EU Integration and Intergovernmental Conferences

Of theories explaining EU integration, intergovernmentalism and institutionalism are the most applicable to the study of bargaining power at IGCs. Neofunctionalism makes fewer predictions about IGC bargaining because it seeks to explain EU integration by examining the creeping competences of EU institutions.¹¹

To the extent that neofunctional studies discuss IGC negotiations, they focus on how supranational institutions, such as the European Parliament and the Commission, shape the preferences of member states, but not why member states win or lose at the bargaining table, the primary focus of this study.¹²

Studies in the neofunctionalist tradition, which argue that examining intergovernmental negotiations is insufficient for understanding EU integration, often recognize that EU treaty negotiation is member state-driven and intergovernmental in nature. Stone Sweet and Sandholtz write that, “EC summits, intergovernmental conferences, and meetings of the Council of Ministers are practically defined by tough, interest-driven negotiations” just as intergovernmentalist literature suggests.¹³

However, neofunctionalists believe that understanding IGCs does not lead to a full understanding of EU integration. This is because, in the view of neofunctionalist scholars, integration is primarily driven by transactions and exchanges between subnational and supranational actors occurring before and after intergovernmental conferences. Marks, Hooghe, and Blank, argue, for example, that contrary to the beliefs of intergovernmentalists, “states do not monopolize the links between domestic and European actors, but are among a variety of actors contesting decisions that are made at a variety of levels.”¹⁴ However, at IGCs both Marks, Hooghe, and Blank, and Stone Sweet and Sandholtz readily admit that states do possess monopoly control over treaty outcomes. Because member states are the only legal signatories of EU treaties, domestic actors can exert influence over treaty outcomes only by directly lobbying member-state governments. IGC bargaining leaves very little room for the subnational–supranational transactions that are the primary focus of neofunctionalists, making their approach less applicable to studies of IGC bargaining.

Other studies in the neofunctional tradition, however, do attempt to address IGCs. This research suggests that intergovernmental negotiations are best understood using a path dependent or historical institutionalist approach because previous interactions between supranational actors and member states shape member-state preferences regarding IGC outcomes.¹⁵

It may indeed be the case that the institutionalized interaction occurring between member states and supranational actors before and after IGCs affects member-state positions at negotiations. However, fully examining these claims would require a model of preference formation, which is beyond the scope of this article.

Studies in the intergovernmental tradition, on the other hand, focus almost exclusively on IGCs but pay significantly less attention to the role of EU institutions in daily politics. Instead, these studies explain EU integration on basis of “grand bargains” between the largest, and by virtue of their size, most powerful, member states.¹⁶

These studies tend to focus on substantive, noninstitutional IGC bargains, such as decisions over monetary union, and view the EU's institutions as a method for credibly committing to EU integration¹⁷ or as agents of member states.¹⁸ Intergovernmentalism often overlooks the fact that there are any number of institutional arrangements that could create credible commitments, and that the precise choices that member states make on institutional issues, such as how to weight votes in the Council of Ministers, empower some member states and supranational actors relative to others.¹⁹ Lastly, by ignoring debates over institutions, intergovernmentalists overlook some of the most controversial and important decisions reached at IGCs, such as the revision of legislative decision-making rules.²⁰

Institutionalist scholars, who analyze EU politics as a function of actors' preferences and institutional constraints, offer strong critiques of both neofunctional and intergovernmental approaches. Tsebelis and Garrett argue that neofunctionalism focuses on institutions as actors who drive integration forward, but they fail to examine institutional constraints, which, along with actors' strategies, lead to equilibrium outcomes. In addition, neofunctionalism, precisely because it is unconcerned with institutional constraints, often overlooks IGC bargaining where the member states negotiate these constraints.²¹

Likewise, intergovernmentalism, by discounting the role of institutions, misses key aspects of IGCs, at which some of the most important and contentious arguments arise over issues of institutional design.

Drawing on the literature from comparative and American politics, institutionalism to date has almost exclusively examined how EU institutions work, without examining how member states design institutions.²²

On the one hand, such a move is laudable. As Tsebelis and Garrett write, “it is simply impossible to analyze institutional choice without first understanding institutional consequences.”²³ On the other hand, this has left a sizeable gap in the literature. Although how institutions in the EU work is well understood, there is significantly less theorizing about how member states choose these institutions at intergovernmental conferences.

