; and (3) knowledge of which choice outcomes
reflect those motivations.
The appeal to, and use by, a choice architect of the foregoing forms of
knowledge is not a linear process. The knowledge-gathering process actually starts
with suspicion (ideally a knowledgeable suspicion) that a chooser’s current choices
do not reflect their authentic motivations. This suspicion will have to be empirically
grounded since the only real evidence we have of a chooser’s motivations prior to
that point are their actual choices. However, behavioural insights suggests that we
285
should view such choices with doubt and that we should at least hesitate in
inferring motivations from choices.
To conclude that a chooser’s present choices do not reflect their authentic
motivations, a choice architect obviously must have knowledge of that chooser’s
authentic motivations. So then, how do nudge proponents determine what choosers’
motivations are, and how do they verify that they are authentic? The authenticity
requirement demands that choice architects verify that the motivations they are
relying on in designing their nudge are the highest-available motivation of that
chooser, such that those motivations are not qualified by context-dependent factors.
Furthermore, those motivations must be independent from undue external
influence.
Oddly, a frequent form of evidence as to choosers’ authentic motivations is
their behaviour in the marketplace. As Thaler (by himself) writes: “If we talk more
about nudges for things like savings, smoking, and weight loss, that is because the
evidence there is strongest that people want such nudges (and pay for them in the
market).”286 Note the last parenthetical. Using the marketplace without caution as an
indicator of choosers’ authentic motivations is contrary to the very behavioural
insights that have led us to doubt individuals’ actual choices as being reflective of
their authentic motivations. If, informed by behavioural insights, we are skeptical
286 Thaler, supra note 275.
286
that individual consumption decisions (e.g. buying a cheeseburger and fries) are
indicative of a person’s authentic motivations, then, without more, we should be just
as skeptical about concluding that other consumption decisions (e.g. joining a gym
or diet program) are now reflective of their authentic motivations.
Another common method used to determine what a hypothetical construct of
a chooser would select is to use survey data as evidence of authentic motivations.
Surveys have the advantage of asking individuals about their motivations – what
they value and believe – in an abstract, detached method. This has the effect of
significantly reducing the possibility that behavioural biases will taint those
motivations as they are accessed and expressed in actual decision-making. In this
regards then, outcome autonomy is superior to doing nothing in the face of
choosers’ behavioural biases.
However, surveys have several drawbacks. First, the flip side of the abstract
benefit of surveys, is that they are just that – abstractions that tend to strip
contextual factors from motivations. They also suffer from the risk of asking the
wrong question and then using the answer as evidence of a motivation that may not
be applicable to a particular choice scenario.
Even if they can be counted on as accurate representations of motivations,
surveys are also problematic because those motivations would be statistical
generalizations and certainly not applicable to each chooser. Since we are interested
287
in autonomy, this represents a problem, as autonomy can generally not be
aggregated (we consider the possibility of aggregating outcome autonomy later in
this chapter). Finally, the process of surveying individuals alters their behaviour as a
nudge would, such that we must wonder whether their answers are a true
expression of their authentic autonomy.
For illustrative purposes, we will use flipping the organ donor consent
default. A driving force behind such proposals is not only that societal welfare will
be improved by enlarging the pool of potential donors, but that chooser autonomy
will also be respected: “Changes in choice architecture would help to ensure more
organs are available, in a way that would not only save lives but also fit with the
wishes of potential donors.”287
Supporters of flipping the default often point to survey data to make their
case that individuals really want to donate (i.e. it is their authentic motivation to do
so), but other systemic and behavioural factors are frustrating that wish. In Canada,
the US and the UK, surveys repeatedly reveal that roughly 90-95% of the population
“support” organ donation.288 We can compare this number with the actual consent
rate of closer to 28-30% (at least in Canada and the UK), suggesting that there must
be a behavioural action-intention gap. In the case of organ donation, the action-
287 Thaler & Sunstein, "Nudge", supra note 1 at 179 (emphasis added). 288 See e.g. Kurtz & Saks, supra note 146 at 782; Ipsos Reid, supra note 146; Organ Donation UK, supra note 146.
288
intention gap is reflected in what is known as the “transplant paradox” – that more
people say they support organ donation, and want to donate, than in fact sign up to
be donors.289 The relevant question then is whether we can count on this “support”
number as evidence of individuals’ authentic motivations?
The first problem is that the question is not the right one – supporting
something (especially when prompted on the phone) is far different from wanting to
do something, particularly when it comes to public goods. Indeed, the survey bears
this out, revealing that a little over half of those that “support” organ donation are
actually “willing to donate” or have already decided to donate.290 It would be
reasonable then to take the approximate 50% of the adult population as having a
higher-order motivation to donate. A choice architect may reasonably rely on such a
number as representing (at least statistically) choosers’ higher-order motivations.
The next piece of knowledge the choice architect must have in crafting an
outcome-autonomy preserving nudge is whether that motivation is authentic. There
are two problems with surveys that go to authenticity. The first is that the survey
may count as an undue external influence (and therefore not reveal an authentic
motivation), since the survey may itself be a nudge. Second, we cannot say that the
higher-order motivations are authentic without regard to contextual factors.
289 Kurtz & Saks, supra note 146. 290 See e.g. Ipsos Reid, supra note 146.
289
Regarding the former, evidence shows that surveying individuals may cause
them to change their later behaviour.291 This phenomenon is known as the mere
measurement effect – that merely measuring individual intent actually has a
measurable effect on that intent. For example, a study of potential blood donors
found that those who received a questionnaire about donating blood in fact went
ahead and donated at a statistically significant higher rate than the control group.292
Using a survey as evidence of authentic motivations may also be questionable
since the survey itself may count as an undue external influence. After all, the
individuals questioned by the survey are not aware of the mere measurement effect,
and so they cannot incorporate that information and choose to reject or accept the
external influence on their motivations. This suggests that their motivations may not
be independent. Granted, the mere measurement effect is usually small, and
unlikely to completely reverse a person’s motivations. However, it is certainly a
further complication to using surveys as evidence of authentic motivations.
Further complicating the issue is the framing effect and surveys. Consider
this recent example reported by Ipsos-Mori in relation to its poll question: “Do you
support or oppose [one of the following statements] in the referendum on Britain’s
291 Alix Peterson Zwane et al, “Being surveyed can change later behaviour and related parameter estimates” (2011) 108:5 Proc Natl Acad Sci U S A 1821. 292 Gaston Godin et al, “Asking Questions Changes Behavior: Mere Measurement Effects on Frequency of Blood Donation” (2008) 27:2 Heal Pschology 179.
290
membership of the European Union?”293 If the inserted statement was “Reducing the
voting age from 18 to 16,” 37% supported it, while 56% opposed the statement.
When the question was framed as “Giving 16 and 17 year olds the right to vote,”
52% indicated support while 41% opposed the statement. These obvious framing-
influenced results raise the issue that with surveys, we cannot really know whether
responses ever reflect the authentic motivations of choosers. Which of the above
statements are we supposed to use to determine authentic motivations? Even more
worrisome perhaps is that choice architects can utilize the framing effect in selecting
and designing surveys so that they tend to get the result they want, which they can
then point to as evidence of authentic motivations, even though the results may
simply be framing-influenced.
Regarding contextual factors, as behavioural insights have repeatedly shown,
individuals tend to have context-dependent motivations. However, it is exactly the
opposite – stable, well formed motivations – that outcome autonomy relies on in the
use of survey data. It is one thing to say that you want to donate your organs in the
abstract, but quite another to agree to donate if you had to spend time and energy
sorting through bureaucratic paperwork. The extremely low actual consent rate
must be some evidence as to this difference. What is most likely the case in this
293 Ipsos-Mori, 2015 Political Monitor (12 December 2015), at 26, online: <https://www.ipsos-mori.com/Assets/Docs/Polls/political-monitor-december-2015-charts.pdf> at 26.
291
situation is that many individuals have weak or partially-formed motivations, such
that the inclusion of minor costs qualifies their willingness to donate. Thus, it is
extremely difficult for choice architects to extrapolate from survey data individuals’
authentic motivations.
Consider a cheeseburger lover – they may want to be healthy and eat well,
but just not by giving up their cheeseburger. This is a common problem with dieters
as well – we may take their words and even their actions (e.g. joining dieting
programs/gyms) as evidence of their authentic motivations to lose weight and be
healthy. However, as the failure of most diets show, perhaps we may more
accurately qualify such motivations with contextual factors: sure they want to lose
weight but it quite another thing to do so by giving up every indulgence and going
to the gym two hours a day.
Another part of the epistemic problem vis-à-vis authenticity is that judgments
by choice architects about choosers’ motivations tend to be made in a selective
vacuum, without considering the possibility that a nudge may also have detrimental
outcome autonomy effects on other, related motivations. Recall that a nudge ought
to respect a chooser’s authentic motivations and authenticity is in part determined by
a person being governed by their highest available higher-order motivation. Thus, if a
nudge increased a higher-order motivation but it was not the highest one available,
and that same nudge actually caused a decrease in outcome autonomy related to the
292
latter, then outcome autonomy is not enhanced or preserved. In other words – a
person may be guided by several motivations, but for our outcome autonomy
purposes, the highest motivation is what matters. And if there is not a clearly
superior authentic motivation, then we must consider the equal claims of competing
motivations.
This matter goes to the supposed distinction between means and ends.
Sunstein and other means paternalists ignore the possibility that means may be ends
in themselves and thus potentially have an equal claim to respect on an outcome
autonomy basis.294 The point is that the journey (or means) sometimes is an end in
itself. As J.S. Mill wrote: “the only freedom which deserves the name, is that of
pursuing our own good in our own way.”295
Sunstein’s GPS is a good illustration of this problem. Sunstein calls a GPS an
“iconic nudge” and cites it as an example of a means paternalistic nudge.296 The user
inputs their end (destination) and the GPS default setting (a nudge within a nudge?)
is to show them the best means to get there – the fastest route. But the destination is
not only the end or motivation at issue. Maybe the traveler is concerned about fuel
economy and wants to avoid a mountainous journey. Maybe they want to take the
shortest route by distance, or the most scenic. The point is that the means are ends in
294 Sunstein, "Why Nudge", supra note 26. 295 Mill, supra note 114 at 18 (emphasis added). 296 Sunstein, "Why Nudge", supra note 26 at 61.
293
themselves. Thus the traveler’s motivation is not just to get a destination, but to get
to the destination in a particular fashion.
The GPS example is useful for illustrative purposes because it shows that our
lives and choices are so complex and interconnected, that it is difficult, if not
impossible, to distill individual’s ends down to one motivation in such a way that
we can confidently nudge them in a particular direction. Thus, to properly nudge
choosers in an outcome autonomy-respecting direction, extremely detailed
knowledge is required about the content of a chooser’s motivations.
Then there are evidential questions arising from circumstances where we
cannot say for certain what a hypothetical autonomous version of a chooser would
select (i.e. we cannot definitively determine their authentic motivations). The first
point to note here is that our standard of evidence should be one of reasonableness.
Because outcome autonomy is the ethical framework of last resort – that is, we are
not at risk of causing a loss to any choosers’ process autonomy – we need not
require perfect knowledge. Reasonable knowledge based on reasonable evidence,
taking into account contextual factors and the possibility of similarly ranked
motivations should be sufficient.
However, it is prudent to be mindful of questions surrounding the reliability
of evidence and the biases of those evaluating evidence. One recent study showed
that different groups of researchers who were provided with the same data set
294
reached entirely divergent sets of conclusions.297 This goes to Glaeser’s concern that
choice architects may be inaccurate in their judgments about choosers’ authentic
motivations because the choice architects may be suffering from behavioural biases
themselves.298 This is a possibility, however it may also be applicable to process
autonomy nudges as choice architects may make inaccurate (biased) judgments
about what nudges cause process autonomy improvements. Due to this universal
risk, we will set this concern aside moving-forward.
Let’s also put the epistemic motivation question aside and assume that we
have perfect knowledge of a chooser’s authentic motivations. The final knowledge
choice architects must have in outcome autonomy approaches is what choice
outcomes match those motivations. This is likely an easy question, since that is why
a choice architect started working towards implementing a nudge in the first place –
if they know what choice outcomes don’t reflect a chooser’s authentic motivations,
then they likely have an idea of what choice outcomes do. Thus, there is little reason
to think that this requirement represents much of an epistemic hurdle to choice
architects.
Overall, we do need to be concerned with the potential for a slippery slope in
the selective application of the epistemic requirements. There is the possibility that a
297 Raphael Silberzahn & Eric L Uhlmann, “Many Hands Make Tight Work” (2015) 526:7572 Nature 189. 298 Edward L Glaeser, “Paternalism and Psychology” (2006) 29:2 Regulation 32.
295
choice architect’s judgments about a person’s authentic motivations will easily slip
into an objective welfarist determination of what the choice architects would choose
or what they think the chooser should choose. When evaluating nudges using
outcome autonomy, we must not become selective to the point that some choice
outcomes are respected and others are not simply because of the objective welfarist
result of those choices. In other words, we cannot simply infer that because a choice
is the “healthy” choice or “financially prudent” choice that it reflects a chooser’s
authentic motivations. We should make no presumptions; evidence is required in
every case in order to justify a nudge on outcome autonomy grounds.
As we have seen throughout this thesis, the best method for getting choosers
to select choices most in line with their authentic motivations is to use nudge
methods that improve a chooser’s process autonomy. A choice architect doesn’t
need to guess what a fully autonomous version of a chooser would choose if the
chooser is actually fully autonomous. That is not to say that we cannot make an
effort to overcome the epistemic problem, but the complex nature of the problem
(and the resulting reliance on hypotheticals) suggests that this will be difficult, and
is the primary practical impediment to utilizing outcome autonomy as an ethical
evaluation tool.
