Using National Survey Respondents as Consumers in an...
Transcript of Using National Survey Respondents as Consumers in an...
IEEE Access 2015-000125
1
Abstract— Plug-in hybrid electric vehicles (PHEVs) offer the
potential to significantly reduce greenhouse gas emissions, if
vehicle consumers are willing to adopt this new technology.
Consequently, there is much interest in exploring PHEV market
penetration models. In prior work, we developed an agent-based
model (ABM) of potential PHEV consumer adoption that
incorporated several spatial, social, and media influences to
identify nonlinear interactions among potential leverage points
that may impact PHEV market penetration. In developing that
model, the need for additional data to properly inform both the
decision-making rules and agent initialization became apparent.
To address these issues, we recently conducted and analyzed an
extensive consumer survey; in this work, we modify the ABM to
reflect the survey findings. A unique aspect is a one-to-one
correspondence between agents in the model and survey
respondents, thus yielding distributions and cross-correlations in
agent attributes that accurately reflect the survey population. We
also implement a used-PHEV market, and allow agents to
purchase new or used compact PHEVs or vehicles of their
current type. Based on our prior survey response analysis, our
modified model includes a PHEV-technology threshold
component, a multinomial logistic prediction of willingness to
consider a compact PHEV based on dynamically-changing
attitudes, and agent-specific delay discounting functions that
predict the amount agents are willing to pay up front for greater
fuel savings. We thus independently account for agents’
discomfort with the new PHEV technology, their desire to drive a
more environmentally friendly vehicle, and their willingness to
pay a higher sticker price for a PHEV. Results of 10 survey-based
ABM scenarios are reported with implications for policy-makers
and manufacturers. We believe close integration of the design of
consumer surveys and the development of ABMs is a key step in
This paper was submitted on 4/14/2015. This work was funded in part by
the United States Department of Transportation through the University of Vermont Transportation Research Center and a workforce development sub-
award from Sandia National Laboratories supported by the U.S. Dept. of
Energy through Inter-Entity Work Order M610000767. M.J. Eppstein is with the Department of Computer Science at the
University of Vermont, Burlington, VT 05405 (email:
[email protected]). D.M. Rizzo is with the School of Engineering, University of Vermont,
Burlington, VT 05405 (email: [email protected]).
B.H.Y. Lee is with the School of Engineering, University of Vermont, Burlington, VT 05405 (email: [email protected]).
J.H.Krupa was with the School of Engineering, University of Vermont, but
is now with Roberge Associates Coastal Engineers, LLC, Stratford, CT 06615 (email: [email protected]).
N. Manukyan was with the Department of Computer Science, University
of Vermont, but is now with Faraday, Inc., Burlington, VT 05401.
developing useful decision-support models; this work serves as an
example of one way to achieve that.
Index Terms— Plug-in hybrid electric vehicles (PHEVs);
agent-based model; market penetration; electric vehicle
adoption; vehicle choice simulation, vehicle choice survey.
I. INTRODUCTION
urrently, transportation accounts for approximately one
third of greenhouse gas (GHG) emissions in the U.S., and
is its fastest growing source [1]. Plug-in hybrid electric
vehicles (PHEVs) offer the potential to significantly reduce
GHG emissions [2]-[4]. The degree to which this can occur
depends on current regional electric power generation sources
[5][6], smart grid technologies to improve grid efficiency and
reduce peak demands [7]-[9], and power generation shifts
from coal to cleaner sources, including natural gas and
renewables (e.g., wind, solar) [10][11]. Potential vehicle-to-
grid technologies could further reduce peak electricity
demands [12][13] and, therefore, have a positive feedback on
the grid efficiency. In addition, PHEVs have projected
lifecycle costs that are lower than internal combustion engines
[14]. From the consumer perspective, PHEVs can provide
large savings in fuel costs without the range limitations of all-
electric vehicles (EVs); the EPA/DOT sticker data for the
2013 Chevy Volt states that the vehicle “will save $6,850 in
fuel costs over 5 years compared to the average new vehicle.”
Realization of these potential PHEV benefits ultimately
depends on consumers’ willingness to adopt this new
technology. Although public awareness of PHEVs is growing
[15], consumers continue to have concerns regarding the new
battery technologies in electric drive vehicles (EVs, PHEVs,
and hybrid-electric vehicles) and the potential inconvenience
of recharging [16]-[22]. While previous work indicates
consumers are willing to pay a premium for greater fuel
efficiency [22][23], initial Chevy Volt sales have fallen short
of projections [24], and are outcompeted by the otherwise
similar gas-powered Chevy Cruze [25].
Consequently, there is significant interest in modeling the
potential market penetration of PHEVs (and other electric-
drive vehicles) under a variety of scenarios (e.g., [26]-[36]; for
a recent review see [37]) to understand the impacts of
potential policies, incentives, energy prices, and other factors
affecting consumer vehicle choice. However, despite this
progress, there exists a need for greater micro-data to properly
inform such models [30][38].
Using National Survey Respondents as
Consumers in an Agent-Based Model of Plug-In
Hybrid Vehicle Adoption
Margaret J. Eppstein, Donna M. Rizzo, Brian H.Y. Lee,
Joseph S. Krupa, and Narine Manukyan
C
IEEE Access 2015-000125
2
For example, in prior work [30] we developed an agent-
based model (ABM) of potential PHEV consumer adoption
that incorporated several spatial and social effects, including
threshold effects [39], homophily [40], and conformity
[41][42], as well as media influences, to identify nonlinear
interactions among potential leverage points that may impact
PHEV market penetration. In that work, the need for
additional data to properly inform both the agent decision-
making rules and agent initialization became apparent.
To address these issues, in a joint collaboration between the
University of Vermont and Sandia National Laboratories, we
performed an extensive online survey of 1,000 adult U.S.
residents using the Amazon Mechanical Turk (AMT) crowd-
sourcing platform [43] in July, 2011. AMT has emerged as a
reliable and relatively inexpensive survey tool [44]-[48]. Our
AMT survey, designed in large part to specifically inform and
improve the ABM first presented in [30], comprised 105
questions regarding consumer demographics, purchasing
decisions, factors influencing vehicle choice (including
PHEVs), attitudes toward the environment and energy, and
discounting questions that assessed how much a respondent
would be willing to pay up front for a vehicle that would
provide various levels of future benefits. After extensive
quality control (described in [49]), 911 out of the 1000
responses were deemed reliable and included in subsequent
analyses. Survey respondents were fairly representative of
national demographics (e.g., see Table 1 and Fig. 1 in [49]).
For example, the number of participants that responded from
47 states across the U.S. was proportional to each state’s 2010
census population (R2=0.91), and the distribution of vehicle
number per household was very similar to national data [50].
Although many survey respondents came from households
with multiple vehicles (they reported an average of 1.7
vehicles per household) and possibly multiple drivers, they
were directed to answer questions regarding their personal
driving habits and the vehicle (past, present, or future)
available for their primary use. Other than household income,
questions on demographics and attitudes related to the survey
participant, not the household. To minimize participation bias,
we specifically ordered the questions so that participants were
not aware that the survey was related to vehicle purchasing
attitudes until the third of six sections, and the PHEV focus
was not revealed until the fifth section.