Outside the EU, international relations literature has recently moved beyond simply explaining the existence of international organizations to examine how rational member states make collective choices about the institutional rules that govern international organizations.²⁴

These studies find that there is a rational basis for the rules that govern international organizations. However, studying multilateral international negotiations, and EU intergovernmental conferences in particular, remains difficult primarily because there is so little information available about the bargaining setting. Negotiations usually occur behind closed doors, making it hard to determine what the bargaining rules actually are. This makes it near impossible to apply canonical legislative bargaining models often used to examine multilateral negotiations, such as the Baron-Ferejohn model.²⁵ This model, and indeed most models of legislative bargaining, assumes that, in addition to identifying the actors, one is able to identify a recognition rule (who has the right to make a proposal), an amendment rule (open or closed rule), and a voting rule. At intergovernmental conferences, and most international bargaining settings, one does not know the recognition rule, the amendment rule, or how, or even whether, voting occurs within the conference.

Nevertheless, work on IGCs, following the institutional design literature, has begun to study how member states make choices over the rules that govern the EU. These studies tend to examine which member states perform well at the bargaining table and which perform poorly.²⁶

Much of this literature has been based on spatial models, while other studies have drawn on the intergovernmentalist tradition, but there have been few serious efforts to test these approaches against one another. In this article, I present a method for extending institutional analyses beyond daily EU politics to IGCs. I test these institutional predictions about the design of EU institutions against intergovernmental predictions. In doing so, I also present methods for formalizing intergovernmental arguments about IGCs, which, to date, have not been formalized. In addition, my analysis adds to the growing literature on the rational design of international organizations by identifying the relevant sources of bargaining power for member states designing international organizations, and by presenting a method for examining bargaining in settings where researchers have little information about the formal bargaining rules.

A Spatial Model of Bargaining

Although institutionalism and intergovernmentalism often make competing predictions about IGC outcomes, these approaches are best understood as special cases of one general spatial model. When member-state preferences are sufficiently close to one another and sufficiently far from the status quo, the predictions of institutionalism and intergovernmentalism are likely to be indistinguishable, at least on the noninstitutional issues discussed by both theories. However, when one or more member states lie close to the status quo relative to the other member states, the predictions made by the two models diverge.

Figure 1 portrays a one dimensional space where three actors, A, B, and C, are bargaining over how to alter the status quo. Imagine that B and C are large member states, Germany and France, while A is a small member state, Denmark. I present two preference scenarios to demonstrate when intergovernmentalism and institutional theory make different predictions. In Scenario (1) of Figure 1, all member states have relatively cohesive preferences and wish to make substantial changes to the status quo. This matches the preference configuration of member states with regard to Justice and Home Affairs at the Amsterdam IGC, the negotiations I use to test these bargaining theories. Issues such as the communitarization of asylum policy and immigration policy met with little resistance and member states moved decision making on these issues from the national level to the community level at Amsterdam. Under this preference configuration, no member state has an incentive to object to a change to the status quo proposed by any other member state. Both institutionalism and intergovernmentalism would also make a prediction on the bargaining line between A and C, making it very difficult to distinguish between these two theories.

In Scenario (2), Denmark (actor A) is located much closer to the status quo than Germany and France (actors B and C). This mirrors the preferences of member states regarding the revision of voting weights in the Council of Ministers. At the time of the Amsterdam negotiations, much legislation could pass in the Council of Ministers by receiving a qualified majority of weighted votes, where large states had more votes than small states, but small states were overrepresented per capita.²⁷

At Amsterdam, large states were in favor of revising Council voting weights both to better reflect member-state population and to make decision making more efficient in an expanded EU. Specifically, many favored a proposal whereby legislation would pass if supported by a double majority of the weighted votes and member-state population. Small member states tended to prefer the status quo that overrepresented them in the Council and saw this new proposal as an attempt by large states to grab power at their expense. Under this preference configuration, the institutional approach would predict an outcome within the unanimity win-set. In other words, it would predict a Council voting weights scheme very close to Denmark's preference. At worst, Denmark would be indifferent between the bargaining outcome and the status quo. The intergovernmental approach, however, does not predict an outcome within the unanimity win-set. Because Germany and France have a large number of resources relative to Denmark, the prediction of the intergovernmental model will lie outside the unanimity win-set on the bargaining line between Germany and France, making Denmark worse off compared to the status quo. Here the approaches make competing predictions that can be tested against one another.²⁸

From this spatial model, I derive a statistical model that I apply to data collected on the IGC leading to the Treaty of Amsterdam. I examine the probability that negotiators choose a specific constitutional outcome from the set of possible outcomes on each issue discussed at the negotiations. Each member state faces the following choice problem: choose to support alternative x on issue i or choose to support the status quo. In the data I use, there is usually only one alternative to the status quo. This means member states can choose to support the status quo or support the proposed alternative. For example, member states can either choose to support the communitarization of asylum policy or to support the status quo of making decisions regarding asylum at the national level.

