WP2012/02

UCD School of Sociology Working Papers often represent preliminary work and are circulated to encourage discussion. Citing without permission from author(s) is prohibited. Any opinions expressed here are those of the author(s) and not those of the UCD School of Sociology, UCD Social Research Centre or UCD Geary Institute. This paper can be downloaded from www.ucd.ie/sociology/research

2

Abstract

A research puzzle for EU scholars interested in decision making is the repeated pattern

of co-operation in the Council of Ministers. There is much evidence to suggest that there

exists a strong norm of cooperation in the EU, which seems to guide the collective EU

decision process and in particular restricts the EU member states from engaging in

bilateral deals with each other to the detriment of other EU member states and the Union

as a whole. Why do individual member states actually comply with this EU norm and

moreover how does this norm operate as a mechanism of cooperation across the

member states in the EU decision process? We argue the norm of cooperation between

member states is sustainable, because the configuration of members’ positions and

interests gives rise to a decision situation that resembles a repeated Prisoner’s Dilemma

(PD). It is rational for individual member states to comply since not doing so would mean

large forgone gains in the future. We test this by making use of two current models of

collective decision making: the Position Exchange Model (PEM) and the Externalities

Exchange Model (EEM). The results suggest that the predictive power of PEM should

vary inversely with the average proportion of ‘winners’ in the data set. More winners

implies that for actors the shadow of the future becomes smaller, since there will be

fewer occasions on which they will actually be in a PD payoff structure.

3

1. Introduction

There are now a variety of different EU studies which suggest that decision

making in the Council of Ministers is characterised by a strong norm of cooperation.

Some studies suggest that the member states will work extensively together to achieve

unanimous agreement with the final collective decisions taken, even where the formal

decision rule procedure allows for a qualified majority vote (Heritier, 1999 ; Richardson,

2001 ; Jönsson C.et al, 1998 ; Lewis, 2003; Schneider et al, 2006). Examples are

provided by different case studies which give in-depth, detailed descriptions of the EU

decision process and map the extended bargaining activity at work in the EU committees

surrounding the Council of Ministers, as well as the European Commission (Wessels,

1998; Westlake, Galloway et al., 2006; Lewis, 2000). Other EU studies suggest there are

a variety of accepted practices, informal rules or “culture of compromise” which

permeate the EU decision process and the participants involved therein (Hayes-

Renshaw et al, 2006; Lewis, 1998; 2005; Heisenberg, 2005). One can imagine that

given the range of different member states of different sizes, resources and sectoral

interests, there will be variation in the level of compliance with the EU norm of

cooperation. In this research we ask: why do member states choose to conform to the

EU norm of cooperation, how does this norm operate as a mechanism of cooperation

across the member states in the EU decision process, and are there conditions when we

can expect to find that member states do not comply with this EU norm of co-operation?

In order to be able to address these research questions, we need to examine the

different decision processes and conditions for cooperative EU decision making.

Traditionally EU scholars have focused their attention on larger scale phenomena such

as understanding the varying trends over time towards European regional integration or

key institutional developments within the EU, such as the Treaty negotiations for

4

example. Research within the neo-functionalist approach has been concentrated on

determining the main drivers of ever closer integration, pointing to the role of

supranational institutions and the logic of spillover effects from one EU policy area to

another. Another major school of EU research, intergovernmentalism, highlights that it is

the member states that control the pace of integration and development of EU policy

competences but that the member states will pursue their domestic interests first above

international or supranational institutions. Given their research focus, it is not surprising

that these two core EU research approaches and others are not well adapted for the

study of processes and conditions of decision making for specific day-to-day issues.

However, the neo-institutionalist perspective is useful in this regard as it

examines the way in which institutions pervade human behavior through norms, rules

and cultural practices (Lewis, 2010; Tallberg, 2010).

Within neo-institutionalism, there co-exist different strands such as sociological,

cognitive, rational choice and historical institutionalism. Rational choice institutionalism is

most relevant for the research presented in this paper. Unlike traditional rational choice,

this variant of the approach begins with the assumption that an actor’s choices are

constrained (i.e ‘bounded rationality’). In this approach the core assumption is that while

actors behave strategically to maximize their own benefits, they are also assumed to

realize that their own goals are best achieved through institutions, which are understood

as systems of rules and inducements to behavior. Generally the rational choice,

institutionalist perspective is combined with a formal deductive modeling approach.

Combining this deductive modeling approach with real decision data allows contrasting

model assumptions about the collective decision process to be set out explicitly, as well

as carefully tested as to their veracity in predicting real decision outcomes (Thomson

and Hosli, 2006). Moreover, within the rational choice field, it is cooperationve game

5

theory, and in particular the focus on bargaining and vote-trading, which provide the key

theoretical basis for the assumptions of collective decision models described below

(Achen, 2006).

This paper begins with the recent empirical research conducted by Arregui,

Stokman and Thomson (2006), who applying the rational choice institutionalist

perspective, compare two types of formal decision models based on contrasting

assumptions about the EU bargaining process. The input data for both kinds of models

is the same: actor policy positions, salience attached to decision issues and actor

resources. The cooperative models, including the Van den Bos Compromise model

(1991) and the Stokman and Van Oosten Position Exchange model (1994), assume that

actors reach collective decisions by making binding decisions, which cannot be reneged

on. In the Van den Bos Compromise model (1991), EU decision making is assumed a

simple, cooperative process involving mutual persuasion or influence amongst the

member states, including a “special role for the Presidency of the Council, probably in

collaboration with the Commission” where strong pressure is exerted to reach a decision

outcome that is acceptable to all (Arregui et al. 2006: pages 131-132). In this model, the

assumptions about the EU decision process are kept very simple: disagreement

amongst the member states is the least favorite option compared with any of the other

alternatives available and as no further assumptions are made explicit in this decision

model about an intervening process prior to the collective decision being taken, the

collective decision outcome is predicted as a simple compromise, which is the weighted

average of member states policy positions. Hence, the model predicts that the collective

decision is reached when all member states’ positions are taken account of, and where

each member state position is weighted by the resources that the member state can

bring to bear on the negotiations and the salience they attach to the issues (Van den

Bos, 1991)

6

The second cooperative model is the Stokman and Van Oosten Position

Exchange model (PEM) (1994). This model builds some more complexity and realism

into the simple compromise model by making a further assumption about the EU

decision process. In the PEM, the EU decision process is assumed to have two stages,

wherein the member states try to increase support for their most salient issues by

bargaining with other member states. In the first stage, a very simple ‘logrolling’ process

is elaborated where a member state is allowed, once-off, to exchange issue support (i.e.

voting positions) with another member state across two issues and this exchange

bargain is assumed to be binding on both member states. In the model, it is also

assumed that this exchange may occur only when the two member states attach

different levels of salience to each of the two issues, as well as holding different policy

positions on each issue. In PEM, the second stage is arrived at when no further

exchanges are possible and compromise is assumed to take place on the basis of the

new voting positions, using the Compromise model solution1.

The non-cooperative model, referred to as the Challenge model (Bueno de

Mesquita, 1994; 2002), assumes that member states try to build a coalition in support of

their own policy positions, by challenging other members’ opposing positions and if

possible, compelling them to shift their positions closer to that of the challenger. In this

model, EU bargaining is viewed as a conflict-based process where member states are

assumed not to be bound by the commitments they make to others during the

negotiations. Unlike the cooperative models, a member state is not concerned with what

benefit others may gain in this process and “power dominance matters more than

convincing arguments according to this mode of conception of political bargaining”

(Arregui et al., page 127).