Our AMT survey yielded several useful results. For
example, more AMT participants (86%) stated that financial
benefits due to fuel savings would be very important in
considering the purchase of a PHEV than those (55%) who
stated that reducing greenhouse gas emissions would be very
important; this is consistent with [17][51][52], who found that
most consumers prioritize financial benefits over
environmental benefits. However, our survey analysis also
showed that those who are most concerned about climate
change had 44.4 times greater odds of being willing to
consider purchasing a PHEV than those least concerned,
whereas those who felt that fuel savings were very important
had only 15.25 times greater odds to consider purchasing a
PHEV than those who ranked this as relatively unimportant.
Like other recent studies that found strong correlations
between pro-environmental attitudes and stated preferences
for alternative fuel vehicles [53][54], this implies that
environmental benefits may actually be a more powerful
motivator for PHEV adoption than financial benefits, for those
who place high importance on environmental concerns. These
findings provide motivation for properly modeling
dynamically changing environmental attitudes when
predicting PHEV market penetration. The complete survey
and all participant-level survey responses are available online
[55]. See [49] for distributions and correlations between
consumer attributes and attitudes, and an in-depth statistical
analysis on several factors we found most influential and
predictive of consumer willingness to consider PHEVs.
In this work, we used the AMT survey results to improve
the ABM decision-making rules for potential PHEV adoption,
and we populate the ABM with 911 agents, each based
directly on the individual responses of one of the 911 survey
participants. ABMs are typically initialized by drawing from
statistical distributions of attributes, informed by survey data
where known. However, data on cross-correlation of many
attributes are rarely available and, even if known, it is
challenging to create random populations that respect
distributions and cross-correlations between all attributes. By
assigning each agent with the attributes of an actual consumer,
distributions and cross-correlations of agent attributes
accurately reflect those in the survey population. We use the
modified ABM to study potential PHEV market penetration
under various scenarios with different incentives, fuel costs
and model assumptions, to illustrate how the improved model
can provide useful insights to policy-makers and
manufacturers. This close integration of consumer survey
design, analysis, and agent-based modeling is an important
advance in creating useful decision-support models for
planning in transportation and other fields.
II. AGENT-BASED MODEL
We adapted the ABM of [30] to closely reflect the findings
of the PHEV survey [49], both in terms of agent attributes and
agent vehicle-purchasing decision options and rules. In this
ABM, agents are models of vehicle consumers. To clearly
distinguish when we are referring to simulated vehicle
consumers in the model and actual vehicle consumers in the
AMT survey, we generally refer to the former simply as
“agents” and the latter as “AMT participants or respondents”.
Readers are referred to [30] for many of the details and
rationale behind the original ABM. Below, we focus on
relevant modifications to this model based on the AMT
survey.
A. Agent Initialization
We built a population of 911 agents directly modeled on the
911 survey respondents, thus obtaining data-driven
distributions and cross-correlations of all agent attributes.
After quality control, 3.8% of the 911 surveys had a few
missing values (specifically, for the attributes used here, there
were 31 respondents with 1 missing attribute, 3 respondents
with 2 missing attributes, and 1 respondent with 4 missing
attributes). Rather than discard these 35 surveys, we opted to
replace the missing values with the median response category
for that question. Most of the survey responses were ordinal or
IEEE Access 2015-000125
3
categorical. For survey responses selected from ordinal bins,
agents were assigned real values drawn from a uniform
random distribution over the selected answer’s bin range to
obtain more continuous distributions of attributes. For daily
per capita vehicle miles traveled (VMT), the bin ranges were
skewed slightly higher than those specified in the survey so
that the resulting distribution more closely matched U.S.
averages [50]. We felt the latter was important in
determining the distance agents might drive in the all-
electric range of a PHEV.
Each agent in the ABM has several associated
demographic attributes modeled directly after survey
responses of exactly one AMT participant, including
individual age, annual household income, residential
location (rural, suburban, urban), and location in the
political spectrum (11% far left, 32% left, 33% center, 19%
right, 4% far right). Each agent also has vehicle attributes
modeled after the responses of its corresponding AMT
participant, including:
(i) The typical number of years of car ownership;
(ii) Personal Daily VMT;
(iii) Age of the vehicle currently available for their primary
use;
(iv) Current vehicle class reported as either compact (24.1%),
midsize (29.5%), full-size (10.3%), minivan or SUV
(20.1%), full-size van or pickup (5.5%). There were an
additional 10.5% reported as no vehicle, motorcycle, or
other; rather than discard these surveys, we assigned them
to the compact vehicle class;
(v) The manufacturer suggested retail price (MSRP) of their
current vehicle;
(vi) The fuel type (in this study, recorded only as PHEV or
non-PHEV), all-electric range (if any) and miles per
gallon (MPG) when not in all-electric mode; and
(vii) Whether they purchased their most recent vehicle new
(32%) or used (60%); of the remaining 8% who said they
had never owned a vehicle, we assigned 4% to new and
4% to used.
Rather than asking AMT survey participants to self-report
the original MSRP and fuel efficiency of their current vehicles
(which we thought would likely be highly unreliable), we
approximated the values for vehicle MSRP and MPG from the
database located at www.cars.com, based on AMT
respondents’ self-reported make, model, and year. MSRP was
averaged over all styles brought to 2011 U.S. dollar
equivalents, assuming 3% annual inflation. MPG was
averaged over all driving modes. See Table I for initial agent
population means for several of the survey-based variables.
In addition, agents have survey-based attitudinal
characteristics representing their levels of concern regarding
GHG emissions, U.S. energy independence, transportation
fuel costs, the importance of projecting one’s environmental
image through vehicle choice, and the potential inconvenience
of recharging. All of these attitudinal characteristics (except
environmental image) are subject to change through social
and/or media influences. We inferred the degree to which
agents are susceptible to media influences (6% low, 39%
medium, 56% high) and social influences (59% low, 33%
medium, 9% high), based on survey responses to a series of
questions asking to what degree various factors had influenced
their current opinions regarding U.S. transportation energy
usage.
Agents were assigned a PHEV-comfort threshold T based
on survey participants’ responses to the question, “Assuming
you were buying a new vehicle, what is your comfort level in
purchasing/leasing the new PHEV technology rather than a
more established type of vehicle, assuming it had all the other
features you desired and was within your budget?” The
participants responded by selecting their comfort threshold,
phrased as a percentage range of PHEVs they would have to
see around them before they would be comfortable
considering a PHEV. The distribution of responses in each
range was similar to that of an identically binned Gaussian
distribution with a mean of 25% and a standard deviation of
48% [49], as shown in Fig. 1. In our initial agent population,
32% had a 0% threshold (innovators), another 4% had a
threshold 0 < T ≤ 5% (early majority adopters), and only 7%
would never consider a PHEV.
Agents were assigned a separate probability of “willingness
to consider a compact PHEV” attribute (W) reflecting whether
survey respondents stated they “would definitely” (28%),
“might” (51%), or “would not” (21%) consider purchasing a
compact PHEV, respectively. Agents corresponding to the
“would definitely” group were assigned W = 1, but for the
remainder we dynamically estimated this probability every
time step, based on a multinomial logistic model described in
the next section.
Finally, survey respondents indicated how much they would
be willing to pay up front for a vehicle (not necessarily a
PHEV) that achieves 6 different specified dollar amounts in
fuel savings per year. These delay discounting responses were
associated with individual agents and used to interpolate the
maximum price premium agents would pay for PHEVs to
achieve estimated annual fuel savings based on projected
gasoline and electricity prices over the agent’s expected years
of vehicle ownership and daily VMT.