Institutional theory suggests that when a state, such as Denmark in Scenario (2) of Figure 1, is located closer to the status quo relative to the alternative, the states supporting the alternative over the status quo will have to somehow gain Denmark's support on that particular issue. This is because Denmark will suffer a loss relative to the status quo if that issue is eventually included in the treaty. Other larger states cannot simply ignore Denmark because if Denmark's total losses are too great, it can veto the entire treaty. As more states, regardless of their size or resources, locate on or near the status quo for a particular issue, it becomes more difficult to gain their support for that issue and more likely that negotiations will fail to alter the status quo. My data does not show member states' exact positions relative to the status quo and alternative. Instead, I know only whether they stated a preference for the status quo or issue inclusion. Therefore, I simply count the number of states preferring the status quo for each issue as the operationalization of institutionalism.

Intergovernmental theory, on the other hand, suggests that the bargaining outcome should reflect both member states' relative bargaining power and preferences. I use Bueno de Mesquita's model of bargaining in the EU Council of Ministers to capture intergovernmentalist logic.²⁹

In this model, each member state faces a choice between an alternative policy and the status quo. Member states cast “votes” equal to the difference in an actor's utility for each option weighted by the actor's capabilities and issue salience. Actor A's quasi-vote, V_A, on a particular issue can be written as follows:

In this equation, C_A captures member state A's relative capabilities or resources. For example, C is much greater for Germany than it is for Denmark because Germany is larger than Denmark. The second term, S_A, reflects how salient a particular issue is for member state A. If a member state cares more about an issue than another member state, its S term is higher. The last part of the equation assumes quadratic utility functions to capture member state A's preference, A, relative to the alternative, X, and the status quo. This term is positive if member state A prefers the alternative over the status quo and negative if the opposite is true. Because the model is multiplicative, V_A is always positive whenever member state A's true preference is closer to the alternative X than to the status quo, and negative otherwise. The absolute magnitude of V_A reflects a member state's relative influence over this particular issue. Again, because the model is multiplicative, there are three ways for a member state to have no influence over a particular issue. The state may have no capability to affect change (C is small), it may not care about the issue (S is small), or it may be indifferent between the proposed alternative and the status quo so the difference in utilities is 0. To make a prediction about whether the status quo remains, V is summed across all member states ([sum ]V). If [sum ]V is positive, the member states collectively choose to alter the status quo; if negative, the status quo prevails. Unlike the institutional model, this approach defines power as the relative capabilities of member states in addition to accounting for member state's preferences. Throughout the rest of the article I refer to this model as the weighted preference model.

To give a concrete example of how these two approaches differ, consider the case of Council voting weights again. In the actual data, a majority of member states (nine of fifteen), all small, prefer the status quo. However, because all of the large member states prefer change, [sum ]V for this issue is relatively positive (2.73 on a scale that ranges from −4 to 4).³⁰

Thus, institutionalism predicts the status quo on this issue while intergovernmentalism predicts change. At Amsterdam, the revision of voting weights was one of the most contentious issues on the table, and member states failed to reform the status quo. On this particular issue, it appears that institutionalism better captures negotiations than intergovernmentalism.

In addition to the prediction of the above model, I use two additional operationalizations of intergovernmentalism in my statistical analysis to be certain that my findings do not hinge upon how I formalize the intergovernmental argument. First, I take the average position of the four largest member states (Germany, France, the UK, and Italy) on each issue. Second, I include two dummy variables that capture when the three largest states (Germany, France, and the UK) either agree to support the status quo or support issue inclusion on each issue.