1 In this respect, according to the Compromise model, the predicted decision outcome on each issue is the

average of all member states voting positions on that issue, where each member state’s position is weighted

by the member state’s resources and the level of salience the member state attach to that issue.

7

Using a large-scale, representative data set of routine EU decisions, Arregui et

al. (2006) tested the cooperative and the non-cooperative models and showed that their

cooperative decision models generated the most accurate model predictions of the EU

decision outcomes. Given the numerous case studies describing the cooperative nature

of the EU decision process, these findings are not too surprising. However it is surprising

that in the Arregui et al. (2006) study, it is the Compromise Model that is overall more

successful, although not statistically better than the PEM, as a predictor of the EU

decision outcomes and this is despite its’ very simple and highly unrealistic model

assumptions.

We argue in this paper that the limited success of the PEM reflects the model

assumptions about the exchange process in the EU negotiations. The PEM does not

require any pair of actors, involved in an exchange of voting positions, to take into

account the potentially negative impact (i.e. negative externalities) of their agreement for

all other actors. But in fact, as Arregui et al. (2006) point out in their study, there are a

high number of instances where there are negative externalities after exchange in their

EU data set (2006). In these instances, the PEM does not seem applicable as a decision

model for the EU negotiation process. However there are some other instances in the

Arregui et al. (2006) study where the PEM is the most successful model at predicting the

decision outcomes, namely where there are highly polarized issues or where the

negative externalities from exchange are lower (Arregui et al, 2006).

In 2008, Dijkstra et al. elaborated a new cooperative model, called the

Externalities Exchange Model (EEM). This model assumes that the actors take account

of the potential for negative externalities for other actors. Moreover the actors can

choose different strategies to restrict the deals they make with other actors, so as to

minimize the potential for negative externalities for others accordingly. The EEM model

8

was applied and tested using the same EU data set, as used in the original Arregui et al.

(2006) study, and was significantly better than the PEM at generating more accurate

predictions of the EU decision outcomes (Dijkstra et al., 2008).

Comparing the results of these cooperative models and in particular the PEM

and EEM (Arregui et al, 2006; Dijkstra et al, 2008) suggests that under alternate

conditions, the bargaining strategy of EU member states will vary and moreover, under

these alternate conditions, member states may choose to comply with or to defect from

the EU norm of cooperation. As mentioned earlier, cooperative game theory provides the

theoretical basis for the elaboration of the various decision model assumptions

discussed in the paper, Extending the application of this cooperative game theory

approach, this paper aims to provide a rigorous study of the conditions in EU collective

decision making under which actors will adhere to the norm of cooperation. In this

regard, we argue that Axelrod’s seminal work on the ‘durable, iterated Prisoner’s

dilemma’ wherein he elaborates the conditions for emergence and durability of

cooperative behavior, is of particular relevance for addressing our research problem and

the development of our research hypotheses (Axelrod, 1984).

In the present paper we argue that the norm of cooperation between member

states exists and is sustainable, because the configuration of members’ positions and

interests gives rise to a decision situation that resembles an n-player infinitely repeated

Prisoner’s Dilemma Game. In fact, it is exactly the negative externalities already

mentioned by Arregui et al. (2006) that give rise to this Prisoner’s Dilemma. We thus

argue that in a repeated setting that contains large negative externalities it is actually

rational for individual member states to comply with the norm of cooperation, since not

doing so would mean suffering from those negative externalities large forgone gains in

the future. We make the argument precise and test its implications by making use of two

9

current models of collective decision making, the Position Exchange Model (PEM) of

Stokman and van Oosten (1994) and the Externalities Exchange Model of Dijkstra et al.

(2008).

The remainder of the paper is organized as follows. In the next section we

expound our theoretical argument and derive our hypotheses. The subsequent section

describes the data used to test the hypotheses, the results of which are discussed in the

section after that. The paper is concluded with a discussion of the results and their

implications.

2. Theory and Hypotheses

The ‘norm of cooperation’ described in the introduction means that (groups of) EU

member states pass up the realization of profitable deals, if these deals harm other EU

members not involved in them. In doing so, member states thus forgo opportunities of

increasing their utility. The question now arises why member states of the EU, would

adhere to such a norm that apparently runs counter to their self-interests. To answer this

question we take a rational choice stance, assuming that actors in the long run do seek

to maximize their self-interests. From this perspective actors, adhering to the norm of

cooperation, abstain from the realization of short-term profits only in order to gain

something more valuable (or to prevent losses) in the future.

2.1. EU decision making and Cooperation Theory

According to Axelrod’s Cooperation Theory (CT; Axelrod 1984) adherence to a norm

of cooperation such as described above can be sustained if at least two criteria are met,

namely: (i) the interaction is characterized by a prisoner’s dilemma-like (PD) payoff

10

structure, and (ii) the interaction horizon is sufficiently long. Thus we will argue in this

section that EU decision making resembles an n-person infinitely repeated PD game.

Requirement (ii) of CT means that there must be a sufficiently high probability that

actors will interact repeatedly in the future. It is self-evident that this is characteristic of

EU decision making processes, where every member country expects to remain a

member in the foreseeable future. The way in which requirement (i) is applicable to EU

decision making is less self-evident and is the topic of this section.

A social situation has a PD-like payoff structure when individually rational behavior

leads to collectively undesirable outcomes (Dawes 1980). Thus, the basic property of

such situations is that if all actors pursue their self-interests (called ‘defection’ in this

context) all will be worse off compared to the situation where actors abstain from the

pursuit of their self-interests (called ‘cooperation’ in this context). The problem however

is that, given what others do, any individual actor is always better off pursuing his self-

interests. Hence a social dilemma.

The principal argument of Cooperation Theory (Axelrod 1984) is that cooperative

behavior (i.e., actors refraining from realizing their immediate self-interests) can

nonetheless be sustained, if the shadow of the future (i.e., the likelihood that the same

actors will repeatedly interact again) is long enough. The idea is that all actors can

threaten to retaliate an ‘act of selfishness’ of any individual actor (i.e., this actor pursuing

his immediate self-interest to the detriment of the other actors) by withholding their

cooperation with that particular actor (or with every actor for that matter) in future

interactions. Moreover, since pursuing their short-term self-interests is itself an

equilibrium strategy in the repeated game (i.e., it is the best thing any actor can do when

everybody else does it too), this threat to withhold future cooperation is credible (e.g.,

Fudenberg and Tirole 1991). The credibility of this threat is exactly what makes it

11

effective: no actor will want to break the cooperative pattern since doing so will destroy

cooperation forever. And with no cooperation everybody, including the actor

contemplating breaking the cooperative pattern, is worse off.

In order to apply CT to EU decision making we need to define exactly what it means

for member states to defect or to cooperate. This is no easy task, since EU member

states typically have a multitude of strategies at their disposal and it is not

straightforward to label any of these ‘defective’ or ‘cooperative’. Thus, we need to make

assumptions concerning the strategy spaces of the EU member states, to make the

problem tractable.