TABLE I
MEAN VALUES OF SEVERAL SURVEY-BASED VARIABLES IN
THE INITIAL POPULAITON OF 911 AGENTS.
Agent Attribute Mean Value
Household Income $54,231
Age 34 years
Daily VMT 43.5 km
(27 mi)
Age of Initial Primary
Vehicle
8.5 years
Expected duration of car
ownership
7.0 years
Fuel Efficiency of Initial
Primary Vehicle
10.2 km/L
(24 mpg)
MSRP of Initial Primary
Vehicle
$28,585
IEEE Access 2015-000125
4
Fig. 2. Synthesized spatial correlations across the 38.9 km2 (15 mi2) simulated domain in normalized values of (a) household income, where red is highest
income and blue is lowest, (b) education level, where red is highest education and blue is lowest, and (c) political leaning, where red indicates far right and
blue indicates far left; (d) agent residence locations (white dots) are generated to be sparse in the rural region (red background), have intermediate spacing in
the suburban region (green background), and are dense in the urban region (blue background).
Fig. 1. Histogram of AMT survey responses of self-reported PHEV comfort threshold ranges, superimposed with a truncated normal distribution binned the
same way.
IEEE Access 2015-000125
5
All attributes described so far were, for each agent, based
directly on survey responses of one AMT participant, thus
respecting the distributions and cross-correlations of these
attributes as observed in the survey population. However,
since survey respondents came from 47 states across the U.S.,
and given our desire to model spatial and social effects of
consumers in a contiguous geographical region, we
synthesized agent locations in a simulated 38.9 km2 (15 mi2)
domain. To do this, we used 10 iterations of a Kohonen self-
organizing map (SOM) [56] to locate the 911 agents in space
based on the normalized values of 4 attributes. This was done
to generate spatial correlations in household incomes (Fig. 2a),
education levels (Fig. 2b), political leanings (Fig. 2c), and
appropriate spacing and grouping of agents based on
respondents from urban, suburban, and rural regions (Fig. 2d).
The resulting spatial correlations (Fig. 3a) do not alter the
observed cross-correlations between these 4 attributes (Fig.
3b), since individual agents were mapped to various locations
without changing their attributes.
Each agent computes a perceived PHEV market share given
vehicles owned by agents in its individual “spatial
neighborhood”. We defined the latter to include all agents
within each agent’s daily VMT/4, based on the assumption that
consumers who drive more perceive more vehicles. The
resulting spatial neighborhoods had a minimum of 31 agents, a
median of 476 agents, and a maximum of 910 agents.
Similarly, social influences were computed within an agent’s
“social neighborhood,” defined to include all agents in its
spatial neighborhood that have annual household incomes
within ±$10,000 and age within ±15 years of its own attribute
values, based on the principle of homophily [40]. The
resulting social neighborhoods had a minimum of 5 agents, a
median of 49 agents, and a maximum of 190 agents.
B. Agent Decision-Making
During each simulated annual time step, all agents are updated
asynchronously in random order. Prior to making any vehicle
purchasing decisions, certain agent attitudes are updated
subject to both external and internal influences. Specifically,
concerns regarding GHG emissions G and U.S. energy
independence E are updated by adding the product of the
average annual change in media coverage conveying the need
to reduce transportation energy and the agent’s susceptibility
to media influence SM. In our simulations, the level of media
coverage was stochastically increased from 0.05 to 0.3 over
the 14 years of the simulation, based on the assumption that
the increase in energy and climate-related impacts will be
reflected by an increase in media coverage. Similarly, concern
regarding fuel costs F is updated by adding the average annual
change in gasoline prices times SM, where gasoline prices are
stochastically increased from $0.911/L ($3.45/gal; the 2011
U.S. average) to various projected values in 2040 (EIA, 2013).
Electricity prices are assumed to increase linearly from
$0.099/kWh in 2011 to $0.108/kWh in 2040 [58]. Four of the
attitude attributes (political leaning, concerns regarding GHG
emission, concerns regarding U.S. energy independence, and
concerns regarding the potential inconvenience of recharging)
are also modified through social influences, subject to the
attitudes of other agents in their social network and their level
of social susceptibility using the social update procedure
described in [30].
Following its attitude updates, an agent probabilistically
decides whether to consider purchasing a new vehicle during
the current year, based on the age of its current vehicle and a
normal probability distribution centered on the agent-specific
number of years the agent expects to own the vehicle. The
agent then estimates the expected annual fuel-cost savings of a
PHEV relative to a vehicle of its current type over the
expected number of years of vehicle ownership, based on its
daily VMT and the projected gasoline and electricity cost (this
Fig. 3. (a) Spatial correlation (Pearson) of household income (I), highest education (E), and political leaning (P) as a function of distance; and (b) all pairwise
cross-correlations (Pearson) between these three attributes and region (R), as observed in the AMT survey.
I,E I,P I,R E,P E,R P,R-0.15
-0.1
-0.05
0
0.05
0.1
0.15
0.2
0.25
attribute pairs
cro
ss-c
orr
ela
tio
n
0 0.5 1 1.5 20.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
neighborhood size
sp
atia
l co
rre
latio
n
household income
highest education
political leaning
(a) (b)
IEEE Access 2015-000125
6
rational computation assumes that consumers will take
advantage of the increasing number of web tools (e.g.,
[59][60]) that provide easy access to this information), and the
maximum expected annual cost of ownership of these two
vehicle types, including loan payments after trading in their
current vehicle. The assumptions regarding vehicle financing
and depreciation are described in [30].
The agent next selects between five vehicle purchasing
options in the following order: (1) a new compact PHEV
similar to a Chevy Volt (with a pre-incentive sticker price of
$40,000, an all-electric range of 61 km (38 mi), an average of
16 km/L (38 mpg) when running on gasoline, and a 16 kWh
battery), (2) a new vehicle similar in type to its current
vehicle, (3) a used compact PHEV similar to a Chevy Volt
(that is not as old as its current vehicle), (4) a used vehicle of
the same type as its current vehicle (but not as old), or (5) no
vehicle purchase. We selected the Chevy Volt as
representative of a first-generation PHEV targeted toward the
general public and with a relatively high all-electric range,
sufficient for the U.S. average daily VMT of 47 km (29 mi)
[50].
Part of the agent decision-making process depends on
whether the agent is even willing to consider purchasing a
compact PHEV; this is considered true if either the agent has a
willingness value of W = 1 or if a uniform random number is
less than the probability predicted by a multinomial logistic
model estimating its likelihood of seriously considering a
compact PHEV, based in part on its dynamically changing
attitudes. Specifically, we used multiple ordinal logistic
regression to build a predictive model of survey respondents
who reported they “would definitely” or “would not” consider
a compact PHEV based on 7 attributes: the agent’s political
leaning, their attitudes regarding the environment and energy,
their concerns regarding fuel costs and recharging, their desire
to project an environmental image through vehicle ownership,
and their current vehicle class, the latter reflecting their
willingness to potentially switch vehicle classes to purchase a
compact PHEV (since most first generation PHEVs are
compact vehicles). In our previous PHEV survey analysis
[49], these 7 attributes were the most predictive set we found;
five-fold cross-validation yielded an average of 80% positive
predictive value when testing the survey data. The logistic
prediction was incorporated into agent decision-making as
described below.