Case Selection and Data

To examine which sources of power best explain IGC negotiations, I use a data set on the 1996 IGC leading to the 1997 Treaty of Amsterdam. Unlike previous IGCs, which member states called voluntarily according to their own timetable, the 1992 Treaty on European Union stipulated that the member states hold an IGC in four years to continue the difficult institutional reforms postponed at the previous Maastricht IGC. This all but guaranteed that at least some member states would prefer the status quo on important issues on the table at Amsterdam, making this a particularly good setting to test institutionalism and intergovernmentalism against each other as they are more likely to make competing predictions.

At the Amsterdam IGC, member states addressed institutional reforms necessary for eastern enlargement, such as reducing the size of the Commission and changing voting weights in the Council to increase legislative efficiency. In addition, Amsterdam provided member states with an opportunity to strengthen the Common Foreign and Security Policy (CFSP) and reduce the democratic deficit by enhancing transparency in the decision-making process and providing the European Parliament, the only directly elected branch of EU government, with more power. The IGC was launched at the Turin Council meeting in March 1996 and concluded at the Council meeting in Amsterdam on 16–17 June 1997. Negotiations fell well short of expectations in several respects. First, as mentioned above, the treaty failed to alter the system of Council voting weights and reduce the number of Commissioners. Second, it did not make significant progress on changing the CFSP. However, the treaty did appoint the secretary general of the EU Council of Ministers as the EU's “high representative” for the CFSP. It also paved the way to move decision making on asylum and immigration politics away from pure intergovernmentalism to allow the involvement of the European Parliament and the Commission, although it stipulated that many decisions in these areas would still require unanimity in the Council. Finally, the treaty granted the European Parliament new powers, as a result of both new investiture rules for the European Commission and the reformed co-decision procedure.³¹

In fact, increasing the power of the European Parliament, and thereby potentially reducing the “democratic deficit,” was one of the few major changes member states successfully undertook at Amsterdam. Given the high stakes of the negotiations, it is even more important to understand why member states failed to reach an agreement on the most controversial items where agreement was most necessary to streamline the EU's institutions before the upcoming rounds of enlargement.

My data set includes the preferences of member states, the Commission, and the European Parliament on 228 issues discussed at the Amsterdam IGC, as well as the location of the status quo and the outcome of negotiations for each of these issues. These data cover the full set of issues discussed over the course of the IGC, including reform of the common foreign and security policy, justice and home affairs, the EU's major institutions (the European Parliament, the Commission, and the European Court of Justice), and the Economic and Monetary Union. The data are constructed from two primary sources. First, I use a report written by the European Parliament taskforce responsible for monitoring and documenting the IGC process. In February 1995, the European Parliament set up a taskforce to monitor the preparatory stages of the 1996 IGC. One of the taskforce's primary goals was to collect member-state and supranational positions on all issues discussed at the IGC. On the basis of publicly available documents, such as memorandums, press reports, and parliamentary committee and plenary sitting hearings, the taskforce summarized the positions of member states and supranational actors on 252 issues.³²

From these 252 issues, researchers under Professor Thomas König identified 228 separate issues for analysis.³³ On each issue, the taskforce indicated whether actors supported the inclusion of the issue in the treaty or preferred to exclude it, allowing the status quo to prevail. Likewise, König's research team coded whether the final treaty included or excluded the issue and whether issues were favorable towards EU integration or not. Because the European Parliament taskforce tended to code member-state support for issues as binary (support an issue or not), König's team coded the vast majority of negotiation outcomes as binary as well. Of the original 228 draft issues, seventy issues were fully included in the treaty. The negotiators came to a lesser compromise on an additional fifteen issues, and 143 issues were dropped entirely, leaving the status quo. These outcomes form the dependent variable for my analysis.

To check the validity of the European Parliament data, I compare the positions listed in the taskforce report with actor positions collected in a separate research project on the Treaty of Amsterdam. Researchers at the Mannheim Center for Social Research conducted a study of the Amsterdam IGC, also collecting data on member-state and supranational positions over issues discussed at the IGC.³⁴

Although the Mannheim data set is less extensive, the issues covered in these data are virtually identical to those in the European Parliament data. The Mannheim data were based on confidential Council reports, public statements by member states, and expert surveys.³⁵ I first recode the Mannheim data so they correspond with the European Parliament data. I am able to match seventy-four of the 228 issues in the European Parliament data to issues in the Mannheim data.³⁶ Within these seventy-four issues, there are 959 actor positions present in both data sets, and of these 959 positions, 803 (84 percent) are in agreement. For those positions that disagree across the data sets, I reexamine the public statements of actors and take the position that most closely corresponds to the public statements.³⁷ Lastly, if an actor's preference is missing in the European Parliament data, but present in the Mannheim data, I fill in the missing preference with the Mannheim value.