In order to do so, we first make the assumption that EU decision making is

fundamentally an exchange process. Thus, we assume that eventual agreement among

EU member states is reached through reciprocal shifts of positions held by these

member states on different issues. Other decision strategies such as persuasion or

seeking confrontation are excluded from consideration. Second, we assume that these

exchanges take place only on pairs of issues, i.e., we assume member states

contemplate only exchanges on pairs of issues. These assumptions lead us to the

consideration of two models of collective decision making called the Position Exchange

Model (PEM; Stokman and van Oosten 1994; Thomson et al. 2006)2 and the

Externalities Exchange Model (EEM; Dijkstra et al. 2008). We will argue that we can use

the PEM as a model for the situation where every member state defects, while the EEM

models the situation where all member states cooperate.

The PEM and EEM model collective decision making as decision making about

controversial issues with single peaked preference functions, as most well-known

models do (Black 1958; Bueno de Mesquita, Newman & Ravushka 1985; Bueno de

2 For a short overview of the application of various bargaining models, including PEM, to EU

decision making, see Sullivan and Selck (2007).

12

Mesquita & Stokman 1994; Steunenberg 1994; Tsebelis & Garrett 1996, and many

others). Decision making may well require simultaneous decisions on several issues. In

these cases, different issues should represent independent controversial elements of the

decision making situation and as a set should cover the full range of possible outcomes.

The dynamics in the decision making process result from actors, with different intensity

and potential, trying to realize their preferred outcome on an issue (their initial position),

whereas per issue only one outcome that is binding for all actors can be chosen. In a

complex situation, possibly involving many actors and issues, actors will try to build a

coalition as large as possible behind their initial positions or behind a position that is as

close as possible to this. This informal bargaining process can be envisaged as

proceeding formal decision making and affecting the final positions of the actors in the

decision making, aiming at a collective outcome that reflects their interests as much as

possible.

The PEM and the EEM are both models of exchange or logrolling of this informal

bargaining process. The PEM models decision makers as only concerned with their own

immediate welfare, implying they will realize any immediate opportunity for gain,

regardless of the consequences for others or for themselves in the future. The EEM

models decision makers as concerned with the welfare of all: actors will only realize a

personal gain if this does not hurt any other actor.

The PEM and EEM assume the same structure of collective decision making. It is

assumed that there exists a finite set M of controversial issues, which can each be

represented by a one-dimensional interval scale. These issues are mutually exclusive

and exhaustive, i.e., an actor can take a position on one issue, irrespective of his

position on other issues (mutual exclusiveness), and the issues together cover the entire

collective decision problem (exhaustiveness). It is further assumed that each actor n

13

from the finite set of actors N, takes a position, nmx , on the scale of each issue m,

representing n’s most preferred outcome of m. Furthermore, each actor n is assumed to

have a salience, nms , for each issue m, expressing the relative importance of issue m to

the actor n. Finally, each actor n has a capability, nc , reflecting n’s potential to affect the

final outcome of each of the issues in M. The actors’ positions, saliences and capabilities

are assumed to be common knowledge. Based on this common knowledge, all actors

are supposed to have a common expected outcome, mO , of each issue m. In both the

PEM and EEM, mO is assumed to be the weighted average of the actors’ positions, with

weights equal to the actors’ capabilities times their saliences, as in equation (1) below:

n

nmn

n

nmnmn

msc

xsc

O (1).

2.2. The PEM and Defection

The basic idea of the PEM is that pairs of actors can mutually increase their utilities

compared to their utilities of the expected outcome in (1) by exchanging their positions

on pairs of issues. The PEM assumes that actors have single-peaked preferences: an

actor’s initial position on an issue represents his preferred outcome, and any deviation of

the final outcome from it, is evaluated as strictly worse. In the PEM the utility of actor n (

nU ) over the outcomes of all the issues in M is assumed to be:

Mm

mnmnmn OxsU || (2).

14

Equation (2) shows that an actor’s utility is assumed to be (i) additive over all issues, and

(ii) decreasing linearly in the absolute distance of the outcome from the actor’s position,

with the salience of the issue determining the rate of decrease.

***

Figure 1

***

In the PEM, two actors are assumed to be able to exchange on a pair of issues only

if they have positions on opposing sides of the expected outcomes on both issues. With

two issues, and their expected outcomes, we can partition the set of actors into four

groups, A, B, C, and D, as is shown in Figure 1. Members of group A are on the left

hand side of the expected outcomes on the interval scales of both issues, those of group

D are on the right side of both issues. Members of group B are on the left hand side of

the expected outcome on issue 1, and on the right hand side on issue 2, with members

of C having opposite positions. From this it immediately follows that according to the

PEM members of A can only exchange with members of D, and members of B can only

exchange with members of C.

Exchange between two actors is mutually profitable only if the actors have different

relative saliences for the two relevant issues. Without loss of generality, consider two

actors, i and j, and two issues, 1 and 2. Assume i and j are on opposite sides of the

expected outcomes of issues 1 and 2. Denote the changes in the expected outcomes on

issues 1 and 2, caused by position shifts of actors i and j, as 1 and 2 , respectively.

Then i and j can only exchange profitably if either (3) or (4) is true.

2

1

1

2

2

1

j

j

i

i

s

s

s

s

(3)

15

2

1

1

2

2

1

j

j

i

i

s

s

s

s

(4)

Equations (3) and (4) show that exchange is only mutually profitable if the exchange

ratio (1

2

) is in between the relative saliences (see Dijkstra et al. 2008 for a proof).

If (3) holds, i shifts his position on issue 1 in the direction of j, whereas j shifts his

position on issue 2 in the direction of i. Issue 1 is then called the supply issue of i and the

demand issue of j, whereas issue 2 is the demand issue of i and the supply issue of j. If

(4) holds, issue 2 is the supply issue of i and issue 1 is the supply issue of j. The latter

situation is depicted in Figure 1.

In the PEM all possible bilateral exchanges are determined for each pair of issues

from M. For each of these exchanges, position shifts are determined such that the utility

gains of the exchange partners are equal and at a maximum. The exchanges are then

listed in the order of the size of the utility gains. The exchange with the highest utility

gains is then executed, and all other possible exchanges involving one or both of the

partners of this first exchange, and in which these partners use the same supply issues

as in this first exchange, are deleted from the list. This process is then repeated with the

remaining exchanges on the list, until the list is empty. Then, (1) is applied to all issues

with the new actor positions, and these are the predictions of the PEM. See Stokman

and Van Oosten (1994) for details about this algorithm.

Thus, the PEM is based on the assumption that pairs of actors engage in bilateral

position exchanges on pairs of issues when such an exchange is mutually profitable for

the two actors involved. However, equations (1) and (2) immediately show that an

exchange between any pair of actors will affect the utility of all actors, not just of the

actors involved in the exchange. To see this, note that position shifts of the two

16

exchanging actors will affect the expected outcome, according to (1). According to (2)

this will affect the utility of all actors, whether or not they are a partner to the exchange.

Such utility effects for actors not directly (i.e., as ‘exchange partners’) involved in the

bilateral exchange are called externalities, and can be either positive or negative. In the

case of positive externalities, bilateral deals cause decision outcomes to become better

for other actors not involved in the deal, whereas in the case of negative externalities,

decision outcomes become worse for actors not involved in the deal. Moreover, one

particular bilateral deal can have positive externalities for some actors and negative

externalities for others.