In conclusion, this section has shown that the epistemic requirements of
outcome autonomy approaches are extremely problematic and in many cases may
296
be practical impediments to using outcome autonomy as an ethical evaluation tool.
It is particularly difficult for choice architects to guess at choosers’ authentic
motivations without more probing into context, and an exploration of what other
motivations may be at play.
8.3. Undecided Choosers
Outcome autonomy nudges have trouble in situations where there are no
clear authentic motivations. Outcome autonomy is a content-based concept: it is not
the act of choosing that matters so much as the content of the choices and whether
they match the content of a person’s motivations. Where there is no content, or the
content is indeterminate, outcome autonomy is effectively “lost,” unable to provide
guidance to choice architects.
As alluded to earlier, individuals with weak motivations pose a problem for
choice architects that wish to deploy outcome autonomy-preserving nudges. At a
minimum, they require a detailed inquiry into contextual factors that may cause
individuals to abandon or change their motivations in a manner that would alter the
selection of outcome-autonomous choice outcomes. Similarly complicating matters
are those choosers who are undecided or who have decided simply not to choose.
Outcome autonomy approaches have no way of accommodating such motivations.
297
Consider our paradigmatic nudge example of the order of food in a cafeteria.
If we know that choosers have motivations for healthy living, we can set up the
cafeteria accordingly. A person whose authentic motivation is to live a long and
healthy life will be better off on an outcome autonomy basis if the healthier food is
placed earlier in line. However, what is a choice architect to do in light of an
individual who has no particular authentic motivation regarding food?
One response might be that the chooser does have an authentic motivation,
but that motivation is to be the type of person that is ruled by their lower-order
desires. But even that is fairly useless for a choice architect that is trying to protect a
chooser’s outcome autonomy. Determining what food will best reflect a chooser’s
lower-order desires is an even more difficult task for choice architects – there is no
way of determining in advance what someone will choose in the moment.
An even greater problem is presented by those choosers who are undecided
on the matter – they have no particular motivation at all. Perhaps a chooser simply
has not thought about the situation, so we cannot say they have a motivation about
what kind of person they want to be. How is the food to be arranged? What choice
outcomes best reflect their autonomy? Outcome autonomy has no answer for this
situation, because it is indeterminate. From the choice architect’s perspective, none
of the food choices have outcome autonomy. However, because our standard is one
of autonomy preservation, if they had no outcome autonomy pre-nudge (which
298
would be the case if they have no relevant authentic motivations), then a shift to
another nudge with little outcome autonomy would not count as an autonomy
reduction. This is another fact in support of our standard of autonomy preservation.
In these cases of undecided choosers, it seems the best option on outcome
autonomy grounds, and assuming that process autonomy nudges are not available,
is to leave choosers to be governed by their lower-order desires. Doing so would
respect whatever minimal autonomy resides in a person being governed solely by
lower-order desires without associated higher-order motivations. This is certainly a
better option than preventing them from choosing at all.
Note that this is a different case from choosers whose motivations are
indeterminate because of epistemic problems. The latter encompass situations where
choice architects know that a chooser has a motivation, but is not able to ascertain
the content of that motivation with any reasonable level of certainty. In that case it is
better not to engage in any choice architecture for fear of causing an unintended
decrease in outcome autonomy. Contrast this with the true undecided chooser – the
choice architect knows that the chooser has no higher-order motivation. In instances
of the truly undecided chooser, choice architects arguably have wider latitude to
engage in choice architecture so long as a chooser’s lower-order desires are
permitted to govern their choice selections. There is little risk in those situations of
299
causing a decrease in outcome autonomy (and assuming process autonomy is
appropriately addressed).
Finally, we need to address circumstances where a chooser has an authentic
motivation not to choose.299 This motivation may be expressed in several ways –
either by delegating a proxy to the choice architect (“you choose for me” or
“whatever you suggest”), or by simply refusing to engage in the desired choice
activity. In either case, it seems that the motivation to not choose is a valid authentic
motivation. So long as it is authentically the highest-order and independent from
undue external influence, that motivation ought to be respected by choice architects
in deploying outcome autonomy strategies.
So what would an outcome autonomy nudge look like that respected an
individual’s authentic motivation to not choose? If the motivation takes the
delegation or proxy form, then the choice architect has carte blanche to select the
relevant option that the chooser will be nudged towards. If, on the other hand, the
motivation is one of an absolute refusal to choose, then out of respect for that
motivation, no nudging may be the most appropriate course so as not to risk a
decrease in outcome autonomy.
If “no choice” is an option within the opportunity set, then that would
obviously be the preferred end result of a nudge. However, in the case of defaults
299 For an excellent exploration of this phenomena see Sunstein, "Choosing", supra note 12.
300
where something must result if the chooser does nothing, choice architects should
set the default such that the burden of commitments on individuals is minimized.
Presuming consent where there is clearly no motivation to consent, nor an
opportunity for the individual to express that consent or lack thereof themselves, is
a far more worrisome outcome vis-à-vis a chooser’s autonomy than presuming “no
consent” where there is no such consent.
8.4. Short-Term Behaviour vs. Long-Term Motivations
A nudge that relies on outcome autonomy for direction requires constant,
ongoing monitoring and adjustment. Individual motivations change over time and
in response to experiences from past choices, so outcome autonomy nudges must
continually adapt to mirror those changes in motivation.
For example, consider a chooser who was previously unmotivated by
environmental concerns but who has, with shifting social norms, become more
environmentally-conscious. A choice option that they may have been nudged
toward earlier may have had high outcome autonomy, but now an environmentally-
friendly option has the higher claim to outcome autonomy, suggesting that the
nudge should be adapted as well to reflect that change. This burden of monitoring
and adjustment speaks in part to the epistemic problem – it is difficult, on a
301
continuing basis, to accurately measure authentic motivations and nudge choosers
towards choices that reflect those (possibly shifting) motivations.
The nature of outcome autonomy being not an actual participatory
expression of a chooser’s authentic motivations also means that decision-making is
never improved, necessitating constant nudging to correct for outcome autonomy
errors as a result of biases and heuristics (i.e. deficient decision-making). People’s
habits for example aren’t eliminated or overcome, they are simply redirected
towards different options. This focus on short-term behaviour guided by extrinsic
motivation rather than long-term, intrinsic motivations is a key shortcoming of
outcome autonomy. This approach necessitates near-constant nudging to account
for those behavioural issues. If nudging were to stop, it is likely that individuals’
option selections would go back to non-outcome autonomy selections. In contrast,
process autonomy nudges provide choosers with learning opportunities and areas
for them to improve their motivations and habits, so that outcomes may improve
organically and for the longer-term.
Similarly, outcome autonomy nudging by the public sector may result in a
“battle of nudging” with the private sector. Imagine a private marketer that nudges
consumers to buy their particular product (shouldn’t be too hard to imagine as it is
the raison d'être of all marketing). This is done not on an outcome autonomy basis
(choosers’ autonomy is not intended to be enhanced), but on a self-interested basis
302
on the part of the marketer. But the desired effect is achieved by working on
outcomes, not on process – a person is not reasoned with to autonomously have an
authentic motivation for that particular product over its competitors. The
government, thinking that choosers would not select that option on an as-if
autonomy basis, then nudges consumers in response in a different direction (e.g.
counter-biasing). The private marketer then responds, and so on. Government does
not have the resources nor is it nimble enough to win a nudge arms race.
Then we have to think about where that leaves the consumer. While the
government may be acting to protect consumer autonomy, private companies may
not, such that the net result for the consumer is a loss of autonomy. Also, by
focusing solely on resulting choices, the benefits of process autonomy are missed
entirely – if individuals were provided the opportunity to discover and express their
own authentic motivations in a participatory manner, then constant nudging and re-
nudging might be rendered unnecessary.
Finally, there is also the signaling effect of outcome-driven nudging by
government. Public outcome-driven nudges effectively license and make acceptable
the technique of avoiding individual process autonomy. The private sector may take
its cue and reason that “if government is doing it, then we can too.” That story will
likely not end well for the autonomy of the consumer, our ostensible motivating
concern.
303
Thus, while nudging on an outcome autonomy basis may be extremely
effective at changing choice outcomes, they necessitate a lifelong commitment on the
part of choice architects and offer little opportunities for individuals to improve
their own decision-making.
8.5. Inapplicability of Outcome Autonomy to Some Choice Scenarios
While outcome autonomy often has a role to play when process autonomy
cannot be called on to evaluate a nudge (i.e. a nudge has no effect on a chooser’s
process autonomy), it’s a concept that perhaps should not have application to all
types of choices. Some choices are inherently about process – that the process of
participating in a particular choice is what gives the choice meaning and we cannot
fathom the choice without that participatory process.
Imagine a heinous nudge focused on voting. We may think that when we
enter a ballot box and cast our vote we do so with full autonomy – we are reflecting
on our options and voting for the candidate that best reflects our autonomous
motivations. However, behavioural insights research suggests that some of those
judgments (about who to vote for) are biased. For example, studies have shown that
during elections voters draw inferences of competence from brief glimpses of the
faces of candidates, contrary to the belief that voting is done on a reflective and
304
deliberative basis (i.e. with full process autonomy).300 Thus we might legitimately
have suspicions that individuals’ votes (choices) do not accurately reflect their
authentic motivations.
However, process autonomy nudges will not be of much help in overcoming
this behavioural bias causing potential motivation-outcome gaps. We can inform
individuals that they are biased, but there is no practical way to eliminate the bias,
particularly because it operates on a generally subconscious level and we’re not
about to eliminate all photograph or video campaign elements. Thus, we may resort
to an outcome autonomy nudge (e.g. a default candidate selection based on an
understanding of voters’ authentic motivations).
Fortunately, however, the epistemic condition may not be met – choice
architects may not have good evidence of voters’ authentic motivations. They may
not confidently know which candidate the majority of affected choosers would
select if unbiased. However, that evidentiary burden may be met in other cases, or at
least a good case for it could be made given the extensive polling data that is
produced each election.301
300 See e.g. Alexander Todorov et al, “Inferences of Competence from Faces Predict Election Outcomes” (2005) 308:5728 Science 1623. 301 Why polls are so often wrong may in fact shed light on this kind of behavioural problem and others related to elections, assuming that we can trust polls as genuine expressions of individual’s authentic motivations (such trust is certainly not a given). But if polls function as such expressions, and yet individuals vote differently than polls predict, perhaps there is a motivation-outcome (intention-action) gap that changes people’s behaviour from their stated intentions. This may be a worthy future project.
305
This is a fantastical example that no reasonable person would advocate.
However, what it illustrates is that outcome autonomy may be useful only for those
types of choices where the process isn’t integral to the choice. Outcome autonomy
has an obvious problem with cases like voting where, even if deficient, we believe
that individuals should be able to exercise whatever autonomy resides in their
decision-making process; their actual participation in their life. Even if we could
resort to outcome autonomy here, we should not, because the outcomes (no matter
how autonomous) are simply irrelevant because it’s the participation (the process)
that matters.
8.6. Determining Effects of Nudges on Outcome Autonomy
Determining whether there has been any change in a chooser’s outcome
autonomy independent of effects on process autonomy is a difficult task. Some
nudges will improve process autonomy yet cause a reduction in outcome autonomy
and vice versa. The possibility of such a result illustrates once again the central
problem with outcome autonomy – its hypothetical nature.
Imagine a simple choice scenario where a chooser has to choose between
three choices – a cheeseburger, soup, or plain salad – from a dining menu. Let’s also
assume that the choices are arranged on the menu in ascending order of
healthfulness, so that the cheeseburger is the least healthy but first and the salad is
306
the healthiest but listed last. Likewise, we’ll assume that the healthfulness of the
choices offsets the satisfaction quality of the food, such that the healthiest food
(plain salad) is the least satisfying and the least healthy food (cheeseburger) is the
most satisfying, with the soup somewhere between the two extremes. Let’s also
assume that all other characteristics of the choices are the same and that the chooser
is deciding solely based on taste (satisfaction) and healthfulness.
Recall that nudges are designed to change choice outcomes; that is the point
of nudging. If nudges did not change choices, they would be useless as modes of
choice architecture. In our scenario then, let’s assume that pre-nudge a person’s
decision-making leads them to select the cheeseburger – the least healthy but most
satisfying food. This can be accomplished for example, by including photographs of
those foods since individuals tend to purchase more of foods that they can visualize.
Or it can be accomplished by other strategic design elements of the menu, such as
noting some as “fan favourites” or “chef’s choices.”
A nudge in this case might use those same techniques to push the healthy
salad instead. So there would be pictures of salads instead of cheeseburgers. The
purpose of the nudge is to increase health-related welfare, so the goal is to get
choosers to select, ideally the salad, but at a minimum, the soup. If they still select
the cheeseburger post-nudge, then the nudge has obviously failed. Note that the
choice architect here is not concerned with maximizing taste-related satisfaction
307
except to the extent that it gets choosers to make healthier choices – they want the
salad and the soup to taste good and be fulfilling so people buy it next time instead
of the cheeseburger.
Since the nudge operates through the same decision-making processes
(predominantly System 1), we’ll assume that there is no change on process
autonomy - the same behavioural biases/process is used to leverage a different
choice outcome in a classic case of re-biasing. So the relevant question for this
section becomes how would such a nudge affect a chooser’s outcome autonomy?
What information do we need to begin that evaluation?
We’ve already discussed the epistemic problems with this scenario. To
evaluate the nudge’s effect on our diner’s outcome autonomy, we need to know the
content of the diner’s authentic motivations. Only by identifying those motivations
can we align choices with those motivations, thus imbuing them with outcome
autonomy.