An agent will purchase a new compact PHEV if all of the
following are true:
(i) The price premium of a new PHEV (less any available
rebates) is within the agent’s willingness to pay extra
(linearly interpolated from their delay discounting
attributes based the amount of fuel costs they expect to
save annually),
(ii) The maximum annual cost of ownership is affordable
(defined as ≤ 20% of the annual household income, a
common rule of thumb [61]),
(iii) The agent is willing to consider a compact PHEV
(estimated from the logistic function as described above),
(iv) The agent is willing to consider buying a new car, and
(v) The number of PHEVs owned by agents within its spatial
neighborhood exceeds the agent’s threshold T (except in
Scenario 2 where we did not use the threshold).
If the agent does not purchase a new PHEV, it then decides
whether to purchase a new vehicle similar in type to its current
vehicle based on whether (a) the agent is willing to consider
buying a new car, and (b) the vehicle is affordable.
If the agent does not purchase a new vehicle, it then
assesses whether to purchase a used vehicle, either a compact
PHEV or a conventional vehicle of the same type as its current
vehicle. We assume that used vehicles of the current vehicle
type are available for all years under consideration, and the
agent considers the most recent used model that is affordable
(based on the vehicle depreciation formula in [30]). For used
compact PHEVs, we implement a used PHEV market; we
track which PHEV model years have been traded in and allow
only these years to be purchased as used PHEVs. The agent
considers the most recent used PHEV that is both available
and affordable.
The agent will purchase the used PHEV if all of the
following are true:
(i) The age of the used PHEV is less than the age of its
current vehicle,
(ii) The difference in price between the used PHEV and the
used vehicle of its current type under consideration falls
within the agent’s willingness to pay extra,
(iii) The agent is willing to consider a compact PHEV, and
(iv) The number of PHEVs owned by agents within its spatial
neighborhood exceeds its threshold T (except in Scenario
2 where we did not use the threshold).
If the agent does not purchase a used PHEV, it will
purchase the used vehicle similar in type to its current vehicle,
provided the vehicle age is less than the age of its current
vehicle. Otherwise, the agent does not buy any vehicle and
retains its current vehicle during this time step.
III. RESULTS
We report on 10 scenarios after 14 years of simulation,
starting from 0% compact PHEVs in 2011 to projected market
penetration in 2025 (Fig. 4, Table II). Similar to the findings
in [30], we did not observe qualitative differences in the
results of different stochastic runs (as evidenced here by the
relative smoothness of the curves in Fig. 4), so individual runs
of each of the 10 scenarios were used to illustrate the relative
model sensitivities to the different parameters in these
scenarios. Since, on average, agents consider buying a vehicle
every 7 years, this simulation period corresponds to
approximately 2 vehicle purchases per agent. In the base case
scenario, we assumed all agents will consider purchasing a
new car, agent thresholds T and initial vehicle classes are as
reported by survey participants; gas prices stochastically trend
upward toward the Annual Energy Outlook 2013 reference
gasoline price projection [58] of $1.14/L ($4.32/gal) in 2040
(in 2011 dollars); and we assume that PHEV incentives exist
for the duration of the 14-year simulation in the form of a
$7,500 federal tax rebate (as currently exists in the U.S.), a
IEEE Access 2015-000125
7
$4,000 manufacturer rebate (as existed for the Chevy Volt in
June, 2013), and a $1,500 state tax rebate (as currently exists
in California). Each of the remaining scenarios varies in
exactly one parameter, as follows. Scenario 2 omitted the use
of the threshold T in the agent decision-making process.
Scenario 3 assumed that agents who had purchased their most
recent vehicle used would not consider purchasing their next
vehicle new. Scenario 4 assumed that all agents were compact
Fig. 4. Market penetration of compact PHEVs for all 10 scenarios described in Table II.
TABLE II.
PERCENTABLE OF COMPACT PHEV’S IN THE SIMULATED FLEET AFTER 14 YEARS OF SIMULATION (STARTING FROM 0% IN 2011
THROUGH 2025) FOR EACH OF THE 10 SCENARIOS DESCRIBED IN SECTION III AND ILLUATRATED IN FIG. 4. VALUES IN BOLD
DIFFER FROM THE BASE CASE.
Scenario Use Thres-
hold T?
All will
consider
new cars?
All initially
compact
car owners?
Predicted
gasoline price
in 2040
Rebate
amounts
(Federal +Manuf.
+State)
Rebate
duration
% compact
PHEVs in
2025
% change
compared to
Base case
1) Base case Yes Yes No $1.14/L
($4.32/gal)
$7,500
+$4,000
+$1,500
NA 17.8% NA
2) No
threshold No Yes No $1.14/L
($4.32/gal)
$7,500
+$4,000
+$1,500
NA 33.9% +16.1%
3) Some used-only buyers
Yes No No $1.14/L ($4.32/gal)
$7,500 +$4,000
+$1,500
NA 8.0% -9.8%
4) All compact
buyers
Yes Yes Yes $1.14/L ($4.32/gal)
$7,500 +$4,000
+$1,500
NA 21.2% +3.4%
5) High gas
prices
Yes Yes No $1.64/L
($6.19/gal)
$7,500
+$4,000 +$1,500
NA 18.6% +0.8%
6) Very high
gas prices
Yes Yes No $2.11/L
($8.00/gal)
$7,500
+$4,000 +$1,500
NA 19.6% +1.8%
7) No state
rebate
Yes Yes No $1.14/L
($4.32/gal)
$7,500
+$4,000 +$0
NA 14.8% -3.0%
8) No state or
manufacturer
rebate
Yes Yes No $1.14/L
($4.32/gal)
$7,500
+$0
+$0
NA 9.9% -7.9%
9) No rebates Yes Yes No $1.14/L
($4.32/gal) $0
+$0
+$0
NA 5.3% -12.5%
10) Rebates end after 5
years
Yes Yes No $1.14/L ($4.32/gal)
$7,500 +$4,000
+$1,500
5 yr 12.7% -5.1%
IEEE Access 2015-000125
8
car owners (as in the scenarios reported in [30]). Scenario 5
assumed a more rapid rise in gasoline prices based on a high
gasoline price projection [58]. Scenario 6 assumed an even
higher rise in gasoline prices (e.g., such as might occur with
an increase in gasoline tax). Scenario 7 assumed there was no
state rebate. Scenario 8 assumed no state rebate and no
manufacturer rebate; and Scenario 9 assumed no PHEV
rebates at all. Scenario 10 assumed that all rebates were
terminated after 5 years. PHEV market penetration for the 10
scenarios is shown in Fig. 4, with details summarized in Table
II.
IV. DISCUSSION
Market penetration of PHEVs (and other EVs) is still very
low and highly variable across the 50 U.S. states, ranging in
2013 from a low of essentially 0% of new car registrations in
Mississippi to a high of 1.6% in Washington and Hawaii [62].
Thus, proper model validation is impossible with such low
numbers, especially considering that agents are based on
survey respondents from 47 states. Furthermore, our ABM is
intentionally relatively simple; for example, we do not attempt
to model other types of EVs (or any other vehicles not owned
by our 911 survey respondents) or make any attempt to
account for future unknown changes in the technology or how
this might impact PHEV price or consumer willingness to
adopt. Our intent with this ABM is not to make accurate or
quantitative predictions, but rather to better understand
potential barriers and leverage points that may impact PHEV
market penetration by improving our originally proposed
PHEV model [30] using agent decision-making rules and
initialization based on data from the recently published PHEV
survey [49].