Once more consider Figure 1. It shows how the position exchange of actors i and j

causes the expected outcome (according to equation (1)) to shift away from the

positions of actors in group C on both issues. Thus, actors in this group experience

negative externalities from the exchange between i and j. Actors in group B, however,

experience positive externalities from this exchange, since the expected outcomes on

both issues shift toward their positions. For actors in groups A and D externalities can be

either positive or negative, depending on how their relative saliences compare to the

exchange rate that i and j agree on (see equations (3) and (4)). See Van Assen et al.

(2003) and Dijkstra et al. (2008) for more elaborate discussions of externalities in

collective decision making.

What is central to the current paper is that the PEM models decision making as a

process of bilateral exchanges on pairs of issues, without taking into account the positive

or negative externalities for other actors. Thus, when actors in the PEM contemplate the

desirability of a potential exchange, they are assumed to only consider their own

immediate utility gains from this exchange, and to disregard any effects for others.

17

Now consider a decision situation in which the negative externalities caused by

bilateral exchanges are so large that, when all bilateral exchanges are consumated

according to the PEM, they outweigh the benefits of exchange for any single actor. In

this case, all actors in the decision situation would have been better off if no bilateral

exchanges had taken place. Thus, the situation resulting from the bilateral exchange

procedure of the PEM is Pareto inferior to the situation in which no bilateral exchanges

take place. However, from the perspective of any single actor in the PEM, completing a

profitable exchange is always better than not completing it, regardless of what others do.

Hence, in a situation with large negative externalities for all actors, we argue that actor

behaviour according to the PEM resembles the strategy profile where all players defect

in the PD.

2.3. The EEM and Cooperation

In the EEM externalities are taken into account. Dijkstra et al. (2008) introduce two

variants of the EEM. In the first (labeled EEMb) actors block exchanges that cause

negative externalities for fellow-members of their ‘group’. The groups considered by the

EEMb are the groups labeled A through D in Figure 1. In the second version (labeled

EEMb&w) actors block all exchanges that entail any negative externality for any actor, no

matter that actor’s group membership.

In the current paper we only consider the latter version of the EEM and simply refer

to it as the EEM. We make this choice because it best reflects the generality of the norm

of cooperation we try to understand. As indicated in the introduction, this norm is

assumed to imply that actors generally try to prevent negative externalities for any other

actor, not just for actors with whom they happen to agree on the two issues under

consideration (i.e., their fellow members of group A, B, C or D).

18

Since the EEM assumes all actors try to prevent negative externalities for others, it

does not explicitly model position shifts of individual actors, but directly models the shifts

in the expected outcomes of the pair of issues. In essence, the EEM assumes that on

each pair of issues actors collectively look for changes in the expected outcomes such

that the utility of no actor is decreased and the utility of at least one actor is enhanced. In

other words, the EEM looks for Pareto improvements with respect to the initially

expected outcome. If more than one such improvement exists, actors choose the

outcome specified by the Generalized Nash Bargaining Solution (GNBS; see Chae and

Heidhues 2004).

In more detail, the EEM model works as follows. In the EEM, equation (1) is

computed for each issue and is taken as this issue’s initially expected outcome (just as

the PEM does). Then, for all possible pairs of issues, alternative outcomes are sought

that constitute a Pareto efficient outcome for all actors. That is, in the EEM only outcome

shifts on pairs of issues are considered that for all actors yield at least as much utility

(according to equation (2)) as the initially expected outcome. In addition, the alternative

outcome cannot be improved for any actor by further shifting on one or both of the

issues, without decreasing the utility of another actor. Thus, in the EEM negative

externalities (compared to the initially expected outcome) for any actor is prohibited, but

positive externalities are allowed.

If there is more than one Pareto efficient alternative to the initially expected outcome

the GNBS is used to find a single prediction. Letting denote the pair ),( 21 of shifts

on issues 1 and 2,respectively, the GNBS is the value of that maximizes the weighted

product of utility gains. More formally,

Nn

r

nP

n)](U[Max (5).

19

The EEM takes as weights the capability of the actors relative to the summed capability

of all the actors. Formally, the weight assigned to actor n’s utility in (5), nr , is:

Ni

i

nn

c

cr (6),

If there exist no Pareto improvements with respect to the initially expected outcomes,

the EEM simply identifies the initially expected outcomes as it’s solution.

With more than 2 issues involved, the EEM implements the following voting

procedure.

(i) Compute (1) for all issues

(ii) Compute the prediction of the EEM for all M(M-1) exchange possibilities

(iii) Actors vote for their most preferred exchange opportunity

(iv) Select from the list of (remaining) issue pairs the one with the highest weighted

votes

(v) Eliminate all issue pairs from the list containing one of the two issues on which

the exchange in (iv) took place

(vi) If the list is not empty after (v), go back to step (iv)

In step (iii) the EEM assumes that each actor votes for that exchange opportunity in the

list that yields him the largest positive utility change. Hence the EEM excludes strategic

voting. In EEM’s voting procedure, an actor’s vote is weighted by the capability of the

actor, relative to the sum of capabilities of all actors in N. The exchange with the highest

sum of weighted votes is executed first. Actors vote only once, at the beginning of the

process. If there is a tie, one issue pair is selected at random. In the data analyzed in

20

this paper, such ties did not occur. See Dijkstra et al. (2008) for the details of this

algorithm.

Now consider again the decision situation in which the negative externalities caused

by bilateral exchanges are so large that, when all bilateral exchanges are consumated

according to the PEM, they outweigh the benefits of exchange for any single actor. We

previously saw how actor behaviour according to the PEM can then be conceived as the

PD strategy profile ‘all defect’, with the associated Pareto inefficient outcome. When

applying the EEM to this situation, no negative externalities exist, and outcomes are

Pareto efficient. Thus, when the actors choose to make their decisions according to the

EEM process, rather than according to the PEM process, this can be interpreted as

actors playing the PD strategy profile ‘all cooperate’. Note that, even if all other actors

choose to ‘comply with the EEM’ (i.e., cooperate), bilateral exchanges according to the

PEM will in general still be possible and will moreover by profitable for the two actors

involved. In other words, when all others cooperate, any pair of potential exchange

partners experiences the temptation to defect.

2.4. Hypotheses

In the previous sections we have theoretically constrained the strategy spaces of the

EU member states to only include exchange processes (either bilateral or group-wise)

concerning pairs of issues. Subsequently we have argued that, in a situation where

negative externalities of bilateral exchanges are very large, the PEM and the EEM

predictions can be regarded as the outcomes of strategy profiles where all actors defect

or cooperate, respectively. With these models in hand we can make requirement (i) of

Cooperation Theory, that the interaction situations have a PD-like payoff structure, more

precise.

21

First, observe that all the individual bilateral exchanges modeled by the PEM are

mutually profitable for the two exchanging actors. Thus, as required by a PD payoff

structure, completing the available exchanges is always the best thing an actor can do in

the short run, regardless of what the other actors do (i.e. whether or not they comply with

the norm of cooperation). However, when all actors follow their immediate self-interests

(i.e., exchange according to the PEM) this might entail large negative externalities for all

actors. The decision situation would resemble a PD payoff structure if these negative

externalities would override the utility gains the actors reap from their own exchanges

and possibly from positive externalities, compared to the situation where actors

abstained from pursuing their short-term private interests. The latter situation (adherence

to the norm of cooperation) is modeled through the EEM. Thus, a PD-like payoff

structure would exist if all actors would be better off (i.e., receive higher utilities

according to equation (2)) when the decision making process followed the EEM

procedure than when it followed the PEM procedure. Indeed, using the same data set

that we use in the current paper and comparing the outcomes predicted by the PEM to

the initially expected outcome of equation (1), Arregui et al. (2006) find that for all

proposals of the European Commission, and summed over all member states negative

externalities of exchange outweigh the sum of gains from exchange and positive

externalities.