These new data are among the best available on the preferences of member states and supranational actors at IGCs, and perhaps international negotiations more generally, but the data are not perfect. Specifically, there is a problem of missing data. Of the 3,876 possible preferences (seventeen actors times 228 issues), 1,065 are missing (approximately 27 percent). Moreover, the vast majority of the recorded preferences support changing the status quo. Of the 2,811 reported preferences, 1,995 preferences support altering the status quo while only 816 prefer the status quo. It appears that member states are much more likely to explicitly support an issue's inclusion than its exclusion from the final treaty. In the original coding of the European Parliament data, König has assumed that a missing preference is the same as favoring the status quo. This, however, is an assumption that must be tested. I test this assumption by handling the missing preferences in a variety of ways.

The most frequent method for handling missing data in political science is simply list-wise deletion.³⁸

This, however, is appropriate only if the missing values occur completely at random (MCAR), meaning that the occurrences of missing values are uncorrelated with the error term in the statistical model. Even if the MCAR assumption holds, list-wise deletion often results in throwing away vast quantities of data. Using list-wise deletion in the current data set would leave only fifty of the original 228 issues. Moreover, if the assumptions of MCAR do not hold, using list-wise deletion leads to biased estimates.

Methods for imputing missing data employing EM algorithms, such as Amelia,³⁹

take advantage of covariance among variables within the data set to impute missing values in a way that does not inflate statistical significance. These methods usually assume that data are missing at random (MAR), meaning that pattern of missing values can be explained by covariation among variables in the data. Any missing data remaining after accounting for covariation is assumed to be the result of randomness. However, imputation methods such as Amelia will not work if the missing data are fundamentally different from the nonmissing data. This type of missing data may result from selection effects created by the strategic considerations of the actors.⁴⁰ Asserting that the missing values in the current data set are equivalent to the status quo implies that the missing data are fundamentally different than the nonmissing data, meaning Amelia is likely to impute incorrect values.

Rather than imputing the missing data using Amelia, I first examine variables that explain the missing values. This will help underscore the reasons why preferences are missing. Moreover, if I am able to predict the missing preferences, I can show that list-wise deletion is inappropriate. Second, I run the analysis on the data varying the assumptions I make about missing values and demonstrate that my results hold across the various assumptions.

First, I present a Poisson model to examine the underlying reasons for missing preferences. My unit of analysis is the actor, and the dependent variable, presented in Table 1, is simply the number of missing observations per actor. Actors vary greatly in their number of missing preferences. Clearly the European Parliament was able to assess its preferences quite well. Surprisingly, it was much less successful at determining the preferences of the other supranational actor, the Commission. Spain, Belgium, and Luxembourg had relatively clear positions, while Ireland and Denmark had missing preferences more frequently. A missing value implies that both the European Parliament taskforce and the Mannheim research team were unable to determine an actor's preference on a given issue. This could occur for several reasons. First, an actor may honestly not have a position, something likely related to issue salience. Actors may simply not care enough to take a stance on issues of little importance to them. For example, landlocked states such as Luxembourg and Austria are not likely to care about fisheries policy. If this is true, assuming that a missing value is the same as the status quo may be a reasonable assumption. If the actors truly do not care about an issue, they should not be willing to spend their limited political capital to attempt to change the status quo. Unfortunately, I cannot test this argument directly because the data do not contain a measure of issue salience. However, this may be related to an actor's size. Smaller member states are not affected by as many issues as large member states. This would imply that small states should have more missing preferences than large states.

Second, member states may not have a preference because the members of government are unable to agree on a position. If the governing coalition cannot agree on a position, they may simply choose to not state a preference. One may also interpret this as supporting the status quo because implicitly it means that the government actors cannot agree on a position that would alter the status quo. To test this, I examine missing preferences as a function of government composition. Specifically I examine whether the member state has a single-party government or a governing coalition.

