Incomplete information and bargaining in the EU: An explanation of first-reading non-agreements

Abstract

According to the logic of standard veto bargaining models with complete information, bargaining in the European Union should never reach the second reading of the ordinary legislative procedure (co-decision). Even so, non-agreement at the first reading occurs frequently in European Union decision making. How can this be explained? Drawing upon game theory on bargaining with incomplete information, two specific predictions with regard to the occurrence of first-reading non-agreements can be generated. First-reading agreements are less likely to occur (1) the greater the distance between the ideal point of the Council and the ideal point of the European Parliament and (2) the greater the salience attached to the proposal. This article combines positional and proposal-specific data and finds robust support for both hypotheses in two different empirical tests.

Keywords

European Union legislative politics incomplete information preferences salience

Introduction

The basic game-theoretic model of veto bargaining infers that when all actors are completely informed about the preferences of the other players, vetoes should not occur. This is because the agenda setter can choose to propose the policy closest to her own ideal point that will also be approved by the veto player (Cameron and McCarty, 2004). According to this logic, bargaining with the possibility of extending to more than one round would always be completed in the first round. The necessary assumption here is complete information. Studies on European Union (EU) decision making usually assume that actors have complete information with regard to the other actors’ positions in the bargaining (Crombez, 1996; Crombez and Hix, 2013; Moser, 1996; Scully, 1997; Steunenberg, 1994; Tsebelis, 1994; Tsebelis and Garrett, 2000).

This assumption makes it easier to model the bargaining process, but nevertheless, some empirics are contrary to it. According to the basic veto bargaining models, when the main decision makers in the EU (the Council of Ministers and the European Parliament (EP)) have complete information with regard to each other’s preferences on co-decision legislation, we should never observe legislation that is adopted at later stages than the first reading. The logic behind this argument is that a ‘no agreement’ at the first reading is costly and can be prevented by striking a deal that suits the configuration of preferences within the Council and the EP. However, legislation is frequently adopted at later stages than the first reading. In the 5th and 6th EPs (1999–2009), 53% of the co-decision proposals were completed at the first reading (Toshkov and Rasmussen, 2012).

Hence, the assumption of complete information can be relaxed in the EU context. While the Council presidency keeps track of the various Council members’ preferences on a particular piece of legislation, the presidency is not necessarily informed of the exact preference distribution in the EP. The same holds for the party group leaders within the EP. They may have control over the preference configuration within their own group but are likely to be less informed about the preference configuration within the Council and within other party groups inside the EP. While the EP negotiates in public, Council debates are traditionally conducted in a less transparent manner (Stasavage, 2004: 688–695). This means that the Council may know more about the EP than the EP knows about the Council. If preferences were completely known inside and across institutions, we would only observe first-reading agreements as bargaining costs increase with the number of decision-making rounds. Nevertheless, empirics show that agreement is reached at all stages in the co-decision procedure. Tsebelis and Garrett (2000) argue that complete information only exists in the final rounds of EU decision making. By then the actors have had the opportunity to learn about each other in previous rounds of bargaining.

This article thus relaxes the assumption of complete information in EU bargaining in order to explain incidences where first-reading agreement cannot be reached across the institutions. Drawing upon veto bargaining theory with incomplete information (Cameron and McCarty, 2004), two specific predictions regarding the occurrence of first-reading non-agreement can be generated. First-reading agreements are less likely to occur (1) the greater the distance between the Council pivot and the median position of the EP and (2) the greater the salience attached to the proposal. This article tests both expectations. The first test utilises positional data on 183 issues nested in 64 proposals adopted before and after the 2004 enlargement (Thomson et al., 2006, 2012). The second test merges various data sources at the proposal level in order to investigate whether the expectations also hold for a larger data sample (656 proposals) that includes other operationalisations of the substantial predictors and important control variables (Häge, 2007, 2011; König and Luig, 2012). Both tests find robust evidence for the empirical predictions generated from veto bargaining models with incomplete information.

First-reading agreements are negatively associated with high salience and large differences in preferences between the EP median and the Council pivot. Hence, while other studies on early agreement in the EU emphasise the democratic drawbacks of this type of decision making (e.g. Farrell and Heritier, 2003; Häge and Kaeding, 2007; Reh et al., 2013), this analysis contributes to the positive aspect of the debate as highly salient and controversial cases are associated with more transparent decision-making modes.