In various studies discussed in earlier sections of this paper, the research presented

there signaled the existence of a norm of cooperation in EU decision making. From the

framework of Cooperation Theory outlined above, we formulate the following

expectations and hypotheses:

22

First of all, one would expect the EEM to better predict the actual outcomes than the

PEM. Indeed, Dijkstra et al. (2008; Table 1) found that the EEM predictions have higher

correlations with the actual outcomes than the PEM predictions.

Second, we expect the number of actors preferring the PEM predictions to the EEM

predictions to be low, i.e., we expect the decision situation to have a PD payoff structure

for almost all actors. Moreover, as the number of issues in the set M increases, the

number of exchange possibilities naturally increases as well. Thus, the potential gains

from exchange will generally increase in the number of issues considered. However, if

negative externalities are as important as we argue they are, they should grow at a

faster rate than the gains from exchange. Therefore, we hypothesize that the number of

actors preferring the PEM predictions to the EEM predictions decreases as the number

of issues considered increases.

Third, if the decision situation is characterized by a PD payoff structure, there should

be no actors that consistently prefer the PEM predictions to the EEM predictions. Note

how according to Cooperation Theory, the sustainability of cooperation hinges on the

fact that actors who pursue their short-run self-interest can be punished in the future.

However, if an actor consistently prefers the PEM over the EEM, punishment in this

sense is not possible. To make precise predictions of how strong this threat of

punishment would have to be to induce an actor to cooperate requires information about

actors’ discount rates (i.e., the degree to which they care about future outcomes).

However, apart from the obvious difficulties of estimating discount rates for countries, we

simply do not have this information available in the data set. Therefore, we formulate the

weaker hypothesis that on average no actor prefers the PEM over the EEM more

frequently than the other actors do. Thus, no actor is significantly less vulnerable than

others to the threat of refused cooperation

23

Fourth, exchanging according to PEM frequently yields negative externalities for

other actors, when compared to the EEM outcome that does not allow any negative

externalities. However, individual exchanges that are part of PEM are still mutually

profitable for the exchange partners. Therefore, this situation resembles a repeated

Prisoner’s Dilemma (PD), where bilateral exchanges represent defection. Cooperation

can emerge in such a situation when the ‘shadow of the future’ is sufficiently dark

(Axelrod 1984). This being the case in EU decision making can explain the relatively

poor performance of PEM in this dataset (Dijkstra et al, 2008). However, there might be

instances even within this dataset (i.e., commission proposals) where some actors gain

under PEM, i.e., the negative externalities do not override the gains from private

exchanges. For these actors these proposals are not a PD. In fact, of the 49 proposals in

our dataset, only 1 is a PD in the sense that all actors lose under PEM. Thus, we expect

that in general PEM will predict better when there are more ‘winners’.

Fifth, powerful actors are interesting exchange partners for others, since they have a

lot to offer (i.e., they can affect the expected outcome more strongly than weaker actors).

Hence, the assumption is that they are sought as exchange partners more than weaker

actors are. Since the same reasoning holds for themselves, they will mainly seek each

other as partners. In the EU data set examined here, France, Germany, Italy and the

UK are the four most powerful actors: they all have the same maximum power index.

Thus, these four powerful actors as a group can strongly affect the extent to which

exchanges take place. Therefore we expect that whether or not they gain or lose under

PEM (compared to the EEM) has a significant impact on the mean absolute prediction

error (MAE) of PEM.

The first hypothesis was already tested in Dijkstra et al. (2008). The remaining

four hypotheses will be investigated in the following sections of the paper. Note how

24

these expectations derived from Cooperation Theory amount to specifying conditions

under which we expect one model (the EEM) to do better than another (the PEM).

3 Research Design

This research uses a large-scale EU data set which was originally collected by Thomson

et al. (2006). This EU data is uniquely suited to this research as it covers a wide range of

EU policy sectors and is an excellent representative sample of EU Commission

proposals. For the selection of these Commission proposals and the collection of model-

based data, 125 expert-based interviews were conducted and these experts were

selected on the basis of their own in-depth knowledge of the proposal negotiations

(Thomson et al. 2006). The Thomson et al. (2006) data set is comprised of 162

controversial issues, which originate from 66 European Commission proposals, and

which were discussed by the Council in the period January 1999-December 2000. The

experts selected these Commission proposals on the basis of three criteria: legislative

procedure or decision rule to which they were subject, the time frame within which they

were introduced and debated, and their political importance. Furthermore, with regard to

each of these Commission proposals, experts were asked to identify those issues to

which attached a considerable level of controversy (Thomson et al., 2006). For a more

detailed discussion of this data collection process, the reader is referred to the original

research by Thomson et al. (2006). Here follows a discussion of the data used for this

present research.

3.1 Data

25

For this research we identified, from the original Thomson et al. (2006) EU data set,

a total of 49 suitable Commission proposals, which contained 137 issues, We required

that the Commission proposals selected should contain at least 2 issues, since the EEM

and PEM models conceive of decision making as a process of exchanges or outcome

shifts on pairs of issues. In this research the maximum number of issues identified per

proposal was six issues. Moreover each of these issues was defined on a uni-

dimensional, interval scale where the preferred policy outcome of each actor for that

issue could be placed. Overall, Commission proposals, which we selected for this

research, are representative of the different EU legislative procedures relevant for the

time period 1999-2000. In this respect, these proposals differ in terms of whether they

were subject to either the Consultation or Co-decision rule and whether Unanimity or

Qualified majority was the relevant formal decision rule for the Council of Ministers. As in

earlier research using this data (Thomson et al., 2006; Dijkstra et al., 2008), the decision

models are applied to each of the Commission proposals separately and it is assumed

that the collective decision negotiations are restricted to issues within each proposal, so

that no exchanges are permitted to take place across issues from different proposal

sets. This reflects the actual EU decision process where each of the Commission

proposals was dealt with at different points in time (Dijkstra et al. 2008).

The actors involved in the decision process include the members of the Council,

which at the time included fifteen Member States, as well as the European Commission

and the European Parliament (Thomson and Stokman, 2006). All these actors were

treated as unitary actors for the purposes of the research.

Each actor’s preferred policy outcome is defined as the actor’s position on that issue.

The actors’ issue positions were standardized so that positions at 0 and 100 defined as

the most extreme positions favoured by any of the actors. In terms of missing data, on

26

average 15.61 of the 17 actors took positions on each of the 162 issues, where 33 of

these issues were dichotomous (i.e. only two possible positions) (Dijkstra et al, 2008).

The data set for this research includes a measure of each actor’s salience for each

issue. The measure of salience of an actor for an issue is the level of importance that the

actor attaches to the issue. An actor’s salience is ranked as a score between 0 and 100.

If an actor’s salience for an issue is rated at 0, this means that the issue is of no

importance at all, whereas a score of a 100 implies that the issue is of the highest

importance to that actor. The capability of an actor is a measure of potential resources

an actor can exert during the negotiations of the issues in a proposal. In this research,

the actors’ capabilities were estimated using the Shapley Shubik Index (Shapley and

Shubik, 1954).