Third, member states may have a preference but they may strategically hide that preference if they feel that their position is unpopular. They may not want to take a losing position for fear that being on the wrong side too often may somehow hurt their bargaining position on other issues. Alternatively, they may not want to publicly state an unpopular position for fear that this could somehow hurt their reputation either with the public at home or with other negotiators. This, again, may be related to member-state size. If intergovernmental theories are correct, and size is a source of strength, small countries are likely to hide their preferences if they feel they cannot influence the outcome over the position of the large countries. It does not, however, imply that missing positions should be coded as preferring the status quo.

Lastly, the original coding of the data implies that if actors prefer the status quo, they are less likely to report their preference, or the taskforce will be less likely to ascertain their preference. Perhaps actors wish to avoid the appearance that they are laggards, and thus prefer to make only positive statements rather than negative statements. To test this directly, I simply examine the relationship between missing preferences and the average position of member states on the issues for which they do have a reported preference. If the number of missing preferences increases as member states favor the status quo more often, this would provide justification for the assumption that a missing preference is in fact the same as preferring the status quo over change.

Using a Poisson model, I examine the number of times a member state's preference is missing. As independent variables, I use the log of the member states' populations, a dummy variable that is one for member states with a multiparty government at the time of Amsterdam, and the average position of the member states on the issues for which they do have a stated preference. I drop the supranational actors from the analysis because they clearly have neither populations nor governments; however, I discuss the Commission's missing values below. The results show that the single best predictor of a missing value is the average stated preference. Member states frequently located on the status quo also have a higher number of missing preferences. Coalition governments lead to a greater number of missing preferences as well. Population, on the other hand, has no effect on how often an actor has missing values. Table 2 presents first differences generated from my Poisson models.⁴¹

To create the first differences, I set log population and average position to their means and multiparty government to 0. I then vary the log population from its minimum value to its maximum value holding the other variables constant. I do the same for average position and multiparty government. The analysis demonstrates that, all else equal, moving from the average position of the member state furthest from the status quo to the position of the member state closest to the status quo increases the number of missing preferences by approximately thirty-four issues. Likewise, member states with multiparty governments have, on average, thirteen more missing preferences than member states with single-party governments. Population size has virtually no effect on how often a member state's preference is missing.

The Poisson model offers strong evidence that König's original assumption that a missing position is equivalent to a status quo position is valid. Member states closer to the status quo clearly have more missing preferences, and a member state's status quo bias is the strongest predictor of missing preferences. Moreover, the variable most likely to cast doubt on the validity of this assumption, log population, is not statistically significant, and has no substantive impact.

Unfortunately, because the Commission lacks a population and government, the Poisson model does not provide any insight into why the Commission has so many missing preferences. It is nevertheless worthwhile to consider why its preferences are missing so often, especially when most models of EU legislative politics assume that the supranational actors have relatively similar preferences.⁴²

A primary reason may simply be the weak position of the Commission at IGCs. Although a Commission representative was present at IGC working group and ministerial meetings, he apparently had little influence. The Belgian representative to the IGC, Franklin Dehousse, has written that the Commission was, “extremely weak during the whole [IGC] process, but especially at the end (one can hardly remember a comment from the Commission during the two days in Amsterdam that had any kind of impact, even a modest one).”⁴³ In addition the European Parliament Taskforce White Paper Volume II contains no mention of any position papers written by the Commission. Because of the Commission's weak position, it may have refrained from making public statements, thus making it difficult to ascertain its position.

Analysis

I run several probit models using whether or not an issue was included in the treaty as my dependent variable. My independent variables are the number of member states on the status quo (which capture the institutional argument), and the weighted preference model along with the average position of the largest states and the two dummy variables (which operationalizes intergovernmentalism). The results demonstrate that institutional theory outperforms the intergovernmental model regardless of how I operationalize intergovernmentalism.

Testing these theories in this way may lead to some objections. Namely, this method treats negotiations that are clearly multidimensional and probably involve both issue linkage and log-rolls as a series of independent decisions. Moreover, I do not provide a model of log-rolling. Nevertheless, given the nature of the data, the method I present is the best way to test these theories against each other. First, because the data lack a measure of saliency other than simple missing preferences, developing and testing a model that takes issue linkage into account would be nearly impossible. Without clearer knowledge of what member states care about the most and the least, it is impossible to determine what trade-offs they might be willing to make.