Let’s assume for our example that our chooser has an authentic motivation to
live as healthy a life as possible, regardless of other, lesser concerns such as taste and
satisfaction. In light of that motivation, we know that the salad has the most
outcome autonomy, the soup a little less, and the cheeseburger the least outcome
autonomy.
308
What is important to note is that the outcome autonomy characteristics of
these choices is the same pre- and post-nudge. In most cases (but not all), you cannot
have a change in outcome autonomy without a change in selected options. Our
hypothetical diner’s motivation to live a healthy life did not change as a result of the
nudge. Nor did the fact that the salad is the healthiest option. All along the two
were aligned as having the most outcome autonomy. Even without a nudge, the
salad would have the highest outcome autonomy value. This is because the chooser
has the same authentic motivations pre- and post-nudge and the autonomy value of
particular options is entirely dependent on their alignment with the chooser’s
authentic motivations. If nudges do not alter a person’s authentic motivations, the
nudge would not have altered the outcome autonomy of the options within the
opportunity set.
In our menu scenario, a nudge would effectively increase a chooser’s
outcome autonomy by leading them to select an option (salad) that is most in line
with their authentic motivations. If the situation were reversed, such that our
chooser selected the salad pre-nudge, then any nudge that caused them to choose
differently would amount to a reduction in outcome autonomy (assuming again that
the nudge did not alter their authentic motivations). To summarize thus far: while a
nudge can alter which option a chooser ultimately selects, a nudge cannot alter the
309
outcome autonomy value of particular options if it does not act on a person’s
authentic motivations, as the latter exist independent of, and prior to, any nudge.
While most nudges won’t affect a chooser’s authentic motivations (and thus
won’t affect the outcome autonomy of particular options), some nudges do have that
effect. As we saw in Chapter 6, educational nudges can be an example of permissible
external influences on motivations. And it is only in cases of permissible external
influences that we can say that a person’s outcome autonomy can change without a
change in actual choice outcomes.
For example, a person may not have an authentic motivation to live a healthy
life until they are informed, or educated about, the importance of a healthy lifestyle.
From that point on, it is conceivable that the chooser would incorporate that view
into their motivational structure; that a good life is a healthy life. Prior to the
educational nudge, unhealthy food choices had greater outcome autonomy than
healthy choices because they were more satisfying. After the nudge, that situation
reversed. The nudge caused the chooser to alter their authentic motivations such
that the unhealthy food no longer carried much outcome autonomy since it was no
longer a reflection of what an autonomous version of that chooser would select.
Now we will consider whether nudges that affect process autonomy also
necessarily affect outcome autonomy. Let’s consider our menu example again,
except that instead of swapping out pictures of cheeseburgers for pictures of salad,
310
our restaurateur includes calorie labels on the menu, as many jurisdictions are now
mandating, and maybe they go a little further than the regulations and grade the
foods as well based on calorie count in a traffic light design. We’ll also assume that
our hypothetical diner reads the calorie labels and includes the information in their
decision-making process. Alas, the cheeseburger remains too tempting and the diner
still selects that option.
The diner’s decision is arguably more informed and one could argue that
their process autonomy has increased as well. Their selection of the cheeseburger is
made on a more informed basis and using the calorie information (ideally) they
have been able to better judge what options best fit into their conception of the good
life. In this case, while the ultimate choice has not changed (they still chose the
cheeseburger), can we say that the outcome autonomy of the choice has improved
since there has been an increase in the process autonomy that led to that choice
outcome?
The answer must be no. There has been no change in outcome autonomy in
this case because there has been no change in the person’s authentic motivations.
While choices can fall along a continuum from non-outcome autonomous to
outcome-autonomous, the outcome autonomy quality of choices cannot be changed
without changing a persons’ authentic motivations. Changing the process by which
choosers select options (even by increasing the process autonomy) does not change
311
whether the choice is one that a hypothetical autonomous version of a chooser
would select (with the exception of nudges that work at the motivation formation
stage of decision-making).
As such, it is likely that a small increase in process autonomy will lead to no
change in outcome autonomy. Our menu example above illustrates this in the case
where the calorie information was added. The diner chose the cheeseburger, just as
they did pre-intervention. Therefore, while there may have been a slight increase in
process autonomy, there was no change in outcome autonomy, because the
cheeseburger was still not the option that the individual would have selected if they
were fully autonomous. Our diner’s authentic motivations to live a healthy life are
unchanged. The fact that they deliberated more because they had greater
information did not change what a fully autonomous construct of themselves would
choose: the healthier salad. That fact remained constant throughout the different
choice scenarios.
What this means is that we do not grade the outcome autonomy of choice
outcomes based on the level of process autonomy used to arrive at those selections.
Instead, we grade the outcome autonomy based on the proximity of the options to
those that a hypothetical construct would have chosen (i.e. alignment with their
authentic motivations). What process autonomy an individual actually possesses
and uses in a particular choice scenario is irrelevant for determining the outcome
312
autonomy of choices, except to the extent that those authentic motivations are altered
by the nudge. It is possible then that there could be a scenario of good, but less than
full process autonomy under which a person selects an option that is not reflective
of their authentic motivations. In this case, even though they have good process
autonomy, the selected option does not carry much outcome autonomy.
One possible explanation for why a person operating under fairly good
process autonomy still selects poor choices on outcome autonomy grounds goes to
the epistemic problem – that we may be misconstruing the content of a person’s
authentic motivations. We may think that a person has an authentic motivation to
live a healthy life, but perhaps that is not their highest-order motivation. This may
happen because a healthy life is not one of their higher-order motivations, it is
conflicting with other similarly ranked motivations, or it is qualified by contextual
factors.
What is clear however, is that any nudge that appears to increase both
process and outcome autonomy is obviously the ideal nudge for autonomy
purposes. If well designed, nudges that cause significant improvements in process
autonomy should, as a side-effect, cause an increase in outcome autonomy. Closely
mirroring that ideal type nudge, would be nudges that increase process autonomy,
but with little to no effect (even collaterally) on a chooser’s outcome autonomy.
313
However, an unfortunate, yet ethically acceptable, side effect of attempts to increase
lexically prior process autonomy may be a decrease in outcome autonomy.
8.7. Aggregating Outcome Autonomy
The epistemic problem is likely the largest practical impediment to using
outcome autonomy as an ethical guide for nudging. A close second has to be that
designing nudges that preserve the outcome autonomy of choosers seems to require
an aggregation of individual outcome autonomy – something like a “utilitarian”
conception of autonomy. And doing so necessarily seems to contradict the
importance of autonomy as individual self-government.
Let’s begin by considering once again our paradigmatic nudge of the order of
food in a cafeteria buffet. This nudge works well for outcome autonomy analysis
because it does not lend itself well to process autonomy improvements. No matter
what kind of information is provided to consumers or how much they are asked to
deliberate on their decisions, the bias to purchase whatever is first in line cannot be
eliminated; the food must appear in some order and the order will still have some
effect. Thus, in that scenario, outcome autonomy may be an appropriate ethical
evaluation tool of last resort.
In our cafeteria example, we’ll assume that the cafeteria proposes switching
from an unhealthy-first to healthy-first arrangement. Instead of donuts and sweets
314
first, salad is first in line. And we’ll assume that the cafeteria manager has good
evidence of all diners’ authentic motivations and which food options match those
motivations. There are also no contextual factors that may qualify those motivations
since they still pass by the food – it’s not as if they have to go to the gym five times a
week to satisfy the motivation. One key qualifier – the healthier food is more
expensive than the unhealthy food (which conveniently happens be true).
Since our evaluative standard is one of autonomy preservation, the key
determinant is whether choosers’ choices post-nudge have less outcome autonomy
than their pre-nudge choices. That is not a difficult standard to imagine being
satisfied.
Let’s say that a plurality (40%) of consumers have an authentic motivation to
live healthy and that the salad would have the greatest outcome autonomy for them.
So on that basis, a healthy-first rearrangement nudge increases their outcome
autonomy. We’ll also assume that another 30% of patrons have no motivations
except to be ruled by their lower-order desires – they will purchase whatever
satisfies their cravings. So their outcome autonomy is unaffected by the nudge since
all choices are still present and visible and they can still act according to their lower-
order desires. Finally, 20% of diners have an authentic motivation to eat frugally.
For those choosers, the option that best matches their motivation is the cheaper
unhealthy food. The remaining 10% of choosers have a variety of other motivations.
315
We’ll also assume that the diners with motivations have equally intensive
motivations. For example, on an intensity grade from 0-1.0, they all have 0.6
intensity. Note however that in practice, determining the intensity of motivations is
another epistemic problem, one that is extremely difficult, if not altogether
impossible. Thus, since intensity cannot be determined with any level of certainty,
we will stick with the “quantity of choosers” approach to make the point about
aggregation.
A frugal-minded diner approaches our post-nudge cafeteria and since they
don’t see the cheaper cinnamon bun right away and they are stuck in front of the
salads, they purchase salads (we’ll set aside for the time being the possibility that
the arrangement may cause a change in authentic motivations). By the time they get
to the dessert section, their plate is full of expensive fruit and salad. So it seems then
that (again assuming that their authentic motivation is to eat the cheapest food), the
nudge reduces their outcome autonomy.
A neoclassical approach would say that perhaps their motivation for the
cheap sweets was not strong enough if they succumbed to the bias in selecting the
fruit. This may be true, but this kind of reasoning cuts both ways. If health-minded
choosers previously selected the sweets because they were arranged first, then their
health-minded motivations may not have been strong. But strength of motivations
does not matter so long as they are authentic – independent and of the highest-order
316
nature. If those requirements are satisfied, then a nudge towards a choice outcome
that does not reflect those (admittedly weak) motivations would have to count as a
reduction in outcome autonomy.
What we see in this example then is that choice architecture affects many
different choosers and those choosers are highly unlikely to share the same
authentic motivations. Some cafeteria patrons may have authentic motivations to be
healthy, some may value taste above health, and some may be guided by frugal
considerations. In other words, goals and motivations are heterogeneous.302 Any
version of the rearrangement nudge is certain to increase outcome autonomy for
some choosers, and reduce it for others. As Sunstein notes, this reality further
underscores the desirability of process-based approaches such as active choosing.303
As tempting as it may be to attempt to easily offset those effects and
determine the net effect on all choosers’ outcome autonomy, the nature of autonomy
complicates such a move. Autonomy, by definition is concerned about the
individual, while nudges are rules or policies of general application driven by
statistical predictability.304 As we saw with process autonomy, we generally cannot
create a meaningful social aggregate of autonomy from a basket of choosers.
302 See e.g. Adrien Barton & Till Grüne-Yanoff, “From Libertarian Paternalism to Nudging—and Beyond” (2015) 6 Rev Philos Psychol 341 at 346. 303 Sunstein, "Choosing", supra note 12. 304 Dan Ariely, Predictably Irrational: The Hidden Forces That Shape Our Decision (New York: Harper Perennial, 2010).
317
The problem however is that if we strictly adhere to this individualistic idea
of autonomy when we speak of outcome autonomy, then we necessarily require that
nudges must enhance or preserve the autonomy of all choosers. This would
essentially eliminate any non-process autonomy-driven nudges from consideration.
No nudge could possibly claim to respect the outcome autonomy of every single
chooser – even assuming that a choice architect could verify that every chooser’s
authentic motivations would be reflected in the nudged choice outcome.
It is extremely difficult to imagine any situation where all individuals share
an authentic motivation such that a nudge that steers choosers towards a particular
option can be said to respect the outcome autonomy of all choosers. Such a situation
may possibly arise in choice scenarios involving more objective choice outcomes
(e.g. cash) as opposed to subjective qualities like taste, health, frugality where
balanced decision-making judgments are prevalent. The former choice scenarios
involve fungible goods like cash where options are otherwise equal – where it can
be assumed that most, if not all, individuals would rather have more cash than less.
However, the scope for such situations to exist in the real world is narrow.
So, we are left with the problem that for outcome autonomy to have any use,
we need to maximize aggregate outcome autonomy (i.e. by nudging in favour of the
majority’s or plurality’s outcome autonomy), but in the face of the principle that
individual autonomy is generally not, and should not be, subject to aggregate
318
maximization.305 Since autonomy is about the self, it is contradictory to think of
having the motivations of others guide one’s outcome autonomy. In other words,
how can we sacrifice the outcome autonomy of the few in favour of the outcome
autonomy of the many? (Recall here that an ideal aggregate calculation would also
require an understanding of motivation intensities, but we will disregard that issue
because of the indeterminacy problem.)
It may be tempting to argue that because nudges are easily reversible – that
individuals can always choose otherwise – that we are not sacrificing the autonomy
rights of the minority, since they still technically exist.306 This seems to conflate
liberty of choice and autonomy. As we’ve noted before, someone may have liberty
of choice, but still not be free to choose otherwise (recall Locke’s prison). We know
that some nudges are opaque to the chooser and they don’t really have an option to
choose otherwise, even though it may seem that way. So by merely being a nudge
definitionally (i.e. preserving a chooser’s opportunity set) does not mean that choice
architecture can maximize the outcome autonomy of the many at the cost of the
outcome autonomy of the few. Instead, we must look elsewhere for such a
justification.
305 For a brief discussion on maximizing/aggregating rights see: Ernest J Weinrib, “Utilitarianism, Economics, and Legal Theory” (1980) 30:3 Univ Tor Law J 307 at 318. 306 See e.g. Rebonato, supra note 112; Amitai Etzioni, “Humble Decision-Making Theory” (2014) Public Manag Rev 1.
319
The answer may, oddly enough, lie in the outcome qualifier of outcome
autonomy. Outcome autonomy is not participatory autonomy; it is not actual self-
governing, but the hypothetical end result of a self-governing construct. We may be
able to aggregate outcome autonomy because we are simply aggregating
instantiations of the hypothetical results of process autonomy. No choosers’ process
autonomy (their participatory self-government) is negatively affected in such cases.