Furthermore for the purposes of later examining the impact of the decision context

on the EU decision making, we describe here the data in terms of (i) the number of

proposals that have 2, 3, …, 6 issues; (ii) the number of proposals in different policy

sectors; (iii) the number of proposals under different decision procedures. In Table 1 we

see that almost half (49%) of the proposals had just two issues and another

approximately 32% of the proposals had no more than 3 issues. This suggests that the

complexity of the EU decision proposals in this data set, as measured by the number of

issues in the proposal set, is modest with just under 19% of the proposals having 4 or

more issues.

Table 1: Number of proposals having 2 -6 issues:

27

No of

Issues

No. of Proposals Percentage

2 24 49,0

3 16 32,7

4 5 10,2

5 3 6,1

6 1 2,0

Total 49 100,0

In Table 2, we examine the proportion of proposals, which belong to policy

sectors and which may be described as more or less “integrated”. Following

Lane and Mattila’s earlier research (1998), in our research we defined Agriculture

and the Internal Market as “more integrated” and all other policy sectors as “less

integrated”. Mattila and Lane’s ‘roll-call of votes research’ proposes that the

deeper the integration of the EU policy sector, the more difficult decision making

by unanimity (2001).

Table 2: Number of proposals in policy sectors

Level of

Integration

Frequency of

proposals

Percent

More (,00) 23 46,9

28

Less (1,00) 26 53,1

Total 49 100,0

In Table 3, while the spread of proposals is reasonably distributed across the

different procedures, Consultation (CNS) and Co-decision (COD), the majority of the

proposals under both these procedures use the Qualified Majority decision rule (QMV)

compared with the Unanimity decision rule (Unam).

Table 3: Decision Rule and Procedure

Decision Rule and

Procedure

Frequency of

Proposals

Percent

QMV CNS 20 40,8

Unam CNS 9 18,4

QMV COD 16 32,7

Unam COD 4 8,2

Total

49

100,0

29

4. Results

Now let us turn to the model based research results. In this research, we argue that the

norm of cooperation between member states in the European Union is sustainable,

because the configuration of members’ positions and interests gives rise to a decision

situation that resembles a repeated Prisoner’s Dilemna (PD). It is rational for individual

member states to comply since not doing so would mean large forgone gains in the

future. We have elaborated this argument through five specific hypotheses and we test

these by comparing the results of two contrasting collective decision model scenarios.

First, the PEM which captures an ‘unfettered’ exchange decision mechanism and

second, the EEM which presents a more ‘restricted’ collective decision mechanism

taking account of the potential for negative externalities.

To evaluate our expectations, we first determined the predictions of the PEM and the

EEM for each issue. Subsequently, we computed for each member state the utility of the

predictions of both of these models, according to equation (2) presented earlier 3. We

determined, for each proposal separately, which of the member states preferred the

PEM prediction to the EEM prediction, and dubbed them ‘winners’. Thus, winners are

member states that have higher utility under the PEM predictions than under the EEM

predictions. Two observations can be made at this point. First, note how for member

states that are not winners (i.e., those that prefer the EEM to the PEM), the proposal has

a PD-like payoff structure: although individual exchanges according to the PEM are

profitable for them, they would be worse off if all actors exchanged according to the

3 We computed for each member state the utility of the predictions of both of these models, according to equation (2).

We did so for each proposal separately, so the symbol M in equation (2) refers to the set of issues belonging to a

particular proposal.

30

PEM. Second, observe how the set of winners may be different for different proposals

(indeed, this is one of our hypotheses).

In this research we found that the overall mean proportion of winners is 0.41. If one

assumes no missing data, this is calculated as an average of 6.56 out of 16 actors being

a winner, (i.e., preferring the PEM prediction to the EEM prediction). This seems to be

quite a large number and runs counter to our expectation that the number of winners

should generally be low. However as we saw earlier in Table 1, there are a reasonably

high proportion of the proposals under analysis (49%) which have only 2 issues, which

suggests that the tendency for higher than expected number of winners is enhanced in

these specific circumstances.

Next, to evaluate our expectation, that the overall number of winners is generally low

and that it is decreasing as the number of issues per proposal increases, we apply a

standard linear regression modeling approach. In our first regression model, the

dependent variable is identified as the proportion of member states per proposal that are

classified as winners. Member states having missing data on at least one of the issues in

a proposal were excluded from the analysis. We define the independent variable as the

number of issues per proposal minus 2. Since only proposals that contain at least 2

issues are included in the analysis, this variable has values from 0 to 4.

In the regression model, we find a constant coefficient of 0.472 (t47 = 11.45, p < 001)

and an expected negative coefficient for the number of issues per proposal of -0.073 (t47

= -2.25, p = 0.029). The high and significant constant indicates that there are sets of

issues (proposals) in which not every actor is strictly worse off in the situation where

everybody exchanges according to PEM. However, the significantly negative value of

the parameter for the independent variable (i.e. which counts the number of issues per

31

proposal) confirms our hypothesis that individually rational exchanges (modelled by

PEM) very quickly become collectively irrational, due to negative externalities.

These regression results indicate that, in a proposal containing only 2 issues,

the proportion of winners is about 0.47. Assuming no missing data, this would

mean that, on average, 7.55 out of 16 actors are classified as winners in

proposals containing 2 issues. However, for each additional issue, the proportion

of winners decreases by approximately 0.07. Again assuming no missing data,

this would mean that for each additional issue the number of winners decreases

by 1.17.

Extending this argument, in a second regression model, we found that, in addition,

the predictive performance of the PEM (as measured by the model’s mean absolute

error per proposal) gets worse as the number of issues increases. For instance, these

findings show that, for a proposal containing 2 issues, the mean absolute error of the

PEM is 19.77. However, this number already increases by 3.98 (20% increase) with the

first additional issue. Moreover, in this second regression model, we sought to examine

whether some aspects of the EU decision context, such as the decision procedure or the

type of policy sector, might facilitate higher or lower levels of cooperation across actors

in the negotiations for different proposals. In this regression model, we controlled for the

decision procedure (see Table 3) and type of policy sector (see Table 2) but neither of

these control variables had a significant effect.

To test our hypothesis that no actor prefers PEM over the EEM more frequently than

other actors do, we examine whether, on average, no actor is a winner more frequently

than the other actors. In order to do this, we determined the ‘winner distribution’ (i.e., the

count of how many times each actor is a winner across all proposals). The results of our

analysis of the “winner distribution” are presented in Table 4 below. The results

32

presented in Table 4 show that statistically all actors are a winner in 16.25 of the 49

proposals and that the winner distribution does not significantly depart from uniformity

(Chi-square = 15.82, df = 15, p = 0.3944). This concurs with the earlier research findings

of Thomson et al. (2006) which points to the instability of actor alignments in the Council

of Ministers.