Since 911 surveys passed our quality-control standards, we
elected to limit our simulated populations to exactly 911
agents with a one-to-one correspondence between survey
respondents and agents, thus preserving distributions and
cross-correlations of respondent attributes. One could
conceivably scale-up this model using multiple copies of these
911 agents in different spatial locations. However, in our
previous work [30], we found no qualitative differences
between simulations with 10,000 agents and those with 1,000
agents, or between different stochastic runs with the same
parameters. Thus, for this study, we judged that 911 agents
were sufficient to identify trends and sensitivities. Ideally, the
model could be scaled up by conducting a larger survey within
a limited geographical region of interest; however, given our
choice to perform the survey using AMT, we could not limit
the geography of respondents beyond specifying they must be
within the U.S.
Of the parameters tested, the PHEV technology comfort
threshold T, reflecting consumer uneasiness with the new
PHEV technology, is shown to be the largest barrier to
potential PHEV market penetration, depressing potential
PHEV sales by nearly 50% in 14 years of simulation (Fig. 4
and Table II, compare Scenario 2 to Scenario 1). Note that our
implementation of this social thresholding effect is motivated
by the classic works of [39] and [62]. In [64] the authors took
an alternative approach by discretizing time into 3 sequential
phases and assigning innovators and early adopters to the first
phase (which assumed 0.1% EV market penetration), early
majority adopters to the second phase (5% EVs), and late
majority and traditionalists to the third phase (10% EVs). In
contrast, our implementation assigned percentage thresholds to
agents within the self-reported ranges of their corresponding
AMT survey respondents. Thus, agents naturally considered
whether to purchase a PHEV after market penetration had
exceeded their personal comfort threshold and the same agent
could make more than one vehicle purchase during the course
of a simulation. Although many of the survey-based agents
were willing to be early adopters of the PHEV technology,
after 14 years the total market penetration only exceeded the
thresholds for 47% of the agents in Scenario 1. In these
simulations, we assumed the heterogeneous agent-specific
threshold values were static; it is possible that as PHEV
technology becomes more widely advertised and familiar to
consumers, these threshold values may decline over time.
Furthermore, we did not model PHEV adoption by
commercial operators, such as taxis, company vehicles, etc. It
is possible that PHEVs may make more rapid inroads into
these commercial fleets and thus impact the market
penetration level perceived by consumers. Certainly, our
results indicate that aggressive steps are needed by
manufacturers (e.g., through educational advertising and
strong battery warrantees) and policy-makers (through public
service announcements, development of public recharging
infrastructure and policies, and through well-publicized
incentives) to help consumers feel more comfortable with the
PHEV technology.
If the PHEV market penetration exceeded an agents’
personal comfort threshold and the multinomial logistic model
predicted that the agent would seriously consider purchasing a
PHEV, we then assessed whether they were willing to pay the
PHEV sticker price, assuming it was within their budget.
Rather than inferring an agent’s willingness to pay extra for a
more fuel efficient vehicle from indirect indicators (e.g., as in
[64]), we used the individual delay discounting functions self-
reported by survey respondents. Thus, our model
independently accounts for agents’ discomfort with the new
PHEV technology, their desire to drive a more
environmentally friendly vehicle, and their willingness to pay
a higher sticker price for a PHEV.
When no consumer agents were willing to switch from
used-car buyers to new-car buyers, PHEV purchases dropped
by 55% (Fig. 4 and Table II, compare Scenario 3 to Scenario
1). Since PHEVs are likely to be scarce in the used-car market
for some time (and consumers may be even more
uncomfortable purchasing a used PHEV, due to a fear of
expected degradation in the battery capacity), and since
economic pressures are causing some U.S. consumers to
switch from new car buyers to used car buyers [65], this will
be a major impediment to increasing the proportion of PHEVs
in the U.S. light-vehicle fleet.
Interestingly, when we increased the number of agents who
initially owned compact cars from 35% to 100% (Fig. 4 and
Table II, compare Scenario 4 to Scenario 1), compact PHEV
market penetration only increased by 3.4% over 14 years. This
is because our agents were modeled after our survey
population, of which a surprising 18% of the survey
respondents indicated they would definitely, and another 39%
IEEE Access 2015-000125
9
stated that they might, consider switching car classes to
purchase a compact PHEV. On the one hand, these results
indicate the limited number of car classes in first generation
PHEVs may not, in itself, be a large barrier to PHEV market
penetration. On the other hand, they also imply that future
availability of PHEVs in more vehicle classes may not have a
huge impact on overall PHEV market penetration.
While we elected to use known attributes of the Chevy Volt
as representative of available PHEV technology throughout
each simulation, one could easily incorporate hypothetical
improvements in PHEV technology over time (e.g., as in [64])
and/or increased availability of PHEV models over time.
However, unless such changes succeed in dramatically
lowering the PHEV comfort thresholds of many consumers
(something we can only guess at), our results indicate that they
may have relatively little impact on predicted PHEV market
penetration. Thus, since we have no way of knowing what
those future changes might be or how they may affect
consumer purchasing decisions, we judged that adding such
assumptions (which are not founded on data) into the model
would introduce unnecessary complexity. We also assumed a
constant 2011 vehicle price throughout the simulations, a
standard assumption in engineering economics. In reality, one
would hope that as PHEV rebates are phased out there will be
a reduction in the MSRP of PHEVs relative to conventional
vehicles due to improvements in battery technology and
manufacturing and economy of scale.
Large increases in gasoline prices (Fig. 4 and Table II,
compare Scenarios 5 and 6 to Scenario 1) had surprisingly
small effects on PHEV market penetration in these
simulations. However, our model assumed that vehicle
consumers used rational computations of fuel costs savings
(such as those now easily obtained through various websites).
In reality, many consumers do not base vehicle purchasing
decisions on financially accurate assessments of alternatives
[23], so a more rapid rise in gasoline prices and/or higher
gasoline taxes, and associated press coverage, may cause a
disproportionately high psychological effect that stimulates
PHEV adoption at a greater rate than rational fuel cost savings
would warrant.
Our simulations illustrate the importance of governmental
and manufacturer rebates in making PHEVs competitive (Fig.
4 and Table II, compare Scenarios 7-10 to Scenario 1). Again,
we assumed that agents took all rebates into account when
making vehicle purchasing decisions. However, even when
these rebates are in place, many consumers may be unaware of
them, particularly manufacturer rebates (that change
frequently, vary with consumer credit ratings, and are often
poorly advertised), and state rebates (which many consumers
may not be familiar with). If PHEVs are to successfully
penetrate the U.S. transportation market to a large extent, it
may be necessary to continue and widely advertise such
rebates until the MSRP of PHEVs can become more cost-
competitive, unless improvements in PHEV technology and
consumer education efforts are successful in significantly
reducing consumers’ PHEV comfort thresholds.
V. CONCLUSIONS
Our prior PHEV market penetration work [30] left us
frustrated with the lack of sufficient data to properly inform
agent vehicle-purchasing decision rules or populate the ABM
with realistic distributions and cross-correlations of agent
attributes. For example, although we employed the well-
established sociological Gaussian distribution to model the
technology adoption lifecycle [66], we could not find data to
properly estimate the mean and standard deviation of this
distribution for new PHEV technology consumer adoption.
We also could not find quantitative data to help predict which
vehicle consumers were likely to seriously consider a PHEV,
especially if this meant switching vehicle class to obtain a first
generation PHEV. Additionally, even if consumers were
willing to consider purchasing a PHEV, we found little
information to help quantitatively predict how much extra they
might be willing to spend up front to obtain greater fuel
savings in the future.