Outcome autonomy can be subject to aggregate maximization because it is
simply a different order of value (a lexically inferior value) than process autonomy,
the latter of which cannot be subjected to an aggregation argument.307 Process
autonomy ought to be fully respected because it centers on individuals actually
participating in their lives by governing themselves and actively expressing their
autonomy, whereas outcome autonomy does not. Being autonomous is about
experiencing and exercising autonomy, not just about outcomes. It does not make
sense to speak of aggregating the experience or exercise of something. But, in the
case of outcome autonomy, we may be able to aggregate the results of the process.
We want our lives and choices to lead to some end, but living and being
human is what happens in between (something more than Nozick’s mere experience
machine).308 This is why the focus of this thesis has been on respecting individual
307 See e.g. Nozick, supra note 115. 308 Ibid at 42.
320
process autonomy, even if doing so leads to losses in outcome autonomy. Process
autonomy is the nexus of self-governing autonomy, whereas outcome autonomy is
simply a hypothetical derivative of the former. However, in cases where process
autonomy is not affected one way or the other (as in cases of unavoidable choice
architecture), outcome autonomy may have value in setting the ethical boundaries
for nudging. And by aggregating outcome autonomy we are providing for more
realizations of process autonomy, thereby respecting process autonomy to a greater
extent than if we did not nudge on an outcome autonomy basis. Therefore, we can
maximize outcome autonomy, but only as a tool of last resort.
8.8. Conditions on Using Outcome Autonomy as a Tool of Last Resort
Given the foregoing, we can resort to outcome autonomy, and then
specifically to aggregating outcome autonomy, only if a set of pre-conditions are
satisfied. Failing to satisfy the following conditions will render any aggregate
maximization of outcome autonomy unethical and thus the relevant nudges will
also be deemed unethical.
(1) No Decrease or Increase in Process Autonomy. A nudge must not cause any
decrease in the process autonomy of any chooser. In other words, nudges must
preserve or enhance the process autonomy of every chooser. We cannot trade off
321
process autonomy gains of the majority at the expense of process autonomy
losses for the minority. This goes to the nature of process autonomy as reflecting
a person’s self-governing autonomy. In other words, process autonomy is not
subject to aggregation. And recall that any increase in process autonomy
effectively trumps a decrease in outcome autonomy due to its lexical priority.
This means that if a nudge increased the process autonomy of choosers (or
provided them at least with an opportunity to do so), then the nudge would
prima facie count as ethical and resort to outcome autonomy effects would not be
necessary.
(2) Personalized Choice Architecture is not Available. Only if there is no change to
process autonomy, can we refer to outcome autonomy to “break the tie” on
ethicality. In such an event, the preferred mode of nudging would be through
personalized choice architecture, for example through personalized default rules
as futuristically imagined by Sunstein.309 These will rarely be available because
nudges tend to be of wide, general application. However, in the realm of private
nudges (e.g. employees), it may be easier to identify an individual’s authentic
motivations and match those motivations to particular choice outcomes through
personalized choice architecture. If such means are available and practical, they
309 See e.g. Sunstein, "Choosing", supra note 12.
322
must be utilized before we can attempt to resort to a nudge that maximizes
outcome autonomy.
(3) Good Evidence of Authentic Motivations. A choice architect must have good
evidence (not just inferences) of a person’s authentic motivations and what
options best match those motivations. In other words, the epistemic concerns
must be minimal in the particular choice scenario. One cannot hope to maximize
the outcome autonomy of affected choosers without an understanding of what
options have the greatest outcome autonomy and which do not. Both the
affirmative and negative elements are necessary for determining where the
majority of choosers’ motivations reside. If we do not have good evidence of
motivations and which motivations comprise the majority, then we should not
be nudging on an outcome autonomy basis.
(4) Maximization of Outcome Autonomy. Finally, only if all of the foregoing
conditions are satisfied, the nudge can be considered ethical if it maximizes the
outcome autonomy of the majority (not just plurality) of affected choosers,
accounting, if possible, for varying intensities of motivations as well. Note that
the maximization of outcome autonomy is key here; it is an aggregate
calculation. The increase in outcome autonomy of the majority of choosers must
be enhanced by a greater net margin than the minority whose outcome
autonomy may be reduced by the nudge.
323
A further qualification is necessary for the proper functioning of the
maximization condition. Outcome autonomy is maximized when a nudge steers the
majority of affected choosers to select an option that reflects their authentic
motivations. The “affected choosers” qualifier is necessary to remove from the
calculation those choosers whose outcome autonomy is unaffected by the nudge.
This may include those whose motivations are not satisfied by their choices pre- or
post-nudge, as well as those choosers who have no authentic motivation or only
have a motivation to be ruled by their lower-order desires. In those cases, a nudge
would not affect their outcome autonomy.
If the situation were such that there was only a plurality of affected choosers
who shared an authentic motivation reflected in the relevant option, then the nudge
could not proceed on an outcome autonomy basis. This would be so because the
nature of outcome autonomy nudges means that the majority of affect choosers
(together excluded from the plurality) could be moved further from their outcome
autonomous options. In other words, because in our calculation we only count those
choosers whose outcome autonomy is affected by the nudge, and because nudges
can only either enhance or decrease (or cause no change to) outcome autonomy, if a
nudge did not affect a chooser (no change in outcome autonomy) or enhance their
outcome autonomy, then it must naturally decrease their autonomy. This would
324
effectively mean that a majority of affected choosers had their outcome autonomy
decreased, even though a clear plurality’s outcome autonomy may benefit from the
nudge. This is simply a reflection of the heterogeneous nature of individual
motivations.
We can now apply the final fifth condition for the maximization of outcome
autonomy to our choice scenario of the frugal and healthy cafeteria patrons. We’ve
already assumed that conditions one to four were satisfied, though in real life, the
satisfaction of the epistemic condition is doubtful. Recall that 40% of our consumers
had an authentic motivation to be healthy, 20% to be frugal, 10% other and the
remaining 30% had no particular motivation. This represents a bit of a hurdle
because there is no majority apparent, just a plurality.
To maximize outcome autonomy, a nudge must enhance the outcome
autonomy of the majority of affected choosers, meaning that we can remove the 30%
with no authentic motivation. The re-arrangement nudge would not affect those
choosers’ outcome autonomy in the negative or positive, so there is no need to
include them in the aggregation calculation. A majority of affected choosers then
emerges from the remaining 70% – the 40% of all choosers (57% of affected choosers)
who have an authentic motivation to be healthy. Thus, a nudge towards the
healthier options would be considered ethical, but only by last resort to outcome
autonomy.
325
8.9. Conclusion
Outcome autonomy is a poor imitation of process autonomy with many
drawbacks, but it is still part of what it means to be autonomous. In many cases,
outcome autonomy acts as a final check on decision-making to ensure that
individuals with deficient levels of process autonomy are indeed self-governing by
helping to ensure that their authentic motivations result in matching option
selection (choices). All else being equal on the process autonomy side of the
equation, a nudge that steers a chooser to a choice outcome that they would choose
as-if autonomous must be considered ethically superior to one that takes a chooser
further from that as-if autonomy (assuming that we can determine what choice
outcome reflects as-if autonomy).
However, protections are required to use outcome autonomy as a tool of last
resort. Resort to outcome autonomy as an ethical guide for nudging must only occur
in cases where process autonomy is unaffected by a nudge. Choice architects must
also have good, reasonable evidence about what an autonomous hypothetical
construct of a chooser would select. Inferences should not be drawn from the
objective status of particular options. If statistical generalizations are the only data
available and personalized nudging is not a feasible option, then the direction of a
326
nudge may be set by maximizing the outcome autonomy of the majority of affected
choosers, proven as such with reasonable evidence.
327
Chapter 9
Synthesizing the Autonomy Effects of Nudges
9.1. Introduction
Part II of this thesis has shown that nudges can affect a chooser’s overall
autonomy by separately affecting a chooser’s process and outcome autonomy. We
have also seen that when evaluating the ethicality of nudges on autonomy grounds,
the appropriate standard of evaluation is one of bare autonomy preservation – to be
ethical, a nudge ought to preserve the autonomy of choosers. However, since we
have also seen that nudges can have differential effects on process autonomy and
outcome autonomy, we need some way to reconcile potential conflicts between the
two types of autonomy.
Given that we consider process autonomy as being lexically superior to
outcome autonomy (which is simply a hypothetical shadow of the former) our
autonomy preservation standard first applies to process autonomy effects of
nudging. Thus, the initial step in our ethical evaluation of any nudge must be to
examine the process autonomy effects. If a nudge causes a reduction in a chooser’s
process autonomy, then the nudge is unethical with respect to autonomy, as process
autonomy is the heart of our understanding of autonomy as effective and
participatory self-government. For the same reasons, a nudge that causes an increase
328
in process autonomy must be considered ethical with respect to autonomy. This
conclusion holds regardless of the same nudge’s effects on outcome autonomy.
While we cannot trade off process autonomy losses for outcome autonomy gains,
we can trade off any amount/number of outcome autonomy losses for any
amount/number of process autonomy gains because of the lexical priority of the
latter.
There are some nudges that will not affect (i.e. neither increase nor decrease)
a chooser’s process autonomy. As nudges are designed to change individual
choices, and outcome autonomy is a characteristic of choices, nudges that change
choices will often have an impact on a chooser’s outcome autonomy independent of
any process autonomy effects. Since outcome autonomy is still autonomy in some
sense, a nudge, if it is having no effect on process autonomy, must at least preserve
the outcome autonomy of choosers to be considered ethical. But we would only ever
refer to outcome autonomy effects as an evaluation tool of last resort – i.e. once it
has been determined that there are no superseding process autonomy effects.
Finally, a nudge that has no effect on process or outcome autonomy will be
considered ethical because it has no effect on the autonomy of choosers – their
autonomy is technically “preserved.”
329
9.2. Table of Combinatory Autonomy Effects
Figure 8 below is a table of combinatory effects of nudging on a chooser’s
autonomy. This table covers all possible combinations of effects on a chooser’s
process and outcome autonomy, with corresponding examples. The remainder of
this chapter will examine each of the nine combinations individually, reaching a
conclusion as to the ethicality of such nudges as indicated in the table below.
Figure 8 – Combinatory Effects of Nudging on Autonomy
Case #
Process Autonomy
Outcome Autonomy
Ethicality Key Examples
1 Decrease Decrease Unethical Subliminal advertising
2 Decrease No Change Unethical Subliminal advertising;
Anchoring charitable donations
3 Decrease Increase Unethical Graphic cigarette warning labels
4 No Change Decrease Unethical Menu item order
5 No Change No Change Ethical Urinal fly
6 No Change Increase Ethical Framing effect re: survival
statistics
7 Increase Decrease Ethical Social norm messaging re:
savings
8 Increase No Change Ethical Penalty defaults re: tax refunds
9 Increase Increase Ethical Active choosing
330
9.3. Unethical Nudges
We will begin with an examination of nudges that should be considered
unethical because of their effects on a chooser’s autonomy. This class covers four
sets of combinations, which we will examine individually in the order indicated in
Figure 9 below.
Figure 9 – Combinatory Effects of Nudging on Autonomy, Unethical Only
# Process Autonomy
Outcome Autonomy
Ethicality Key Examples
1 Decrease Decrease Unethical Subliminal advertising
2 Decrease No Change Unethical Subliminal advertising;
Anchoring charitable donations
3 Decrease Increase Unethical Graphic cigarette warning labels
4 No Change Decrease Unethical Menu item order
(1) Reduction in Process Autonomy & Reduction in Outcome Autonomy
Any nudge that decreases both a chooser’s process and outcome autonomy
must be considered unethical. This is as clear cut a case of unethicality as can be. No
form of choice architecture can be considered ethical on autonomy grounds if it
operates through a less autonomous choosing process (i.e. motivational and
decisional autonomy) and steers the chooser towards a choice outcome that they
would not have chosen if they were autonomous. Even if we were to disregard the
331
lexical priority of process autonomy and combine the nudge’s effects on the two
types of autonomy on an aggregate basis, the net effect of the nudge is still a
reduction in autonomy.
Because a nudge of this class fails at our first stage of autonomy analysis –
process autonomy – we need not technically even consider outcome autonomy
effects. However, suffice to note that a nudge of this class would also fail at the
outcome autonomy stage because choosers end up with a choice outcome that is less
reflective or further from, their authentic motivations (i.e. what they would select
with full process autonomy). So, not only is a nudge of this type providing choosers
with less of an opportunity for participatory self-government, but it is taking them
further from where they would have hypothetically ended up with the former, as
compared to their pre-nudge state.
Our key example here is subliminal advertising – think of subliminal
messaging that encourages organ donation (unrealistic, but possible). As we saw in
Chapter 6, some forms of subliminal advertising work by covertly altering a
chooser’s higher-order motivations. Because of its manipulative covertness
preventing a chooser from accepting or rejecting the influence on their motivations,
the resulting motivations are considered inauthentic. The donation messaging
encourages a chooser to want to donate – to develop a motivation to donate without
their participation in developing that motivation. As such, this nudge would fail to
332
preserve the motivational autonomy of choosers and thus fails to preserve the
process autonomy of choosers. On that ground alone, the nudge should be rendered
unethical.
However, it gets worse, because the nudge provides choosers with an
inauthentic motivation (and assuming that the chooser is able to express those
motivations at the option evaluation level), the end result will be the selection of a
choice outcome (donating) that reflects an inauthentic motivation. Since it is possible
that an inauthentic motivation (donating) is different from a chooser’s hypothetical
authentic motivation (not to donate), the nudge also fails to preserve a choosers’
outcome autonomy. In sum, nudges with this combination of dual negative
autonomy effects are clearly unethical.