Table 4: Winner Distribution across all actors

Actor Aus Belg Den

Com/EP Fin Fra Ger Gre Ire Ita Lux NL Por Sp Swe

UK Total

Obs. 11 12 12 16 14 24 17 16 16 15 12 21 14 18 17 25

260

Exp. 16.25 16.25 16.25 16.25 16.25 16.25 16.25 16.25 16.25 16.25 16.25 16.25 16.25 16.25 16.25 16.25

260

34

This result taken together with the Pareto inefficiency of the PEM helps explain the

existence of a cooperative norm in such a decision setting, where (i) unfettered bilateral

compromises (in the form of exchanges) have very large social costs, and (ii) there is no

consistent group of actors that would benefit from insisting on bilateral compromise as the

dominant mode of decision making. In other words, the costs that would result from bilateral

agreements would be borne by all actors alike (or ‘in turn’). Moreover these results imply that

the shadow of the future is a real threat to these actors. The losses (from negative

externalities) are generally (when everybody exchanges) larger than gains (from exchange

and positive externalities), and no actor can evade future punishment, since no actor is

consistently a ‘winner’. In this regard, Arregui et al. (2006) have shown that, using the same

EU data set as in this present research, the losses due to negative externalities are indeed

larger than gains, summed over all actors and all proposals. Therefore, Cooperation Theory

offers a plausible explanation of the observed EU norm of cooperation.

There are now remaining two hypotheses be tested. If actors exchange according to

PEM, then this yields negative externalities for other actors and the more actors that

experience negative externalities, the less attractive the PEM for the actors overall.

However, in some circumstances and for some actors, the negative externalities do not

outweigh the gains from private exchanges. Hence, we turn to our fourth hypothesis which

proposes that in general PEM will predict better when there are more winners. Related to

this, is the fifth hypothesis which proposes that the four most powerful actors (i.e. France,

Germany, Italy and the UK) can strongly affect the extent to which exchanges with potential

negative externalities can occur in the collective decision process. In other words, we expect

that whether or not they gain or lose under PEM (compared to the EEM) has a significant

impact on the mean absolute prediction error (MAE) of PEM.

Moving to test these final two hypotheses, the first step was to examine which actors stand

to gain or lose from implementing PEM versus EEM. In order to do this, we compute, for

each Commission proposal, the difference of payoffs under PEM and under EEM and then

35

correlate the absolute error of PEM and the Root Mean Square Error (RMSE) of PEM with

the utility gains per country. This yields the result that for every country (i.e. actor) except

Greece, there is the expected negative correlation: the higher a country's gain (or the lower

its loss) the lower PEM's prediction error. Moreover, the results show that the utility gains

under PEM of France and Germany had the highest correlation with the prediction error of

PEM. However, our analysis showed that none of these correlations were significant and

which may reflect the low number of cases (i.e. 49 Commission proposals).

To test the fifth hypothesis, we examined the structure of the externalities and

identified which countries bear 'similar consequences' from PEM. Applying a factor analysis

approach to the spread of utility gains per proposal of the countries, the analysis revealed 4

components4. The results presented in Table 5 show that the first component is by far the

largest and contains 11 actors: Austria, Belgium, the European Commission, Finland,

France, Greece, Ireland, Italy, Luxembourg, Portugal and Spain. The second and third

components comprise Germany and the Netherlands and then Denmark and Sweden

respectively. The fourth component has just one actor, the UK. Notably, three of the four

most powerful actors (i.e France, Germany and the UK) occupy different components which

suggests that there is very limited overlap in the proposals wherein even pairs of the most

powerful four actors can be ‘winners’ under PEM (i.e. PEM is profitable for them compared

to EEM).

Table 5: Factor Analyses on Actor Utility Gains per Proposal

Rotated Component Matrixa

Component

1 2 3 4

Austria_mean ,858 -,123 ,390 -,180

4 Each of the four components had a eigenvalue greater than 1 and after rotation, each country was assigned to

the component where it had the highest loading.

36

Belgium_mean ,964 -,067 ,006 ,041

Commision_mean ,719 -,111 -,403 -,312

Denmark_mean -,311 ,520 ,644 -,105

Finland_mean ,927 ,084 ,167 -,062

France_mean ,831 -,386 ,069 -,020

Germany_mean ,082 ,876 -,125 -,205

Greece_mean ,876 -,233 -,152 ,180

Ireland_mean ,912 ,234 -,143 -,147

Italy_mean ,905 ,221 -,169 -,135

Luxembourg_mean ,969 ,033 -,099 ,139

NL_mean ,111 ,814 ,350 ,184

Portugal_mean ,907 ,149 -,331 ,140

Spain_mean ,886 ,267 -,119 -,249

Sweden_mean ,004 ,011 ,993 -,005

UK_mean -,053 -,065 -,045 ,989

Extraction Method: Principal Component Analysis.

Rotation Method: Varimax with Kaiser Normalization.

a. Rotation converged in 6 iterations.

For example, when we examined the proposals wherein France and Germany are both

winners under PEM (compared to EEM), the analysis showed that there are only 9 proposals

in which both these countries gain from PEM. Furthermore, when we compared the

prediction error on these 9 issues with the remaining available issues5, the results, presented

5 36 issues are used for this comparison and 4 issues were excluded from this analysis since either

France or Germany had missing values on these issues.

37

in Table 7, showed that indeed the average prediction error of PEM is lower in these 9

cases, and is marginally significant.

Table 7: T-Test with France and Germany both winning

Fr_Gr_both

_win N Mean

Std.

Deviation Std. Error Mean

PEM_RMSE 1,00 9 23,4678 17,40127 5,80042

,00 36 29,2401 12,82581 2,13764

PEM_Absolute_Error_mean 1,00 9 19,1364 13,54816 4,51605

,00 36 25,9505 11,26914 1,87819

Extending this to an analysis to the mean gain of all (non-missing) actors for each proposal,

we found a negative, but not significant (one-sided p-value 0.1055), correlation with the

prediction error of PEM. Moreover, for each individual country, with the exception of Greece,

the mean gain under PEM is also negatively related to MAE under PEM. However no single

country alone significantly affects the MAE of PEM.

Using regression modelling, and controlling for the decision procedure as well as the

EU policy area, we systematically explored the effect of different independent variables on

the MAE of PEM. The results showed that the overall number of winners and the overall

mean gain are each negatively related to the MAE of PEM but neither has a significant

impact. However, when we focused our regression analysis on the impact of different

combinations of the four most powerful actors (i.e France, Germany, Italy and the UK), our

38

results showed that all together as a single group, the four most powerful actors do

significantly (i.e. p=0.033) and inversely affect the MAE of PEM, as expected. We also split

up the group of these four powerful actors in pairs or in groups of three to check whether the

mean gain under PEM of these subgroups negatively and significantly affects the MAE of

PEM. The results showed that this was true for the following pairs and triplets: (France,

Germany p= 0.039), (France, UK p=0.026), (Germany, UK p=0.023), (France, Germany, UK

p= 0.011), (France, Italy, UK p=0.051), (Germany, Italy, UK p=0.043). These results suggest

that there is sufficient, albeit indirect, evidence that whenever PEM results in large negative

externalities for the four powerful countries, countries abstain from bilateral exchanges.

5 Discussion and Conclusion

This paper addresses the research problem of why Member States’ bargaining strategies

seem guided by a strong norm of co-operation when engaged in collective decision making

in the EU Council of Ministers. Given their very diverse sectoral and political interests, as

well as contrasting resources, why do member states behave in this co-operative manner?

The present research makes use of a recent and very comprehensive data set of EU policy

decisions in the Council of Ministers. In this research we draw on Co-operation theory and

apply the insights from the Prisoner’s Dilemna Game to identify how the configuration of

members’ positions and interests makes it rational for individual member states to comply

with the norm of cooperation, since not doing so would mean large forgone gains in the

future. We apply two alternative collective decision models, PEM and EEM, to identify the

conditions for ‘unfettered’ or ‘restricted’ exchange bargaining in collective decision making in

the EU Council of Ministers.