Secondly, empirical tests that try to reproduce the multidimensionality of the issue space are difficult and suffer from different drawbacks. Often researchers use scaling techniques to extract the two or three most prominent dimensions, then use these dimensions to test multidimensional theories rather than treating each observation as an independent event.⁴⁴

Through aggregation, this approach may help deal with some of the problems related to issue linkage, however it throws away large amounts of data. With only two or three dimensions, further statistical models are not possible. Instead, to determine which member state “wins” at the bargaining table, the researcher is reduced to examining which member states lie closest to the outcome in the issue space and then trying to guess which theory this distance best approximates. Moreover, locating the status quo and the outcome of the negotiations is not straightforward. If one scales the data including the status quo and outcome locations, the researcher treats the status quo and outcome as if they are negotiators themselves, and could potentially affect the locations of the other actors in the issue space, something that is nonsensical. The preferences affect the negotiation outcome. The outcome cannot logically affect the preferences.

Although not perfect, the approach presented in this article offers an improvement over other existing approaches to testing bargaining theories. Leaving further concerns aside, I discuss how the variables are constructed. While operationalizing institutionalism by counting the number of member states expressing support for the status quo is straightforward, constructing the weighted preference model, which captures intergovernmentalism, is less so. First, to capture the member states' relative capabilities, I divide each member state's 1997 gross domestic product (GDP) by total EU GDP in 1997.⁴⁵

Because historically the most important aspects of EU integration have been economic and related to the common market, I use GDP to capture the relative bargaining power of member states on the assumption that having a larger market is likely to be the greatest source of bargaining leverage.⁴⁶ To capture the saliency term in the weighted preference model, I take advantage of the fact that missing preferences may in fact contain information. Even though the Poisson model has demonstrated that missing preferences are most likely equivalent to a preference for the status quo, a missing preference may provide information about the intensity of that preference. It is reasonable to think that a member state is more willing to go to bat for a stated preference than for an unstated one. Accordingly, I construct the saliency term as a dummy variable equal to 1 if that member state's preference is known, and 0 if it is missing. Even though this is a crude measure of saliency, it may be better than dropping the term entirely. Lastly, to construct the difference between the expected utility for issue inclusion and the status quo, I assume that the position of the status quo is −1 and the position of issue inclusion is 1. This allows member states to be indifferent between issue inclusion and the status quo for two different reasons. First, they may be truly indifferent, meaning that they have a stated preference, but that preference lies at 0 halfway between issue inclusion and the status quo, or they may have missing preference on that issue, which would mean the saliency dummy equals 0.⁴⁷

To further test my assumptions about missing preferences I run my primary model, the outcome of treaty negotiations as a function of the number of actors on the status quo and the weighted preference model, making different assumptions about missing preferences. First, I simply use list-wise deletion. Next, I make two different assumptions about the nature of missing preferences to impute the missing data. I first assume that missing preferences are identical to status quo positions and missing data provide no information about issue saliency. Here the saliency term drops out of the weighted preference model. Second, I assume that a missing position implies a less intense status quo preference. In this model I construct three variables: the number of member states truly preferring the status quo (that is, those member states not missing and indicating a preference for the status quo), the number of member states missing, and the prediction of the weighted preference model using missing preferences to indicate issue salience. Finally, I rerun the model assuming missing preferences provide information about issue intensity, but I also include a variable that weighs the number of missing member states by the percent GDP those states contribute to the EU's total GDP. If significant, this variable would support intergovernmentalism because it would suggest that large states' missing preferences are more important than those of small states. Table 3 presents the results of these various models. All models are run with robust standard errors so I do not inflate statistical significance as a result of the possible nonindependence of the observations due to log-rolling.

Looking at Table 3, there are two questions to be answered. First, which variable best explains the treaty outcome, and second, do the assumptions about missing data affect the results? Although the Poisson model has demonstrated that MCAR is an inappropriate assumption given these data, I first run a model with list-wise deletion. The subset of issues that do not contain any missing data is clearly not a random sample of all issues. Instead, as one might expect, it includes many of the most important and contentious issues on the table, such as extending qualified majority voting to various issue areas including Justice and Home Affairs, changing the voting weights in the Council of Ministers, and altering the number of commissioners. Even though this may not be a random sample of issues, it is an interesting sample to examine as it contains the issues that member states arguably cared most about. Running the model on this subset of issues ameliorates concerns that my results are due to small states getting their way on issues that large states do not care about. In this model, number sq has a strong effect, both substantively and statistically, while weighted preference is not statistically significant.