(2) Reduction in Process Autonomy & No Change in Outcome Autonomy
For the same reasons as in combination (1) above, a conclusion of “unethical”
is attached to any nudge that reduces process autonomy but without any effect on
outcome autonomy. Since process autonomy is our primary concern, the initial
reduction in process autonomy is the determinative factor rendering a nudge of this
class unethical.
The subliminal advertising example in (1) also applies in this case, however
with one small difference – the ultimate choice outcome that the chooser is nudged
333
towards is equally reflective of that chooser’s authentic motivation as whatever
choice outcome they were selecting on a pre-nudge basis. Consider a subliminal
nudge that encourages smoking a particular brand of cigarettes, instead of smoking
writ-large. If a chooser’s authentic hypothetical motivation is to be the type of
person that smokes, then simply nudging them towards a different brand should
have little effect on that chooser’s outcome autonomy. However, we must
emphasize again that resort to outcome autonomy would not be necessary since the
nudge would be rendered unethical solely on process autonomy grounds because it
works on motivations covertly.
Another example may help here. One study showed that when individuals
are asked to contribute to charity, the introduction of an anchoring reference (i.e. a
specific amount of money) increases the contribution amount from those that
contribute (while preserving their opportunity sets).310 This kind of nudge decreases
a chooser’s process autonomy because the anchor triggers biased System 1-reacting
(a loss of decisional autonomy). However, this anchoring nudge does not really
change the outcome autonomy of choice outcomes – a $20 donation versus a $4
donation cannot really be said to reflect a different hypothetical authentic
motivation. The motivation is the same – to donate to charity, which is why it is
310 Cynthia Fraser, Robert E Hite & Paul L Sauer, “Increasing Contributions in Solicitation Campaigns: The Use of Large and Small Anchorpoints” (1988) 15:2 J Consum Res 284.
334
important that the nudge only alters the size of the contribution and does not cause
more people to contribute. Hence, the nudge reduces process autonomy but has no
effect on outcome autonomy.
(3) Reduction in Process Autonomy & Increase in Outcome Autonomy
A more complicated situation arises when a nudge causes a reduction in
process autonomy, but that reduction could arguably be “offset” by an increase in
outcome autonomy. It is for these kind of nudges where the lexical priority of
process autonomy comes into play. If we are at all interested in nudges respecting
an individual’s autonomy – participatory self-government – then we must reject any
nudge that causes a decrease in process autonomy, no matter the effect on outcome
autonomy.
Most nudges in this category will represent efforts at counter-biasing. As we
saw in Chapter 6, counter-biasing occurs when a chooser pre-nudge is subject to
behavioural biases that may be causing the chooser to select an option that is not
consistent with their authentic motivations (i.e. contrary to their outcome
autonomy). The choice architect then introduces a nudge that engages a new bias or
heuristic designed to overpower or neutralize the existing bias and cause the
chooser to select a different option, hopefully one with greater outcome autonomy.
335
Consider our familiar example of graphic warning labels on cigarette
packages. As we have already seen, they are designed to neutralize individuals’
optimism bias surrounding health risks associated with smoking. The pre-existing
and expressed optimism bias is not removed by the nudge, it is simply
overwhelmed by the salience and emotional factors encouraged by the graphic
labels. A chooser who is biased pre-nudge arguably has their decisional autonomy
enhanced because they are better able to accurately evaluate their options. But for
those choosers who do not suffer from optimism bias pre-nudge (or suffer to a less
extent than compensated for by the graphic labels), the graphic labels result in those
choosers selecting with less decisional autonomy, and thus the nudge would fail to
preserve their process autonomy. As we saw in Chapter 6, since we cannot aggregate
the process autonomy effects of different choosers (because doing so runs counter to
the very principles of autonomy), any decrease in process autonomy experienced by
any chooser renders the nudge unethical.
However, we could make the argument that many of those choosers affected
by the graphic labels nudge have an authentic motivation to not smoke.311 Thus,
because the nudge steers those smokers to not smoke, the nudge may be said to
increase their outcome autonomy. When compared to the pre-nudge choice outcome
311 Evidence from surveys of smokers (i.e. how many want to quit) support this argument, however, as we have noted, we need to be careful with drawing inferences of authentic motivations from survey responses.
336
of smoking, not smoking seems to better reflect those choosers’ hypothetical
authentic motivations. But as we have noted many times, process autonomy losses
cannot be traded off for outcome autonomy gains. Nudges that respect or promote a
chooser’s process autonomy attempt to provide a chooser with an opportunity for
autonomous choosing, such that we can say that the person is in fact self-governing.
In terms of respecting a chooser’s autonomy, it is not enough to simply steer
choosers towards a choice outcome that the choice architect believes the chooser
would select if they were truly self-governing. Self-governance means that the
individual actually governs themselves. The experience or means of being self-
governing are what gives autonomy its value. We may also have concerns over the
epistemic question of whether a choice architect can accurately ascertain what that
chooser would choose if they were fully autonomous. Overall then, we can never
trade off gains in outcome autonomy (if the gains can even be accurately
ascertained) for losses in process autonomy. Hence, the idea of lexical priority of
process autonomy (i.e. any effect on process autonomy is determinative of
ethicality) and why nudges of this nature remain unethical.
(4) No Change in Process Autonomy & Decrease in Outcome Autonomy
This category of nudges covers those that have no effect on process
autonomy, but nonetheless steer choosers toward choices that are arguably not
337
reflective (or at least less reflective) of their outcome autonomy. For example, a
nudge of this type may utilize existing biases to redirect choosers to a different
choice outcome (i.e. re-biasing). A bias that is simply re-directed toward a different
choice outcome is unaffected for process autonomy purposes. To count as a
reduction in outcome autonomy, the different choice outcome the chooser selects
must be further from what they would hypothetically choose under ideal process
autonomy conditions than their pre-nudge selection.
Let’s consider for our key example a nudge that re-arranges a restaurant
menu. Behavioural insights show that individuals are predisposed to purchase more
of what is listed earlier in the menu.312 We will assume here that our chooser has an
authentic motivation to be as healthy as possible. Our imaginary restaurant serves a
variety of food, as most do. Some dishes are obviously unhealthy (e.g. fries) while
others will be obviously healthy (e.g. mixed berries) while many others will lie
somewhere in between the two extremes – for example, a Caesar salad or a
sandwich. For illustration purposes we’ll stipulate that pre-nudge, our chooser was
selecting a sandwich for lunch, as the sandwich listing was at the front of the menu,
and we know that individuals are biased to select more of what they encounter first.
312 Eran Dayan & Maya Bar-Hillel, “Nudge to nobesity II: Menu positions influence food orders” (2011) 6:4 Judgm Decis Mak 333.
338
Our restaurant manager can rearrange the order of the menu, utilizing the
fact that individuals will choose more of whatever comes first. Such a nudge causes
no reduction in process autonomy, since there is no alteration to choosers’
deliberation and the relevant bias is already at play pre-nudge. There would also not
be any unbiased choosers made biased as a result of the nudge (as in the case of
counter-biasing). Our chooser’s motivation to be healthy is also unchanged by the
ordering of the menu, since the nudge doesn’t change motivations, nor would it
count as a covert external influence. Thus, in both pre-nudge and post-nudge choice
environments, the chooser’s process autonomy remains deficient, but unchanged.
Our imaginary restaurant manager may not share or care about our diner’s
authentic motivation for healthy living and may instead want to maximize their
profits. Since the highest margin foods tend to be the fried foods, the manager as
choice architect rearranges the listings of foods such that the fries are listed first
instead of the sandwiches as they were pre-nudge. Our health-minded diner
succumbs to the temptation and purchases fries instead of the sandwich they were
selecting pre-nudge. Thus, post-nudge in this case, the choice outcome they select is
even further from their authentic motivation (to be healthy) than the pre-nudge
sandwich. Therefore, we can say that the diner suffered a reduction in their outcome
autonomy – they end up worse off than they were pre-nudge since they are further
from what the would have selected if they had full process autonomy. These kind of
339
nudges – that have no effect on a chooser’s process autonomy but cause a reduction
in outcome autonomy – must also be considered unethical.
9.4. Ethical Nudges
This section covers the remaining sets of combinations – the ethical nudges –
which we will examine individually in the order indicated in Figure 10 below.
Figure 10 – Combinatory Effects of Nudging on Autonomy, Ethical Only
# Process Autonomy
Outcome Autonomy
Ethicality Key Examples
5 No Change No Change Ethical Urinal fly
6 No Change Increase Ethical Framing effects re: survival
statistics
7 Increase Decrease Ethical Social norm messaging re: savings
8 Increase No Change Ethical Penalty defaults re: tax refunds
9 Increase Increase Ethical Active choosing
(5) No Change in Process Autonomy & No Change in Outcome Autonomy
Little needs to be said about nudges that have no effect on process or
outcome autonomy. They are barely ethical, since they permit choice architects to
knowingly leave choosers with deficient levels of process or outcome autonomy.
However, since our standard of evaluation is one of autonomy preservation and the
autonomy of choosers is unchanged, a nudge of this nature meets that standard.
340
The best example here may be a classic one mentioned by Sunstein and
Thaler – the addition of a sticker of a fly (the insect) to urinals in the men’s
washroom at Schiphol Airport in Amsterdam.313 The presence of the sticker reduced
the unpleasant amount of urine spillage. The nudge had no real effect on process
autonomy – it did not encourage deliberation or System 2 “thinking” – it simply
gave the men something to aim at in an equally mindless fashion. Nor did the
nudge have much effect on outcome autonomy – men still urinated which was
presumably their motivation for going in the washroom in the first place.
(6) No Change in Process Autonomy & Increase in Outcome Autonomy
This category includes those nudges that have no effect on process autonomy,
but instead increase the outcome autonomy of choosers. Because of the special
status of outcome autonomy as a hypothetical derivative of process autonomy (and
therefore a weak version of autonomy), we must be careful when reaching a
conclusion of ethicality solely based on gains to outcome autonomy. This hesitation
stems from the problems we saw with using outcome autonomy as an evaluative
concept in Chapter 8 – it should only be a tool of last resort. Recall that in order for
us to be able to maximize outcome autonomy on an aggregate basis (as any possible
nudge would require), there can be no lexically prior effects on process autonomy.
313 Thaler & Sunstein, "Nudge", supra note 1 at 4.
341
There are two parts to any analysis of a nudge that resorts to outcome
autonomy for determination of ethicality. The first part looks at whether there are
any process autonomy effects. If, and only if, there are not, can we turn to the
outcome autonomy effects of the nudge (subject to conditions for aggregate
maximization such as knowledge, clear majority of affected choosers, etc.). If a
reduction in process autonomy is found, then the nudge in question is unethical and
we need not proceed to examine the effects on outcome autonomy. Likewise, but on
the ethical side of the ledger, for nudges that increase the process autonomy of
choosers. If a nudge has no actual effect on process autonomy, only then can we
refer to outcome autonomy effects.
Recall from Chapter 8 that in order to find that a nudge has increased the
outcome autonomy of choosers several conditions must be met. The choice architect
must have good, reliable evidence of the authentic motivations of affected choosers.
In order to say that a nudge has steered most (if not all) affected choosers towards
choices that better reflect their authentic motivations when compared to their pre-
nudge choice outcomes, we must know the content of their motivations. Only if a
choice architect has that knowledge can they determine whether a nudge maximizes
the outcome autonomy of the majority of affected choosers.
We most often have to resort to outcome autonomy determinations for our
ethical conclusion in cases of re-biasing, where choosers’ existing biases/modes of
342
cognition are re-directed to different choice outcomes. If those new choice outcomes
align with choosers’ authentic motivations, then we can say that the nudge increases
their outcome autonomy.
Our key example of framing of survival statistics may help illustrate this class
of nudges. Behavioural insights have repeatedly shown that individuals tend to be
subject to framing effects – that is, how logically identical information/choices are
presented (framed) affects the choice outcomes that choosers select. In the case of
cancer patients, O’Connor’s study showed that patients who are presented with a
positive frame (probability of survival is presented) are more apt to choose toxic
treatment over quality of life improvements than those patients presented with a
negative frame (probability of dying is provided).314 The framing effect is
particularly prevalent at low survival probabilities, as O’Connor theorized that the
positive frame gives hope while the negative frame suggest giving up hope (the
word “survive” is not even present).315
The above is but one example of how framing works. Let’s evaluate a nudge
of this nature (flipping from positive to negative frame or vice-versa) for autonomy
ethics. We’ll stipulate that our nudge flips the style of probability presentation from
a negative frame pre-nudge to a positive frame post-nudge. The first thing to note is
314 Anette M O’Connor, “Effects of framing and level of probability on patients’ preferences for cancer chemotherapy” (1989) 42:2 J Clin Epidemiol 119. 315 Ibid at 124.
343
that the nudge seems to have no positive or negative effect on patients’ process
autonomy. Choosers’ process autonomy post-nudge is just as deficient as it was pre-
nudge and they are being informed with logically identical information in both
cases.
The next step is to refer to outcome autonomy. In order to use outcome
autonomy to find this kind of nudge as ethical, the choice architect must have
knowledge that the majority of choosers have an authentic motivation to live as long
as possible and the choice architect has reliable access to evidence of the sort. If so,
then we can say that the treatment option is the choice outcome that has the highest
outcome autonomy for the majority of choosers. Thus, it seems that the nudge has
no effect on process autonomy and increases outcome autonomy (on an aggregate
basis).
One important point to note about this case is that process autonomy
improvements are possible. That is, a nudge could instead be designed so as to
enhance the process autonomy of choosers. Instead of flipping the frame to enhance
outcome autonomy, the nudge can de-bias the chooser by reducing the framing
effect. This is accomplished by presenting what O’Connor calls a “mixed frame” –
messaging that included both the probability of surviving and dying. Switching
from a negative frame pre-nudge to a mixed frame post-nudge could be said to have
process autonomy enhancements.