The theory and research results presented in this paper suggest that the predictive

power of PEM should vary inversely with the average proportion of ‘winners’ in the data set.

More winners implies that for actors the shadow of the future becomes smaller, since there

will be fewer occasions on which they will actually be in a PD payoff structure. The results

39

presented in this paper support our hypotheses and contribute significant research insights

to the analysis of the core mechanism by which co-operative bargaining strategy dominates

EU decision processes in the Council of Ministers. Moreover the research presented here

yields a more general hypothesis concerning the conditions determining the predictive power

of the PEM and the EEM. Future forthcoming research in this field will apply this analysis to

a much larger and more varied decision data set to test these general hypotheses.

40

Figures and Tables

Figure 1. Exchange between actors Ai and Dj on issues 1 and 2. 1O and 2O indicate

the expected outcomes on issues 1 and 2, respectively, before the exchange. A, B, C, and D

indicate groups of actors

Issue 1

A

B

C

D

O1

i j

Issue 2

A

C

B

D

O2

i j

41

References

Achen, C.H (2006) “Chapter 4, Institutiomnal realism and bargaining models” Pp. 86-123 in

R.Thomson et al. (eds) The European Union Decides. Cambridge University Press.

Arregui, J., F.N. Stokman and R. Thomson (2006) “Compromise, Exchange and Challenge

in the European Union” Pp. 131-132 in R.Thomson et al. (eds) The European Union

Decides. Cambridge University Press.

Axelrod (1984) The Evolution of Cooperation New York: Basic Books.

Bueno de Mesquita, Bruce, D.Newman, and A. Rabushka. 1985. Forecasting political

events. The Future of Hong Kong. New Haven, CT: Yale University Press.

Bueno de Mesquita, Bruce, and Frans N. Stokman (Eds). 1994 European Community

Decision Making: Models, Applications and Comparisons. New Haven: Yale University

Press.

Bueno de Mesquita, B 2002 Predicting Politics. Columbus: The Ohio State University Press.

Black, Duncan, 1958. Theory of Committees and Elections. Cambridge: Cambridge

University Press.

42

Chae, Suchan and Paul Heidhues. 2004 “A group bargaining solution.” Mathematical Social

Sciences 48(1): 37-53.

Dawes, Robyn M. 1980. “Social Dilemnas” Annual Review of Psychology 31: 169-193

Dijkstra, Jacob, Marcel A.L.M. van Assen & Frans N. Stokman. 2008. ‘Outcomes of collective decisions with externalities predicted.’ Journal of Theoretical Politics, 20(4): 415-441. Elgstrom, O and C. Jonsson (2000) “Negotiation in the European Union: bargaining or

problem-solving?” Journal of European Public Policy, No. 7, Vol 5 Special Issue pp 684-704.

Fudenberg, D and J.Tirole (1991) Game Theory Massachusetts: MIT Press

Hayes-Renshaw, F and H. Wallace (2006) The Council of Ministers.(2nd edition). London:

Palgrave. Heritier, A (1999) Policy-making and Diversity in Europe: Escaping Deadlock.

Cambridge University Press.

Jönsson C., Bjurulf B., Elgström O., Sannerstedt A. and Strömvik M. (1998)

“Negotiations in Networks in the European Union” International Negotiation, Volume

3, Number 3, 1998 , pp. 319-344(26)

Lewis, J (1998) “The Institutional problem-Solving Capacities of the Council: the Committee

of Permanent Representatives and the Methods of Community”. Max-Planck-Institute fur

Gesellschaftsforschung. Discussion Paper 90, pp 824-838.

43

Lewis, J. (2000) The Methods of Community in EU Decision-Making and Administrative

Rivalry in the Council’s Infrastructure“ Journal of European Public Policy No 7 pp 261-89.

Lewis, J. (2003) “Institutional Environments and Everyday EU Decision Making : Rationalist

or Constructivist?” Comparative Political Studies 36: 97-124

Lewis, J (2005) “The Janus Face of Brussels: Socialisation and Everyday Decision Making in

the European Union”. International Organisation, Vol. 59, No. 4. pp 937-971.

Lewis, J. (2010) ‘How institutional environments facilitate co-operative negotiation

styles in EU decision-making’, Journal of European Public Policy 17(5): 648–64.

Mattila, M (1998) “Why so much unanimity? Voting in the EU Council of Ministers”. Paper

presented at the 4th biannual workshop of WG-7 Mariechamn, Aland. August 5-9, 1998.

Mattila, M and J. Lane (2001) “Why Unanimity in the Council?: A Roll Call Analysis of

Council Voting”. European Union Politics. Vol 2, No 1 pp31-52.

Richardson, J (2001) European Union: Power and Policy-Making. (2nd edition) London:

Routledge.

44

Schneider, Gerald, B. Steunenberg and M. Widgren “Chapter 11. Evidence with insight: what

models contribute to EU research” in Thomson et al. (eds) The European Union Decides.

Cambridge University Press.

Shapley, Lloyd S., and M. Shubik (1954) “A method for evaluating the distribution of power in

a committee system” American Political Science Review Vol 48, pp 787-792.

Steunenberg, Bernard 1994 “Decision making under different institutional arrangements:

Legislation by the European Community”. Journal of Theoretical and Institutional Economics

150: 642-69.

Stokman, Frans N., and Jan M.M. Van den Bos. 1994 “The exchange of voting positions in

the EC Council.” Pp 161-184 in European Community Decision Making: Models,

Applications and Comparisons, edited by B. Bueno de Mesquita and F.N. Stokman. New

Haven: Yale University Press.

Sullivan, J and T.J. Selck (2007) “Political preferences, revealed positions and strategic

votes: explaining decision-making in the EU Council”. Journal of European Public Policy, No.

14, vol 7 pp 1150-1161

Tallberg, Jonas(2010) 'Explaining the institutional foundations of European Union

negotiations', Journal of European Public Policy, 17: 5, 633 — 647

45

Thomson, R and F.N. Stokman (2006) “Chapter 2. Research Design: measuring actors’

positions, saleinces and capapilities” in Thomson et al (ed.) The European Union Decides.

Cambridge University Press.

Thomson, R and M. Hosli “Chapter 1. Explaining legislative decision-making in the European

Union” in Thomson et al (ed.) The European Union Decides. Cambridge University Press.

Tsebelis, George and Geoffrey Garrett, 1996 “Agenda setting power, power indices, and

decision making in the European Union”, International Review of Law and Economics

16:345-61.

Van Assen, M., Stokman, F.N and Van Oosten, R.C.H. 2003. “Conflict measures in

cooperative position exchange models of collective decision making,” Rationality and Society

15:64-84.

Van den Bos, J. M. M. 1991. Dutch EC Policy Making. A Model-Guided Approach to

Coordination and Negotiation. Amsterdam: Thela Thesis.

Wessels, W (1998) “Comitology: fusion in action. Politico-administrative trends in the EU

system” Journal of European Public Policy. Vol 5, No. 2 pp 209-34. Westlake. M, Galloway,

D, and T. Digneffe (2006) The Council of the European Union (3rd edition). John Harper

Publishing.