To address these issues, we recently conducted and
analyzed an extensive consumer survey designed, in large
part, to specifically inform our PHEV market penetration
ABM [49]. In the current study, we modified our ABM to
reflect the survey results. Rather than generating a synthetic
population based on the statistical distributions and cross-
correlations found in the survey (a non-trivial task because of
the multitude of cross-correlated and non-linearly related
variables), we created an agent population with a one-to-one
correspondence between survey respondents and agents, with
each agent’s attributes based as directly as possible on the
responses of one survey participant, including a variety of
attitudes and susceptibility to social and media influences.
Survey questions were carefully phrased to tease out different
aspects of potential PHEV adoption barriers, including fear of
new PHEV technology independent of other vehicle features
or cost (and where consumers saw themselves on the PHEV
technology adoption lifecycle), willingness to pay more
upfront to achieve various savings in fuel costs down the road
(independent of the fuel type), attitudes regarding the
environment and energy, stated willingness to consider a first
generation PHEV with the understanding that they would only
come in compact models, etc. We then redesigned the ABM to
independently include these complementary aspects into agent
vehicle purchasing decisions. Our revised model includes a
PHEV-technology comfort threshold component, a previously
validated multinomial logistic prediction of willingness to
consider a compact PHEV [49], based on consumer attitudes
including environmental and energy concerns (which could
dynamically change in our model, subject to social and media
influences), and agent-specific delay discounting functions
that predict the amount agents are willing to pay up front for
greater fuel savings.
Some of the survey responses were contrary to our original
assumptions. For example, it is worth noting that, while our
survey results confirmed that the comfort threshold T could
reasonably be modeled with a truncated Gaussian distribution
(Fig. 1), the observed 48% standard deviation of this
distribution was much larger than the 20% standard deviation
assumed in our previous work [30]. The higher standard
deviation reflects many potential early adopters, but also
IEEE Access 2015-000125
10
implies that adoption rates will remain slow since many
consumer may not consider a PHEV until market penetration
is already quite high. Cross-correlations in many attributes
(e.g., Fig. 3b) were also lower than initially assumed in [30].
In this work we have avoided the necessity to make
assumptions on these distributions and cross-correlations by
modeling the attributes of individual agents on those reported
by individual survey respondents.
We report insights from several simulated scenarios. Our
results indicate the fear of new PHEV technology remains a
large barrier to widespread PHEV adoption among new car
buyers. To date, there has been relatively little advertising or
media attention designed to educate or allay consumer fears
regarding this technology. Since a growing majority of U.S.
vehicle purchasers elect to buy used rather than new cars, this
will greatly limit the rate of PHEV market penetration, since
PHEVs are not likely to become a large part of the used-car
market in the near future. Surprisingly, our agent population
was relatively insensitive to our restricting available PHEVs to
compact cars and was also relatively insensitive to rising
gasoline prices. However, our simulations confirm that
governmental and manufacturer rebates will remain necessary
to make PHEVs competitive until the sticker price of these
vehicles comes down.
ABM predictions will always be subject to a wide range of
uncertainties associated with various assumptions.
Nonetheless, since manufacturers and policy-makers are
required to make decisions in the face of such uncertainties,
models can be useful decision support tools, if adequately
grounded in data [67]. We believe that close integration of
consumer surveys and the design of agent-based models is a
key step in the development of useful decision-support models
in the transportation research community and beyond.
In their recent review of electric-drive vehicle market
penetration models, [37] recommend that there needs to be a
greater connection between consumer surveys and electric-
drive vehicle adoption rate modeling. This study illustrates
how one can make that connection.
ACKNOWLEDGEMENTS
This work was funded in part by the United States Department
of Transportation through the University of Vermont
Transportation Research Center and a workforce development
sub-award from Sandia National Laboratories supported by
the U.S. Dept. of Energy through Inter-Entity Work Order
M610000767. We thank D. Brad Lanute, Diann E. Gaalema,
Kiran Lakkaraju, and Christina E. Warrender for their
participation in the PHEV survey data collection. We also
thank cars.com (www.cars.com) for allowing us to
automatically scrape their website for MPG and MSRP values.
REFERENCES
[1] EPA, 2013. National Greenhouse Gas Emissions Data, United States
Environmental Protection Agency. http://www.epa.gov/climatechange/ghgemissions/usinventoryreport.html
. Accessed July 17, 2013.
[2] Duvall, M., E. Knipping, M. Alexander, L. Tonachel, C. Clark, 2007. Environmental Assessment of Plug-In Hybrid Electric Vehicles, Volume
1: Nationwide Greenhouse Gas Emissions ( No. 1015325). Electric
Power Research Institute, Palo Alto, CA.
[3] Jaramillo, P., C. Samaras, H. Wakeley, K. Meisterling, 2009.
Greenhouse gas implications of using coal for transportation: Life cycle assessment of coal-to-liquids, plug-in hybrids, and hydrogen pathways.
Energy Policy 37, 2689–2695.
[4] Smith, W.J. Plug-in hybrid electric vehicles—A low-carbon solution for Ireland? 2010. Energy Policy 38, 1485–1499.
[5] Lewis, A. M., J. C. Kelly, G. A. Keoleian, 2012. Evaluating the life
cycle greenhouse gas emissions from a lightweight plug-in hybrid electric vehicle in a regional context. In 2012 IEEE International
Symposium on Sustainable Systems and Technology (ISSST) 2012, 1–6.
[6] Samaras, C., K. Meisterling, 2008. Life Cycle Assessment of Greenhouse Gas Emissions from Plug-in Hybrid Vehicles: Implications
for Policy. Environmental Science & Technology 42, 3170–3176.
[7] Cowan, K.R., 2013. A New Roadmapping Technique for Creatively Managing the Emerging Smart Grid. Creativity and Innovation
Management 22, 67–83.
[8] Hernandez, L., C. Baladron, J. M. Aguiar, L. Calavia, B. Carro, A. Sanchez-Esguevillas, D. J. Cook, D. Chinarro, J. A. Gomez, 2012. Study
of the Relationship between Weather Variables and Electric Power
Demand inside a Smart Grid/Smart World Framework. Sensors 12, 11571–11591.
[9] Saber, A.Y., G. K. Venayagamoorthy, 2012. Resource Scheduling
Under Uncertainty in a Smart Grid with Renewables and Plug-in Vehicles. IEEE Systems Journal 6, 103–109.
[10] Connelly, M.C., J. A. Sekhar, 2012. U. S. energy production activity and
innovation. Technological Forecasting and Social Change 79, 30–46. [11] Klemes, J.J., P. S. Varbanov, D. Huisingh, 2012. Recent cleaner
production advances in process monitoring and optimisation. Journal of Cleaner Production 34, 1–8.
[12] Andersen, P.H., J. A. Mathews, M. Rask, 2009. Integrating private
transport into renewable energy policy: The strategy of creating intelligent recharging grids for electric vehicles. Energy Policy 37,
2481–2486.
[13] Andersson, S.-L., A. K. Elofsson, M. D. Galus, L. Göransson, S. Karlsson, F. Johnsson, G. Andersson, 2010. Plug-in hybrid electric
vehicles as regulating power providers: Case studies of Sweden and
Germany. Energy Policy 38, 2751–2762. [14] Offer, G.J., D. Howey, M. Contestabile, R. Clague, N. P. Brandon, 2010.
Comparative analysis of battery electric, hydrogen fuel cell and hybrid
vehicles in a future sustainable road transport system. Energy Policy 38, 24–29.