344
However, because our standard is one of mere autonomy preservation, we
need not consider all possible enhancements to process autonomy. There is no
obligation on choice architects to improve chooser autonomy; only to not reduce
autonomy by their intervention. This also reflects our purpose of developing a
framework that can be used to evaluate nudges as they are, not all possible nudges.
A choice architect can select any ethical nudge; they do not have to select the best
possible ethical nudge.
In sum, if nudges have no effect on process autonomy, the next best thing
(ethically) is to provide improvements to outcome autonomy. All things considered,
it is better for a chooser’s autonomy for that chooser to be steered towards selecting
an option that better reflects their authentic motivations (when compared to their
pre-nudge choice).
(7) Increase in Process Autonomy & Reduction in Outcome Autonomy
There are a small minority of nudges that will have opposite effects on
process and outcome autonomy. In combination class (3) we saw nudges that reduce
process autonomy while ostensibly increasing outcome autonomy. In this section we
are interested in the opposite – nudges that increase process autonomy but that also
cause a reduction in outcome autonomy. We’ll assume here that the epistemic and
345
aggregation problems have been overcome and our determination of outcome
autonomy is definitive.
As with all our other categories, any actual effect on a chooser’s process
autonomy is determinative of the ethicality of that nudge. Since there is an increase
in process autonomy here, the nudge must be classified as ethical. As we have seen
repeatedly, we can trade off losses to outcome autonomy for gains in process
autonomy because the latter are actual improvements in autonomy as participatory
self-government. Process autonomy is about providing individuals with an
opportunity to be self-governing and a nudge of this nature provides just that.
We do not subject nudges of this class to an aggregation calculus with
potentially offsetting or overwhelming reductions in outcome autonomy because
outcome autonomy is a hypothetical derivative of process autonomy. It would be a
perverse result if the original (process autonomy) were rendered subordinate to the
derivative (outcome autonomy).
Most nudges that provide choosers with greater process autonomy will also
lead to an increase in outcome autonomy since choosers will be more likely to select
choice outcomes reflective of their authentic motivations. However, sometimes
nudges can cause an increase in process autonomy, but backfire in results.
Our key example here is of a nudge that provides relevant information to
choosers, thereby enhancing their decisional autonomy, but that information
346
actually leads them to select a choice outcome with less outcome autonomy than
pre-nudge. We’ll use as an example social norm messaging in the retirement savings
context. Beshears et al. conducted a field study to measure how providing peer
information would affect employee savings participation.316 Non-participants or low
savers in a company’s 401(k) plan were provided with letters encouraging them to
sign-up, with the trial group letter including information about peer enrolment and
savings habits.
The goal of the nudge was of course to increase savings participation. A
libertarian paternalist could argue that choice outcomes of joining the plan or
increasing contributions have high outcome autonomy – it is what the choosers
would select if they had full decisional autonomy. For the purposes of illustration,
we’ll assume it to be correct – that the majority of choosers have as one of their
authentic motivations the desire to save, or save more.
As we saw in Chapter 6, this information is relevant to the chooser,
potentially assisting them in selecting options in line with their authentic
motivations. Providing peer information as this nudge does then seems to be an
increase in process autonomy at the decisional autonomy level. If a chooser’s
authentic motivation is not only to save, but also to “keep up with the Joneses” then
the provision of this information is relevant to their decision-making and will help
316 Beshears et al, supra note 263.
347
them select the choice that best reflects those motivations. So the nudge could be
said to increase process autonomy (depending on the circumstances and
peculiarities of the choosers).
But a nudge of this nature may actually cause a decrease in outcome
autonomy, if we evaluate the choice outcomes as ideally reflecting either the
motivation to save or the motivation to keep up with their co-workers. Beshears et
al.’s study showed that for a subset of workers (unionized non-participants in the
plan), exposure to the peer information caused a reduction in plan enrolment.317 This
means a reduction in saving, arguably pushing those choosers further from realizing
their authentic motivations to save more or keep up with their peers. The cause of
these unexpected negative results (“boomerang effects”) is unknown, particularly
the locus in unionized non-participants.318 The authors entertain several hypotheses,
the best being that the non-participants were de-motivated by the provision of peer
information.
Regardless of the cause, the foregoing is an excellent example of a nudge that
arguably causes an increase in the process autonomy of choosers but that also causes
a decrease in outcome autonomy. However, since the process autonomy effects are
317 Ibid. 318 Ibid.
348
controlling, nudges of this class must still be categorized as autonomy preserving
and thus, ethical.
(8) Increase in Process Autonomy & No Change in Outcome Autonomy
Some nudges will be ethical because they cause an increase in process
autonomy but do not affect outcome autonomy. These nudges will usually be ones
that de-bias, provide information, or provide opportunity for deliberation so that a
chooser’s process autonomy is enhanced, but the greater process autonomy is not
enough to change behaviour.
Our key example here is of ineffective penalty defaults. A 2001 study by
Bronchetti et al. studied attempts to increase savings by low-income tax filers.319 The
study examined altering the default form of tax refunds. A set of low-income tax
filers were defaulted into automatically receiving some portion of their tax refund in
US Savings Bonds (opt-out), whereas pre-nudge they could obviously use their cash
refund on anything they wish, including Savings Bonds. The study found, “the opt-
out default had no impact on savings behaviour.”320
Defaults are normally considered a form of re-biasing, meaning that they will
typically have no effect on a chooser’s process autonomy. Whatever biases are
319 Erin Todd Bronchetti et al, "When a nudge isn’t enough: defaults and saving among low-income tax filers", (2011) NBER Working Paper Series 16887. 320 Ibid.
349
leading individuals to select the default pre-nudge (inertia, loss aversion) remain
post-nudge (since the nudge only affects the content of the default) and so the nudge
has not affected a chooser’s process autonomy. However, as we saw briefly in
Chapter 6, some defaults can change the options so fundamentally that a chooser’s
deliberative, System 2 is activated in response. Penalty defaults (normally used in
contract law) are one such default – the default option is viewed so negatively by the
chooser that they are essentially forced into an active choosing scenario. Thus, a
nudge that switches the content of a default rule to a penalty default may actually
increase a chooser’s process autonomy by increasing their deliberation.
The foregoing study seemed to have exactly that effect. The new default of
getting US Savings Bonds instead of cash was viewed negatively by choosers
because the choosers already had spending plans for the cash.321 As such, they opted
out at a significant enough rate to render the nudge ineffective at changing
behaviour (i.e. changing choice outcomes). Since choice outcomes were not changed,
the nudge had no effect on outcome autonomy.
In sum, nudges can provide choosers with a greater post-nudge opportunity
for process autonomy and yet still be ineffective at changing choices and thus
outcome autonomy. However, since the chooser is still better off with more process
321 Ibid.
350
autonomy, a nudge of this type, though ineffective on outcome autonomy grounds,
is still considered ethical (and, in fact, effective on autonomy grounds).
(9) Increase in Process Autonomy & Increase in Outcome Autonomy
The final class of combinatory effects on autonomy include the most ethical
nudges. These nudges increase chooser’s process autonomy and those effects filter
down to a chooser’s selection of choice outcomes with greater post-nudge outcome
autonomy. Needless to say these nudges are one of the preferred modes of nudging
– the overall increase in autonomy is unquestionable.
However, some nudges in the preceding two classes may actually be more
ethical on autonomy grounds, because our ethical analysis is first and foremost
determined by process autonomy effects. So while any nudge that increases process
autonomy is considered ethical, those that increase process autonomy to the greatest
extent are considered the most ethical.
Regardless of the magnitude of process autonomy improvements, nudges in
this class are clearly ethical. One such example would be effective active choosing.
We saw in Chapter 6 that, instead of using plain defaults, a nudge can provide
choosers with the same choices (respecting opportunity set preservation) in a
manner that activates a chooser’s deliberative, System 2.
351
Nudges of this nature are particularly applicable to current default-based
systems. Instead of asking choosers to select one option while the other option(s)
results if they do nothing, an active choosing nudge would require choosers to
consider all options without one being the default, thereby avoiding the “sticky”
behavioural issues (i.e. lack of process autonomy) associated with defaults. Ideally,
such a nudge would also include an option “to choose not to choose.”
Depending on the particular scenario and the strength of our epistemic
evidence regarding choosers’ authentic motivations, a nudge of this nature may also
cause an increase in outcome autonomy. We can assume that in most cases (but not
necessarily all) a chooser who deliberates more on their choices will select the option
most in line with their authentic motivations (i.e. with high outcome autonomy). In
the pre-nudge default world, some choosers’ option selections had high outcome
autonomy and some had low outcome autonomy, depending on the matching of
motivations to outcomes. The same may be true post-active choosing nudge, but
hopefully to a lesser extent since we would be able to say that all choosers turned
their minds to the choice, even if slightly. That is a greater improvement than a
default system where many choosers may not be aware of a choice or may not think
about it – there is a higher likelihood of choosers selecting options that align with
their authentic motivations
352
Once again though, while outcome autonomy improvements are preferred
from an overall autonomy perspective, process autonomy improvements (or
reductions) are what really count for our ethical analysis. Outcome autonomy
benefits are beneficial side-effects at best because they flow from, and because of,
process autonomy. In sum, any nudge that increases the process autonomy of
choosers is ethical.
9.5. Conclusion
This chapter has shown that nudges can have nine sets of combinatory effects
on process autonomy and outcome autonomy. Nudges can be either ethical or
unethical on autonomy grounds. Where a nudge affects a chooser’s process
autonomy, the nature of that effect (positive/negative) is determinative. Nudges that
preserve or enhance the process autonomy of choosers are ethical. Nudges that
decrease the process autonomy of choosers are unethical. In both cases, effects on
outcome autonomy are non-determinative for our ethical conclusions.
Outcome autonomy is only to be used as an evaluation tool of last resort
because it is a derivative and inferior concept vis-à-vis our understanding of
autonomy as participatory self-government. Outcome autonomy effects are
determinative in two instances. First, where there is no effect on process autonomy
and a reduction in outcome autonomy, such a nudge is classified as unethical.
353
Second, where there is no effect on process autonomy, an increase in outcome
autonomy will render a nudge ethical.
354
Chapter 10
Conclusion
We have seen that nudges are experiencing rapid growth and thought to hold
tremendous potential as regulatory tools. Scarcely a week goes by when a new
nudge isn’t trumpeted as a possible solution to some social ill. Despite the potential,
much work needs to be done on teasing out the nuance of nudge so that we can
more properly evaluate proposed nudges, both ethically and functionally.
Hopefully, this thesis has accomplished some of this work, equipping
policymakers and regulators with a few of the tools necessary to knowledgeably
work with nudges. By coupling a liberty of choice identification screen with an
autonomy evaluation framework, we can narrow the scope of nudges that
policymakers and regulators consider.
In Part I we presented an original framework for the identification of nudges
using a liberty of choice constraint that turns on the preservation of opportunity
sets. This reflects the political promise of nudging as a “soft” form of behaviour
change, distinct from “harder” methods such as command-and-control regulations
and prohibitions.
However, just because an intervention qualifies definitionally as a nudge
does not mean that such nudging should be practiced. Instead, as we saw in Part II,
355
nudges can be subject to an additional form of screening – namely, ethical
evaluation using autonomy preservation as our guide. Autonomy as participatory
self-government requires, at a minimum, that nudges preserve choosers’ process
autonomy as it existed pre-nudge. The outcome autonomy reasoning implicit in
libertarian paternalism should only be used an ethical tool of last resort.
Another key conclusion of this thesis is that nudges cannot be evaluated
broadly as a homogeneous class. Each nudge is different and highly fact-specific,
and this thesis, while pointing to some of the possible differences, does not address
each possible variant. For example, in each instance, we need to make
determinations of opportunity set preservation, level of cognition and authentic
motivation, etc. The pre-nudge world is highly variable – the same nudge may
reduce autonomy in some circumstances and have no effect in others. The
frameworks herein merely provide the tools to undertake the identification and
ethical evaluation analysis; the analysis within this thesis is simply illustrative (i.e.
we have not considered all possible nudges). Thus we should consider the
frameworks developed in this thesis – for identification and evaluation of nudges –
to be adaptive.
This thesis also does not take a position on whether nudges are an ethically
superior form of intervention to non-nudge forms of choice architecture. Indeed, it is
a separate question as to whether some interventions that limit choice (and,
356
therefore, are not nudges, strictly speaking) may also enhance autonomy. But within
the realm of nudging, we can say whether some are ethical or not on autonomy
grounds. On autonomy grounds, nudges should enhance the autonomy of choosers,
but at a minimum, they must not reduce choosers’ autonomy. In other words, it is
wrong for a nudge to reduce the autonomy of choosers, but not wrong for a nudge
to fail to enhance autonomy.
The original contributions of this thesis also open doors for future work.
More work needs to be done on understanding how nudges affect psychological
opportunity costs of choosers (e.g. how are we to classify social stigma?). And for
the proper functioning of our autonomy evaluation framework, we need a better
grasp of the epistemic question surrounding individual motivations – how are we to
confidently judge the locus and authenticity of a person’s higher-order motivations.
In this thesis, we have considered an evaluative framework using autonomy;
we do not make the case that autonomy ought to be the overriding consideration.
Thus, another question is whether there are ethical considerations other than
autonomy that could, potentially, “save” a nudge found to be unethical on
autonomy grounds.