Rather than dropping issues with missing data, Models 2, 3, and 4 make reasonable assumptions about the nature of missing preferences. Model 2 assumes that a missing value is exactly the same as a preference for the status quo, while Models 3 and 4 assume that a missing preference contains information about issue saliency. In all three models number sq is statistically significant and weighted preference is never significant. Interestingly, in Model 3 when the number of missing preferences is separated out from the number of preferences for the status quo, both variables are significant, but the effect of number sq becomes stronger compared to Model 2 and it is larger than the effect of number missing. This comparison provides further evidence that a missing preference is a less intense preference for the status quo. Finally, the GDP-weighted missing preferences variable in Model 4 is not statistically significant, further indicating that weighting preferences by member-state size does not provide any additional information about bargaining outcomes.

Because these are probit models, it is difficult to determine the substantive effect of the coefficients directly. Instead, to demonstrate the effect of member-state size on the probability of issue inclusion, I run simulations examining the case of Council voting weights. On this issue the nine smallest member states preferred the status quo while the six larger states (Germany, France, the UK, Italy, Spain, and the Netherlands) preferred a double majority voting rule. For the moment, assume a hypothetical scenario in which Denmark decided to join the coalition of large states on this issue, leaving only eight member states on the status quo. I examine the predicted drop in the probability that the double majority proposal is implemented when Denmark moves from supporting the proposal to supporting the status quo and when Germany moves from supporting the proposal to supporting the status quo. If intergovernmentalism is correct, one would expect a much larger change in probability when Germany shifts its position compared to when Denmark shifts its position.

According Model 2, if Germany switched its position to support the status quo rather than the double majority proposal, the probability that the proposal would be included in the treaty would drop by 0.046 from 0.288 to 0.242. On the other hand, if Denmark switched to support the status quo instead of Germany, the probability of including the proposal would fall by 0.054, a slightly greater decrease than the hypothetical situation in which Germany switched positions. Models 1, 3, and 4 provide similar results. The coefficients from Model 3 suggest that when Germany switches from supporting the proposal to supporting the status quo, the probability of issue inclusion decreases by 0.06. Likewise, when Denmark shifts to support the status quo, the probability of issue inclusion decreases by 0.05, an almost identical change compared to when Germany shifts. Model 4 suggests that Germany switching position decreases the probability of issue inclusion by 0.09 compared to 0.07 for Denmark. Only Model 1, the list-wise deletion model, suggests that perhaps there is a larger effect for Germany switching positions compared to Denmark (a decrease of 0.13 compared to a 0.05 decrease). However, a simulated 95 percent confidence interval suggests that one cannot rule out that the effect of Germany's shift is 0, while one can rule this out for Denmark. In no instance is there a statistical difference between the changes in the probabilities when the two states switch positions. As institutionalism suggests, the preferences of all states, both large and small, seem to matter equally.

To test these results further, I pit two different operationalizations of intergovernmental theory against the number sq variable. Because the above results demonstrate that a missing preference is similar to a preference for the status quo, I calculate all variables here assuming a missing preference is equivalent to a preference for the status quo. First, I operationalize intergovernmentalism as the average position of the four largest member states, Germany, France, the UK, and Italy, on each issue. Second, I examine whether issues are more likely to be included in or excluded from the final treaty if the three largest member states take a common position. I include two dummy variables in the analysis: large state include, which equals 1 if Germany, France, and the UK agree to include an issue in the treaty and 0 otherwise, and large state exclude, which equals 1 if Germany, France, and the UK agree to exclude an issue from the treaty. I also interact these dummies with number sq. Intergovernmental theory would suggest that number sq should matter less if the three largest member states prefer change and more if they prefer the status quo. This would imply a positive coefficient on the interaction large state include * number sq and a negative coefficient on the interaction large state exclude * number sq. Table 4 presents the results of my analysis.

The models demonstrate that my findings hold for various operationalizations of intergovernmentalist theory. Regardless of how I conceptualize intergovernmentalism, the variable associated with institutionalism, number sq, best explains the treaty outcome. It is the only variable in either model to achieve statistical significance. Both large state average and large state include have the wrong sign, meaning that, if anything, large states have less power than small states. The dummy variables and interaction terms in Model 6 are not statistically significant. A likelihood ratio test reveals that they are also not jointly significant either (χ² (4) = 0.98, p = 0.91).⁴⁸