[15] Accenture, 2011. Plug-in electric vehicles: Changing perceptions,
hedging bets. Accenture end-consumer survey on the electrification of private transport. ACC11-0320/7-1792, 48 pp.
http://www.accenture.com/Microsites/accenturesmartsolutions-
electricvehicles/Documents/Accenture_Plug-in_Electric_Vehicle_Consumer_Perceptions_FINAL.PDF. Accessed
July 17, 2013.
[16] Calfee, J.E., 1985. Estimating the demand for electric automobiles using fully disaggregated probabilistic choice analysis. Transportation
Research Part B: Methodologies. 19, 287–301.
[17] Caperello, N.D., K. S. Kurani, 2012. Households’ Stories of Their Encounters With a Plug-In Hybrid Electric Vehicle. Environment and
Behavior 44, 493–508.
[18] Deloitte Consulting LLP, 2010. Gaining traction: A customer view of electric vehicle mass adoption in the U.S. automotive market.
http://www.deloitte.com.br/publicacoes/2007/MFG.Gaining_Traction_c
ustomer_view_of_electric_vehicle_mass_adoption.pdf. Accessed July 17, 2013.
[19] Hidrue, M.K., G. R. Parsons, W. Kempton, M. P. Gardner, 2011.
Willingness to pay for electric vehicles and their attributes. Resource and Energy Economics 33, 686–705.
[20] Musti, S., K. M. Kockelman, 2011. Evolution of the household vehicle
fleet: Anticipating fleet composition, PHEV adoption and GHG emissions in Austin, Texas. Transportation Research Part A: Policy and
Practice 45, 707–720.
[21] Sovacool, B.K., R. F. Hirsh, 2009. Beyond batteries: An examination of the benefits and barriers to plug-in hybrid electric vehicles (PHEVs) and
a vehicle-to-grid (V2G) transition. Energy Policy 37, 1095–1103.
[22] Zypryme Research and Consulting, 2010. The Electric Vehicle Study. http://www.zpryme.com/SmartGridInsights/The_Electric_Vehicle_Stud
y_Zpryme_Smart_Grid_Insights_Airbiquity_Sponsor_December_2010.
pdf. Accessed July 17, 2013. [23] Turrentine, T.S., K. S. Kurani, 2007. Car buyers and fuel economy?
Energy Policy 35, 1213–1223.
IEEE Access 2015-000125
11
[24] Woodall, B., P. Lienert, B. Klayman, 2012. Insight: GM’s Volt: The
ugly math of low sales, high costs. Reuters. http://www.reuters.com/article/2012/09/10/us-generalmotors-autos-volt-
idUSBRE88904J20120910. Accessed July 17, 2013.
[25] Valdes-Dapena, P., 2012. High price soured Chevy Volt sales. CNNMoney,
http://money.cnn.com/2012/03/05/autos/volt_sales_analysis/index.htm.
Accessed July 17, 2013. [26] Binny, M.P., K.M. Kochelman, and S. Musti, 2011. Evolution of the
Light-Duty Fleet: Anticipating Adoption of Plug-In Hybrid Electric
Vehicles and Greenhous Gas Emissions Across the U.S. Fleet. In Transportation Research Record: Journal of the Transportation
Research Board, No. 2252, Transportation Research Board of the
National Academies, Washington, D.C., 107-117. [27] Brown, M., 2013. Catching the PHEVer: Simulating Electric Vehicle
Diffusion with an Agent-Based Mixed Logit Model of Vehicle Choice.
Journal of Artificial Societies and Social Simulation. 16, http://jasss.soc.surrey.ac.uk/16/2/5.html.
[28] ElBanhawy, E, R. Dalton, E.M. Thompson, and R. Kotter, 2012. A
Heuristic Approach for Investigating the Integration of Electric Mobility Charging Infrastructure in Metropolitan Areas: An Agent-Based
Modeling Simulation. In Proceedings of the 32nd International
Symposium on Environment Friendly Energies and Applications (EFEA), 74-86.
[29] Elgowainy, A., Y. Zhou, A.D. Vyas, M. Mahalik, D. Santini, and M.
Wang, 2012. Impacts of Charging Choices for Plug-In Hybrid Electric Vehicles in 2030 Scenario. In Transportation Research Record: Journal
of the Transportation Research Board, No. 2287, Transportation Research Board of the National Academies, Washington, D.C., 9-17.
[30] Eppstein, M.J., D. K. Grover, J. S. Marshall, and D. M. Rizzo, 2011. An
agent-based model to study market penetration of plug-in hybrid electric vehicles. Energy Policy, 39, 3789-3802.
[31] Higgins A, P. Paevere, J. Gardner, G. Quezada, 2012. Combining choice
modeling and multi-criteria analysis for technology diffusion: An application to the uptake of electric vehicles. Technological Forecasting
and Social Change, 79, 1399-1412.
[32] McManus, W. and R. Senter, Jr., 2009. Market Models for Predicting PHEV Adoption and Diffusion. Transportation Research Institute
Report, UMTRI-2009-37, University of Michigan,
http://141.213.232.243/bitstream/handle/2027.42/64436/102399.pdf. Accessed July 17, 2013.
[33] Sullivan, J.L., I. T. Salmeen, and C. P. Simon, 2009. PHEV marketplace
penetration: An agent based simulation. Transportation Research Institute Report UMTRI-2009-32, University of Michigan,
http://hdl.handle.net/2027.42/63507. Accessed July 17, 2013.
[34] Shafiei, E., Thorkelsson, H, Asgeirsson, E.I., Davidsdottir, B., Raberto, M., and Stefansson, H., 2012. An agent-based modeling approach to
predict the evolution of market share of electric vehicles: A case study
from Iceland. Technological Forecasting and Social Change, 79, 1638-1653.
[35] Yabe, K., Y. Shinoda, T. Seki, H. Tanaka, and A. Akisawa, 2012.
Market penetration speed and effects on CO2 reduction of electric vehicles and plug-in hybrid electric vehicles in Japan. Energy Policy, 45,
529-540.
[36] Zhang, T., Gensler, S., and Garcia, R., 2011. A study of the diffusion of alternative fuel vehicles: an agent-based modeling approach. Journal of
Product Innovation Management, 28, 152–168.
[37] Al-Alawi, B.M. and Bradley, T.H., 2013. Review of hybrid, plug-in hybrid, and electric vehicle market modeling studies. Renewable and
Sustainable Energy Reviews, 21, 190-203.
[38] Daziano, R.A., and E. Chiew, 2012. Electric vehicles rising from the dead: Data needs for forecasting consumer response toward sustainable
energy sources in personal transportation. Energy Policy, 51, 876-894.
[39] Granovetter, M., 1978. Threshold models of collective behavior. American Journal of Sociology, 83 (6), 1420–1443.
[40] McPherson, M., L. Smith-Lovin, and J. M. Cook, 2001. Birds of a
feather: Homophily in Social Networks. Annual Review of Sociology, 27, 415-444.
[41] Axelrod, R., 1997. The Dissemination of Culture: A Model with Local
Convergence and Global Polarization. The Journal of Conflict Resolution, 40 (1), 203–226.
[42] Bednar, J., Page, S., 2007. Can game(s) theory explain culture?: The
emergence of cultural behavior within multiple games. Rationality and Society, 19(1), 65–97.