Finally, several questions arise for policymakers. First, there is the question
whether individual autonomy can be traded off for aggregate social welfare – i.e.
whether nudges that result in objective welfare gains but autonomy losses should be
357
in the policymaker’s regulatory toolbox. Second, there is the question whether our
autonomy preservation framework should be applied in developing a more robust
consumer protection framework for regulating private use and exploitation of
behavioural insights. While this thesis has not addressed these questions
specifically, hopefully it has provided the important first steps for informing this
sort of policy analysis.
358
Bibliography
Abadie, Alberto & Sebastien Gay. “The impact of presumed consent legislation on
cadaveric organ donation: A cross-country study” (2006) 25 J Heal Ecnomics
599.
Almashat, Sammy et al. “Framing effect debiasing in medical decision making”
(2008) 71:1 Patient Educ Couns 102.
Ariely, Dan. Predictably Irrational: The Hidden Forces That Shape Our Decision (New
York: Harper Perennial, 2010).
Armstrong, Katrina et al. “Effect of Framing as Gain versus Loss on Understanding
and Hypothetical Treatment Choices: Survival and Morality Curves” (2002) 22
Med Decis Mak 76.
Arnold, Chris. “Why Is It So Hard to Save? U.K. Shows It Doesn’t Have To Be”, NPR
(23 October 2015) online: <http://www.npr.org/2015/10/23/445337261/why-is-it-
so-hard-to-save-u-k-shows-it-doesnt-have-to-be?>.
Barton, Adrien & Till Grüne-Yanoff. “From Libertarian Paternalism to Nudging—
and Beyond” (2015) 6 Rev Philos Psychol 341.
Baumeister, Roy F. “Ego Depletion and Self-Regulation Failure: A Resource Model
of Self-Control” (2003) 27:2 Alcohol Clin Exp Res 281.
Baumeister, Roy F. “Yielding to Temptation: Self-Control Failure, Impulsive
359
Purchasing, and Consumer Behavior” (2002) 28:4 J Consum Res 670.
The Behavioural Insights Team. Applying Behavioural Insights to Organ Donation:
preliminary results from a randomised controlled trial (December 2013) online:
<https://www.gov.uk/government/publications?departments%5B%5D=behavi
oural-insights-team>.
The Behavioural Insights Team, online: <http://www.behaviouralinsights.co.uk/>.
Ben-Shahar, Omri & John A E Pottow. “On the Stickiness of Default Rules” (2006) 33
Florida State Univ Law Rev 651.
Bennhold, Katrin. “Britain’s Ministry of Nudges”, The New York Times (7 December
2013), online:
<http://www.nytimes.com/2013/12/08/business/international/britains-ministry-
of-nudges.html>.
Besharov, Gregory. “Second-Best Considerations in Correcting Cognitive Biases”
(2004) 71:1 South Econ J 12.
Beshears, John et al. “The Effect of Providing Peer Information on Retirement
Savings Decisions” (2015) 70:3 J Finance 1161.
Binder, Martin & Leonhard K Lades. “Autonomy-Enhancing Paternalism” (2015)
68:1 KYKLOS 3.
Blumenthal-Barby, J S & Hadley Burroughs. “Seeking Better Health Care Outcomes:
The Ethics of Using the ‘Nudge’” (2012) 12:2 Am J Bioeth 1.
360
Bovens, Luc. “The Ethics of Nudge” in Till Grune-Yanoff & SO Hansson, eds,
Preference Change: Approaches from Philosophy, Economics and Psychology (New
York: Springer, 2008).
Brighton, Henry & Gerd Gigerenzer. “Are rational actor models ‘rational’ outside
small worlds?” in S Okasha & K Binmore, eds, Evolution and Rationality:
Decisions, Cooperation, and Strategic Behaviour (Cambridge: Cambridge
University Press, 2012) 84.
Bronchetti, Erin Todd et al. “When a nudge isn’t enough: defaults and saving among
low-income tax filers”, (2011) NBER Working Paper Series 16887.
Camerer, Colin, et al. “Regulation for Conservatives: Behavioral Economics and the
Case for ‘Asymmetric Paternalism’” (2003) 151:3 Univ PA Law Rev 1211.
Capron, Alexander Morgan. “Are There Public Health Domains in ‘Domain-
Specific’ Health Nudging?” (2015) 15:10 Am J Bioeth 47.
Choi, Candice. “Fewer McDonald’ s customers ordering soda with Happy Meals”,
(25 June 2015), online: Consumerist.com
<http://consumerist.com/2015/06/25/mcdonalds-fewer-happy-meals-are-being-
ordered-with-sodas/>.
Christman, John, ed. The Inner Citadel: Essays on Individual Autonomy (New York:
Oxford University Press, 1989).
Christman, John. “Introduction” in John Christman, ed, The Inner Citadel: Essays on
361
Individual Autonomy (New York: Oxford University Press, 1989).
Conly, Sarah. “Three Cheers for the Nanny State”, The New York Times (24 March
2013) online: <http://www.nytimes.com/2013/03/25/opinion/three-cheers-for-
the-nanny-state.html>.
Dayan, Eran & Maya Bar-Hillel. “Nudge to nobesity II: Menu positions influence
food orders” (2011) 6:4 Judgm Decis Mak 333.
Department of Health. “Driving up organ donations”, (2011) online: GOVUK News
Stories <https://www.gov.uk/government/news/driving-up-organ-donations>.
Druckman, James N. “Using Credible Advice to Overcome Framing Effects” (2001)
17:1 J Law, Econ Organ 62.
Dworkin, Gerald. “Autonomy and Behavior Control” (1976) 6:1 Hastings Cent Rep
23.
Dworkin, Gerald. “Lying and Nudging” (2013) 39:8 J Med Ethics 496.
Dworkin, Gerald. The Theory and Practice of Autonomy (Cambridge: Cambridge
University Press, 1988).
Etzioni, Amitai. “Humble Decision-Making Theory” (2014) Public Manag Rev 1.
Felsen, Gidon & Peter B Reiner. “How the Neuroscience of Decision Making Informs
Our Conception of Autonomy” (2011) 2:3 AJOB Neurosci 3.
Fischer, Mira & Sebastian Lotz. "Is Soft Paternalism Ethically Legitimate? – The
Relevance of Psychological Processes for the Assessment of Nudge-Based
362
Policies" (2014) No 05-02, Cologne Graduate School Working Paper Series,
Cologne Graduate School in Management, Economics and Social Sciences,
online: <http://EconPapers.repec.org/RePEc:cgr:cgsser:05-02>.
Frankfurt, Harry G. “Freedom of the Will and the Concept of a Person” in John
Christman, ed, The Inner Citadel: Essays on Individual Autonomy (New York:
Oxford University Press, 1989).
Fraser, Cynthia, Robert E Hite & Paul L Sauer. “Increasing Contributions in
Solicitation Campaigns: The Use of Large and Small Anchorpoints” (1988) 15:2
J Consum Res 284.
Frey, Erin & Todd Rogers. “Persistence: How Treatment Effects Persist After
Interventions Stop” (2014) 1:1 Policy Insights from Behav Brain Sci 172.
Gerber, Alan S & Todd Rogers. “Descriptive Social Norms and Motivation to Vote:
Everybody’s Voting and so Should You” (2009) 71:1 J Polit 178.
Gigerenzer, Gerd & Henry Brighton. “Homo Heuristicus: Why Biased Minds Make
Better Inferences” (2009) 1 Top Cogn Sci 107.
Gigerenzer, Gerd & Thomas Sturm. “How (far) can rationality be naturalized?”
(2012) 187 Synthese 243.
Gill, Nick & Matthew Gill. “The limits to libertarian paternalism: two new critiques
and seven best-practice imperatives” (2012) 30 Environ Plan C Gov Policy 924.
Gino, Francesca. “Why the U.S. Government Is Embracing Behavioral Science”, (18
363
September 2015) online: Havard Business Review <https://hbr.org/2015/09/why-
the-u-s-government-is-embracing-behavioral-science>.
Glaeser, Edward L. “Paternalism and Psychology” (2006) 29:2 Regulation 32.
Glod, William. “How Nudges Often Fail to Treat People According to Their Own
Preferences” (2015) 41:4 Soc Theory Pract 599.
Godin, Gaston et al. “Asking Questions Changes Behavior: Mere Measurement
Effects on Frequency of Blood Donation” (2008) 27:2 Heal Pschology 179.
Grynbaum, Micheal M. “Health Panel Approves Restriction on Sale of Large Sugary
Drinks”, The New York Times (13 September 2012) online:
<http://www.nytimes.com/2012/09/14/nyregion/health-board-approves-
bloombergs-soda-ban.html?_r=0>.
Haan, Thomas de & Jona Linde. “Good Nudge Lullaby” (2012) [unpublished].
Hadar, Liat & Sanjay Sood. “When Knowledge is Demotivating: Subjective
Knowledge and Choice Overload” (2014) 25:9 Psychol Sci 1739.
Hagger, Martin S et al. “Ego depletion and the Strength Model of Self-Control: A
Meta-Analysis” (2010) 136:4 Psychol Bull 495.
Hagman, William et al. “Public Views on Policies Involving Nudges” (2015) 6 Rev
Philos Psychol 439.
Hammerstein, Peter. “Evolutionary Adaptation and the Economic Concept of
Bounded Rationality -- A Dialogue” in Gerd Gigerenzer & Reinhard Selten,
364
eds, Bounded Rationality (Cambridge: The MIT Press, 2001).
Handelman, Sapir. Thought Manipulation: The Use and Abuse of Psychological Trickery
(Santa Barbara: ABC-CLIO, 2009).
Hanks, Andrew S, David R Just & Brian Wansink. “Smarter Lunchrooms Can
Address New School Lunchroom Guidelines and Childhood Obesity” (2013)
162:4 J Pediatr 867.
Hanna, Jason. “Libertarian Paternalism, Manipulation, and the Shaping of
Preferences” (2015) 41:4 Soc Theory Pract 618.
Hansen, Pelle Guldborg. “Should We Be ‘Nudging’ for Cadaveric Organ
Donations?” (2012) 12:2 Am J Bioeth 46.
Hansen, Pelle Guldborg & Andreas Maaløe Jespersen. “Nudge and the
Manipulation of Choice: A Framework for the Responsible Use of the Nudge
Approach” (2013) 4 Eur J Risk Regul 3.
Hausman, Daniel M & Brynn Welch. “Debate: To Nudge or Not to Nudge” (2010)
18:1 J Polit Philos 123.
Hoch, Stephen J & George F Loewenstein. “Time-Inconsistent Preferences and
Consumer Self-Control” (1991) 17:4 J Consum Res 492.
Hodson, John D. “The Principle of Paternalism” (1977) 14:1 Am Philos Q 61.
Ipsos-Mori. 2015 Political Monitor (12 December 2015), at 26, online:
<https://www.ipsos-mori.com/Assets/Docs/Polls/political-monitor-december-
365
2015-charts.pdf>.
Ipsos Reid. Views Toward Organ and Tissue Donation and Transplantation (Ottawa,
2010).
John, Peter, Graham Smith & Gerry Stoker. “Nudge Nudge, Think Think: Two
Strategies for Changing Civic Behaviour” (2009) 80:3 Polit Q 361.
Johnson, Eric J & Daniel Goldstein. “Do Defaults Save Lives?” (2003) 302:5649
Science 1338.
Johnson, Samuel. “Status Quo Bias in Decision Making” (1988) 1 J Risk Uncertain 7.
Jolls, Christine & Cass R Sunstein. “Debiasing Through Law” (2006) 35:1 J Legal
Stud 199.
Jones, Philip & John Cullis. “‘Individual Failure’ and the Analytics of Social Policy”
(2000) 29:1 J Soc Policy 73.
Jones, Rhys, Jessica Pykett & Mark Whitehead. Changing Behaviours: On the Rise of the
Psychological State (Cambridge: Edward Elgar, 2013).
Kahneman, Daniel. Thinking, Fast and Slow (Toronto: Anchor Canada, 2013).
Kahneman, Daniel, Jack L Knetsch & Richard H Thaler. “Anomalies: The
Endowment Effect, Loss Aversion, and Status Quo Bias” (1991) 5:1 J Econ
Perspect 193.
Kenrick, Douglas T et al. “Deep Rationality: The Evolutionary Economics of
Decision Making” (2009) 27:5 Soc Cogn 764.
366
KIRO Team 7. “Stealth Donation Collection Earns Parks Millions”, KIRO TV (2010),
online: kirotv.com <http://www.kirotv.com/news/news/stealth-donation-
collection-earns-parks-millions/nDRqp/>.
Klein, Daniel B. “Reply to Sunstein” (2004) 1:2 Econ J Watch 274.
Klick, Jonathan & Gregory Mitchell. “Government Regulation of Irrationality: Moral
and Cognitive Hazards” (2006) 90 Minn Law Rev 1620.
Korobkin, Russell. “Libertarian Welfarism” (2009) 97:6 Calif Law Rev 1651.
Kruglanski, Arie W & Gerd Gigerenzer. “Intuitive and Deliberate Judgments Are
Based on Common Principles” (2011) 118:1 Psychol Rev 97.
Kurtz, Sheldon F & Michael J Saks. “The Transplant Paradox: Overwhelming Public
Support for Organ Donation vs. Under-Supply of Organs: The Iowa Organ
Procurement Study” (1996) 21 J Corp Law 767.
Legacy of Life. “How to Be a Donor,” online:
<http://www.legacyoflife.ns.ca/how_to_be_a_donor/index.html>.
Loewenstein, George et al. “Warning: You are about to be nudged” (2015) 1:1 Behav
Sci Policy 35.
Lunn, Peter D. “Are Consumer Decision-Making Phenomena a Fourth Market
Failure?” (2015) 38 J Consum Policy 315.
Lusk, Jayson L. “Are you smart enough to know what to eat? A critique of
behavioural economics as justification for regulation” (2014) 41:3 Eur Rev
367
Agric Econ 355.