[43] Amazon Mechanical Turk. https://www.mturk.com. Accessed July 17,
2013. [44] Mason, W., Watts, D.J., 2009. Financial incentives and the
“performance of crowds”, in: Proceedings of the ACM SIGKDD
Workshop on Human Computation, HCOMP ’09. ACM, New York, NY, USA, pp. 77–85
[45] Buhrmester, M., Kwang, T., Gosling, S.D., 2011. Amazon’s Mechanical
Turk: A New Source of Inexpensive, Yet High-Quality, Data? Perspectives on Psychological Science 6, 3–5.
[46] Alonso, O., Mizzaro, S., 2012. Using crowdsourcing for TREC
relevance assessment. Information Processing & Management 48, 1053–1066.
[47] Gardner, R.M., Brown, D.L., Boice, R., 2012. Using Amazon’s
Mechanical Turk website to measure accuracy of body size estimation and body dissatisfaction. Body Image 9, 532–534.
[48] Rand, D.G., 2012. The promise of Mechanical Turk: How online labor
markets can help theorists run behavioral experiments. Journal of Theoretical Biology 299, 172–179.
[49] Krupa, J.S., D.M.Rizzo, M.J. Eppstein, D. B. Lanute, D. E. Gaalema, K.
Lakkaraju, and C.E. Warrender, 2014. Analysis of a consumer survey on plug-in hybrid electric vehicles. Transportation Research Part A: Policy
and Practice, 64, 14-31. Available online DOI:
10.1016/j.tra.2014.02.019. [50] NHTS, 2009. Summary of Travel Trends: 2009 National Household
Travel Survey. United States Department of Transportation, Federal
Highway Administration. http://nhts.ornl.gov/2009/pub/stt.pdf. 2009. [51] Ozaki, R., Sevastyanova, K., 2011. Going hybrid: An analysis of
consumer purchase motivations. Energy Policy 39, 2217–2227. [52] Graham-Rowe, E., Gardner, B., Abraham, C., Skippon, S., Dittmar, H.,
Hutchins, R., Stannard, J., 2012. Mainstream consumers driving plug-in
battery-electric and plug-in hybrid electric cars: A qualitative analysis of responses and evaluations. Transportation Research Part A: Policy and
Practice 46, 140–153.
[53] Ziegler, A, 2012. Individual characteristics and stated preferences for alternative energy sources and propulsion technologies in vehicles: A
discrete choice analysis for Germany. Transportation Research Part A:
Policy and Practice, 46, 1372-1385. [54] Schuitema, G., Anable, J., Skippon, S., Kinnear, N, 2013. The role of
instrumental, hedonic and symbolic attributes in the intention to adopt
electric vehicles. Transportation Research Part A: Policy and Practice, 48, 39-49.
[55] The PHEV survey data is available at:
http://www.cs.uvm.edu/~meppstei/PHEVsurvey [56] www.cars.com. Accessed July 17, 2013.
[57] Kohonen, T., 1990. The self-organizing map, In Proceedings of the
IEEE, 78(9), 1464-1480. [58] EIA, 2013. Annual Energy Outlook 2013 Early Release Overview.
United States Energy Information Administration. DOE/EIA-0383ER.
[59] DOE, 2013a. Vehicle Cost Calculator. United States Department of Energy, Alternative Fuels Data Center.
http://www.afdc.energy.gov/calc/. Accessed July 17, 2013.
[60] DOE, 2013b. My Plug-in Hybrid Calculator. United States Department of Energy. http://www.fueleconomy.gov. Accessed July 17, 2013.
[61] MSN Money, 2013. Car Loan Calculator. http://money.msn.com/car-
buying/car-loan-calculator. Accessed July 17, 2013. [62] Ingram, Antony, 2014. Which state has the highest percentage of electric
cars? No, not CA.
http://www.greencarreports.com/news/1090040_which-state-has-the-highest-percentage-of-electric-cars-no-not-ca. Accessed March 26, 2015.
[63] Watts, D., 2002. A simple model of global cascades on random
networks. In Proceedings of the National Academy of Sciences, 99(9), 5766–5771.
[64] Bockarjova, M., Knockaert, J., Rietveld, P., and L. Steg. 2014. Dynamic
consumer heterogeneity in electric vehicle adoption. Transportation Research Board Annual Meeting, paper 14-1597.
[65] Taylor, P, 2012. NADA DATA 2012: State-of-the-industry report.
http://www.nada.org/NR/rdonlyres/C1C58F5A-BE0E-4E1A-9B56-1C3025B5B452/0/NADADATA2012Final.pdf. Accessed July 17, 2013.
[66] Beal, G. M., E. M. Rogers, and J. M. Bohlen, 1957. Validity of the
concept of stages in the adoption process. Rural Sociology, 22(2), 166–168.
[67] Bastani, P., Heywood, J.B., and Hope, C. 2012. The effect of uncertainty
on US transport-related GHG emissions and fuel consumption out to 2050. Transportation Research Part A: Policy and Practice, 46, 517-
548.
IEEE Access 2015-000125
12
Margaret J. Eppstein is Professor and
Chair of the Department of Computer
Science at the University of Vermont.
She received the B.S. degree in zoology
from Michigan State University in
1978, and the M.S. degree in computer
science and the Ph.D. degree in
environmental engineering from the
University of Vermont in 1983 and
1997, respectively. She was the founding Director of the
Vermont Complex Systems Center from 2006–2010 and
became Chair of Computer Science in 2012. Her research
interests involve forward and inverse modeling and analysis of
complex systems in a wide variety of application domains,
including biological, environmental, technological, and social
systems.
Donna M. Rizzo, Professor in the
School of Engineering, holds
undergraduate degrees in Civil
Engineering from the University of
Connecticut, Fine Arts from the
University of Florence, Italy and a M.S.
and Ph.D. from the University of
California, Irvine, and University of
Vermont, respectively. Her research
focuses on the development of new
computational tools to improve the understanding of human-
induced changes on natural systems and the way we make
decisions about natural resources.
Brian H.Y. Lee is an Assistant Professor in Transportation
Systems at the University of Vermont; his primary
appointment is in the School of Engineering, Civil and
Environmental Engineering Program, and his secondary
appointment is in the Transportation Research Center. He
received his Doctorate degree from the
University of Washington in the
Interdisciplinary Ph.D. Program in Urban
Design and Planning in 2009, his M.S.
degree from Northwestern University in
Civil and Environmental Engineering in
2003, and his Bachelor of Applied
Science degree from the University of
British Columbia in Civil Engineering in
2001. He is an applied researcher with expertise in
transportation and land use analysis and empirical modeling.
He has keen interests in multimodal transport and a record of
interdisciplinary collaborations on policy-relevant projects.
Joseph S. Krupa earned his B.S. in Civil
Engineering and his MS in Civil &
Environmental Engineering at the
University of Vermont in 2009 and 2013,
respectively. He was a Municipal Civil
Engineer for the City of New Haven, CT,
from 2013-2014; since 2014 he is an
Engineer at Roberge Associates Coastal
Engineers, LLC, in Stratford, CT.
Narine Manukyan received a BS in
Applied Mathematics and Informatics at
Yerevan State University in Armenia in
2008, MS in Computer Science at the
University of Vermont in 2011, and PhD
in Computer Science at the University of
Vermont in 2014. Her research interests
include artificial intelligence,
evolutionary computation, complex
networks, data mining, machine learning
and modeling complex systems. Starting in 2015, she became
Lead Data Scientist at Faraday, Inc., Burlington, VT.