Madrian, Brigitte C & Dennis F Shea. “The Power of Suggestion: Inertia in 401(k)
Participation and Savings Behavior” (2001) 66:4 Q J Econ.
May, Kathryn. “Government’s New Innovation ‘Hub’ Open to New Thinking”, (11
February 2015), online: Ottawa Citizen <http://ottawacitizen.com/business/local-
business/governments-new-innovation-hub-open-to-new-thinking>.
Mckenna, FP, DM Warburton & M Winwood. “Exploring the limits of optimism: the
case of smokers’ decision making” (1993) 84:3 Br J Psychol 389.
McNeil, Barbara J et al. “On the Elicitation of Preferences for Alternative Therapies”
(1982) 306 N Engl J Med 1259.
Mill, John Stuart. On Liberty (London: Filiquarian Publishing, 2006).
Mitchell, Gregory. “Libertarian Paternalism Is An Oxymoron” (2005) 99:3 Northwest
Univ Law Rev 1245.
Mitchell, Gregory. “Why Law and Economics’ Perfect Rationality Should Not Be
Traded for Behavioral Law and Economics' Equal Incompetence” (2002) 91:1
Georgetown Law J 67.
Mols, Frank et al. “Why a nudge is not enough: A social identity critique of
governance by stealth” (2015) 54 Eur J Polit Res 81.
Morewedge, Carey K et al. “Debiasing Decisions: Improved Decision Making With a
Single Training Intervention” (2015) 2:1 Policy Insights from Behav Brain Sci
368
129.
Müller, Sabine & Henrik Walter. “Reviewing Autonomy: Implications of the
Neurosciences and the Free Will Debate for the Principle of Respect for the
Patient’s Autonomy” (2010) 19 Cambridge Q Healthc Ethics 205.
NHS Blood and Transplant. “Organ Donation and Transplantation: Activity Report
(2011/12)” (2012), online: <http://www.odt.nhs.uk/uk-transplant-
registry/annual-activity-report/>.
Nozick, Robert. Anarchy, State, and Utopia (Basic Books, 1974).
O’Connor, Anette M. “Effects of framing and level of probability on patients’
preferences for cancer chemotherapy” (1989) 42:2 J Clin Epidemiol 119.
Organ Donation UK. “Summary of Market Research Findings” (2015) online: NHS
Blood & Transplant
<https://www.organdonation.nhs.uk/statistics/latest_statistics/>.
Organ Donation Wales, online: <http://organdonationwales.org/?lang=en>.
Pettit, Philip. “Freedom as Antipower” (1996) 106:3 Ethics 576.
Quigley, Muireann. “Nudging for Health: On Public Policy and Designing Choice
Architecture” (2013) 21:4 Med Law Rev 588.
Rebonato, Riccardo. “A Critical Assessment of Libertarian Paternalism” (2014) 37:3 J
Consum Policy 357.
Redmon, Kevin Charles. “Who’s Saving Electricity in Your Neighborhood?”, Pacific
369
Standard (March 2012), online: <http://www.psmag.com/nature-and-
technology/whos-saving-electricity-in-your-neighborhood-39932>.
Rozin, Paul et al. “Nudge to nobesity I: Minor changes in accessibility decrease food
intake” (2011) 6:4 Judgm Decis Mak 323.
Saghai, Yashar. “Salvaging the concept of nudge” (2013) 39:8 J Med Ethics 487.
Sandbu, Martin E. “Valuing Processes” (2007) 23 Econ Philos 205.
Schnellenbach, Jan. “Nudges and norms: On the political economy of soft
paternalism” (2012) 28 Eur J Polit Econ 266.
Scoccia, Danny. “Paternalism and Respect for Autonomy” (1990) 100:2 Ethics 318.
Selinger, Evan & Kyle Powys Whyte. “What Counts as a Nudge?” (2012) 12:2 Am J
Bioeth 11.
Service Ontario. “Organ and tissue donor registration,” online:
<https://www.ontario.ca/page/organ-and-tissue-donor-registration#section-1>.
Shah, Anuj K, Sendhil Mullainathan & Eldar Shafir. “Some Consequences of Having
Too Little” (2012) 338:6107 Science (80- ) 682.
Silberzahn, Raphael & Eric L Uhlmann. “Many Hands Make Tight Work” (2015)
526:7572 Nature 189.
Slovic, Paul. “The Construction of Preference” (1995) 50:5 Am Psychol 364.
Smith, N Craig, Daniel G Goldstein & Eric J Johnson. “Choice Without Awareness:
Ethical and Policy Implications of Defaults” (2013) 32:2 J Public Policy Mark
370
159.
Soman, Dilip & Maggie Wenjing Liu. “Debiasing or rebiasing? Moderating the
illusion of delayed incentives” (2011) 32 J Econ Psychol 307.
Stanovich, Keith E & Richard F West. “Individual differences in reasoning:
implications for the rationality debate?” (2000) 23 Behav Brain Sci 645.
Stewart, Hamish. “A Critique of Instrumental Reason in Economics” (2008) 11:01
Econ Philos 57.
Stewart, Hamish. “The Limits of the Harm Principle” (2010) 4 Crim Law Philos 17.
Sugden, Robert. “The Opportunity Criterion: Consumer Sovereignty Without the
Assumption of Coherent Preferences” (2004) 94:4 Am Econ Rev 1014.
Sugden, Robert. “Why incoherent preferences do not justify paternalism” (2008) 19
Const Polit Econ 226.
Sunstein, Cass R. Simpler: The Future of Government (New York: Simon & Schuster,
2013).
Sunstein, Cass R. Why Nudge? The Politics of Libertarian Paternalism (New Haven:
Yale University Press, 2014).
Sunstein, Cass R. Choosing not to Choose: Understanding the Value of Choice (New York:
Oxford University Press, 2015).
Sunstein, Cass R. “The Storrs Lectures: Behavioral Economics and Paternalism”
(2013) 122:7 Yale Law J 1826.
371
Sunstein, Cass R. “Was Mayor Bloomberg a Nanny?” (2014) 1:May Harvard Public
Heal Rev.
Sunstein, Cass R. “Legal Interference with Private Preferences” (1986) 53:4 Univ
Chicago Law Rev 1129.
Sunstein, Cass R. “Switching the Default Rule” (2002) 77 New York Univ Law Rev
106.
Sunstein, Cass R. “The Ethics of Nudging" (2014) [unpublished, available online:
<http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2526341>].
Sunstein, Cass R. "Do People Like Nudges?" (2016) [unpublished, available online:
<http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2604084>.
Sunstein, Cass R & Richard H Thaler. “Libertarian Paternalism Is Not an
Oxymoron” (2003) 70:4 Univ Chicago Law Rev 1159
Tengland, Per-Anders. “Behavior Change or Empowerment: On the Ethics of
Health-Promotion Strategies” (2012) 5:2 Public Health Ethics 140.
Thaler, Richard. “The Argument Clinic”, (2010), online: Cato Unbound
<http://www.cato-unbound.org/2010/04/16/richard-thaler/argument-clinic>.
Thaler, Richard H & Shlomo Benartzi. “Save More Tomorrow: Using Behavioral
Economics to Increase Employee Saving” (2004) 112:1 J Polit Econ s164.
Thaler, Richard H & Cass R Sunstein. Nudge: Improving Decisions About Health,
Wealth, and Happiness (New York: Penguin Books, 2009).
372
Thaler, Richard H & Cass R Sunstein. “Libertarian Paternalism” (2003) 93:2 Am
Econ Rev 175.
Thomas, Ayanna K & Peter R Millar. “Reducing the Framing Effect in Older and
Younger Adults by Encouraging Analytic Processing” (2012) 67:2 Journals
Gerontol Ser B Psychol Sci Soc Sci 139.
Todorov, Alexander et al. “Inferences of Competence from Faces Predict Election
Outcomes” (2005) 308:5728 Science 1623.
Trillium Gift of Life Network. “Registration Stats”, online:
<https://beadonor.ca/scoreboard>.
Tversky, Amos & Daniel Kahneman. “Rational Choice and the Framing of
Decisions” (1986) 59:4 J Bus S251.
Veetil, Vipin P. “Libertarian paternalism is an oxymoron: an essay in defence of
liberty” (2011) 31 Eur J Law Econ 321.
Wansink, Brian. “Mindless Eating: The 200 Daily Food Decisions We Overlook”
(2007) 39:1 Environ Behav 106.
Weinrib, Ernest J. “Utilitarianism, Economics, and Legal Theory” (1980) 30:3 Univ
Tor Law J 307.
Weinstein, Neil D. “Unrealistic Optimism About Susceptibility to Health Problems”
(1982) 5:4 J Behav Med 441.
Weinstein, Neil D. “Unrealistic Optimism About Future Life Events” (1980) 39:5 J
373
Pers Soc Psychol 806.
Weinstein, Neil D & William M Klein. “Resistance of Personal Risk Perceptions to
Debiasing Interventions” (1995) 14:2 Heal Pschology 132.
Welsh Government. “Update to Survey of Public Attitudes to Organ Donation: Key
figures from Wave 10” (2 January 2016), online: <http://gov.wales/statistics-
and-research/public-attitudes-organ-donation/?lang=en>.
White, Mark D. The Manipulation of Choice: Ethics and Libertarian Paternalism (New
York: Palgrave Macmillan, 2013).
Whyte, Kyle Powys et al. “Nudge, Nudge or Shove, Shove—The Right Way for
Nudges to Increase the Supply of Donated Cadaver Organs” (2012) 12:2 Am J
Bioeth 32.
Wilkinson, T M. “Thinking Harder About Nudges” (2013) 39:8 J Med Ethics 486.
Wilkinson, T M. “Nudging and Manipulation” (2013) 61:2 Polit Stud 341.
Zink, Sheldon, Rachel Zeehandelaar & Stacey Wertlieb. “Presumed vs Expressed
Consent in the US and Internationally” (2005) 7:9 Ethics J Am Med Assoc.
Zwane, Alix Peterson et al. “Being surveyed can change later behavior and related
parameter estimates” (2011) 108:5 Proc Natl Acad Sci U S A 1821.
374
Appendix I – Glossary
Term Definition
Active choosing A de-biasing nudge that enhances the decisional
autonomy of choosers by requiring some choice, even if
that choice is to not choose.
Authentic
motivations
A chooser’s motivations that are of a higher-order nature
and independent from undue external influences such
that we can say that they reflect a chooser’s authentic self
(their fundamental values and beliefs).
Behavioural biases Anomalies in decision-making identified from
behavioural insights research.
Behavioural insights Research findings about how humans behave and make
decisions. Usually arising from the fields of behavioural
economics, social psychology and cognitive psychology.
Budget constraint Total set of finite resources available to a chooser to apply
to a particular choice problem (local opportunity set) or
all choice problems (global opportunity set).
Choice architecture Intentionally designing the context of choice.
375
Choice(s) A chooser’s selected options resulting from the decision-
making process.
Choosing The process of selecting an option from an opportunity
set.
Costs All economic and non-economic costs associated with a
particular option in an opportunity set.
Counter-biasing Nudges that, in an attempt to neutralize an existing
expressed bias, introduce the expression of a new or
additional bias (e.g. graphic warning labels on cigarette
packages to counter optimism bias).
De-biasing Nudges that function by improving the process
autonomy of choosers, often by providing information or
time for deliberation (e.g. active choosing).
Decisional autonomy The autonomy of a chooser found at the option
evaluation stage of decision-making. Decisional
autonomy is found in choice evaluation that is unbiased,
informed and deliberative.
Heuristics Shortcuts or “rules of thumb,” predominantly generated
by System 1.
376
Higher-order
motivations
Motivations that reflect a chooser’s values and beliefs;
tend to be about lower-order desires – motivations about
motivations. Generated by a chooser’s System 2.
Independent
motivations
A chooser’s motivations that are not the product of
undue external influences.
Libertarian
paternalism
A theory advanced by Sunstein and Thaler that nudges
are to steer choosers towards choices that they would
have selected if they possessed full decisional autonomy.
Liberty of choice
constraint
The requirement that choice architecture must preserve
the opportunity set of choosers; the key tool used in
defining nudges from other forms of choice architecture.
Lower-order desires Automatic desires, urges or cravings in response to
choice stimuli with little reflection or deliberation.
Generated by a chooser’s System 1.
Meta-motivations Motivations about other motivations, or about the
choosing process.
Motivational
autonomy
The autonomy of a chooser found at the motivation
formation stage of decision-making. Motivations must be
authentic to carry motivational autonomy.
377
Nudge(s) A choice architectural intervention that satisfies the
liberty of choice constraint by preserving a chooser’s
opportunity set.
Opportunity cost(s) Costs relating to finite resources that, if spent on a
particular option or opportunity set, cannot be used on
other options within the same set, or in other opportunity
sets.
Opportunity set(s) The set of options and their associated costs that a
chooser faces in a given choice problem.
Option(s) The members of an opportunity set.
Outcome autonomy The autonomy found when resulting choices (selected
options) match a chooser’s authentic motivations.
Price That subset of costs related to the economic cost of the
option itself, excluding transaction costs.
Process autonomy Motivational autonomy and decisional autonomy; the
participatory element of self-governance.
Re-biasing Nudges that have no effect on the expression of
behavioural biases, instead simply re-directing existing
378
biases to a different option selection and resulting choice.
(e.g. switching which option is a default).
System 1 A fast, automatic, reactive/instinctual style of cognition;
the first part of Kahneman’s dual process theory.
System 2 A slow, reflective, deliberative style of cognition, the
second part of Kahneman’s dual process theory.
Undue external
influences
External influences on a chooser’s motivations that
operate in such a fashion that a chooser does not have
the opportunity to accept or resist the influence.