Co-decision bargaining and early agreements in the EU

The co-decision procedure was introduced under the terms of the Maastricht Treaty (1992). One of the main reasons for establishing this procedure was to empower the EP and hence to increase the democratic legitimacy of the EU (Crombez, 1997). The reform of the co-decision procedure by the Amsterdam Treaty in 1999 not only made the EP and the Council of Ministers (hereafter the Council) coequal legislators, it also paved the way for early agreements in the first reading of the procedure (Farrell and Heritier, 2003). The possibility of early agreements reduced bureaucracy in the sense that policies already settled at first reading no longer needed a second reading to be formally adopted. About a third of all dossiers decided under this legislative procedure in the fifth term of the EP (1999–2004) were so-called fast-track legislation completed at the first reading (Rasmussen, 2011). The growing number of first-reading agreements has spurred discussion on possible negative consequences on democracy (Farrell and Heritier, 2003, 2004; Shackleton and Raunio, 2003). First-reading agreement means that the Council and the EP have reached an informal compromise before the legislation reaches the plenary stage in the EP at the first-reading stage. This informal process is called trialogues since representatives from the three legislative institutions – the Commission, the EP and the Council – are involved. Bargaining in trialogues has been criticised for a lack of transparency and for shutting minority interests out from the negotiating table (Farrell and Heritier, 2003; Häge and Naurin, 2013).

Despite this type of criticism, empirical studies investigating factors that determine early agreements paint a mixed picture of the democratic consequences of fast-track legislation. On the positive side for democratic legitimacy, early agreements are less likely to take place the greater the distance between a rapporteur from a large party group and the EP median (Rasmussen, 2011). Furthermore, the Council presidency cannot exploit informal negotiations in trialogues in order to secure policy gains (Häge and Naurin, 2013; Reh et al., 2013). Policies reached early on are no closer to the positions held by the main informal actors – the Council presidency and the EP rapporteur’s party group – than policies reached at later stages (Rasmussen and Reh, 2013). There is also evidence for the EP, under co-decision, having more influence on early agreements than on agreements reached in the conciliation committee at the end of the procedure (Häge and Kaeding, 2007). This finding suggests that the Council prefers to conclude early over agreement in the conciliation committee, in which costs are greater for the Council than for the EP (due to the former’s limited time and resources).

Although the above findings do mitigate the criticism of fast-track legislation to some extent, several studies also note negative consequences of early agreements on democracy. For instance, when the legislative workload and the number of negotiators increase, decisions are more likely to be taken at an early stage (Reh et al., 2013). While the efficiency gains of concluding early are indisputable, the legislative transparency and minority legitimacy suffer. The democratic accountability is also reduced due to the ministers being less involved in bargaining on fast-track legislation than in bargaining at later stages of the co-decision procedure (Häge and Naurin, 2013). Instead, Council bureaucrats negotiate early agreements on behalf of the ministers.

This article endeavours to contribute to the debate primarily by investigating the effects of preference divergence and salience on early agreements. While preference divergence and salience have been investigated thoroughly for other research questions related to EU bargaining and decision making (e.g. Aksoy, 2012; Cross, 2013; Häge, 2007; Schneider et al., 2010; Thomson, 2011; Thomson et al., 2006), the effects of these variables can be further explored in the setting of early agreements. In terms of democratic accountability, it would be preferable if highly salient and controversial cases were to be decided in a more transparent way than informal bargaining in trialogues. Previous studies on fast-track legislation provide some evidence that greater preference divergence reduces the probability of early agreements. Rasmussen (2011) finds that greater distance between the EP rapporteur (who is in charge of drafting the proposal content inside the EP) and the EP median on the left–right dimension is negatively associated with early agreements. This finding is contingent on the rapporteur stemming from one of the large party groups within the EP. Reh et al. (2013) show that greater policy distance between the national political parties of the rapporteur and the presiding minister of the responsible Council formation at the time of political agreement reduces the probability of trialogue bargaining. This effect is small and only significant at the 10% level. Reh et al. (2013) also investigate how salience in terms of media coverage affects the probability of early agreements. While the expected effect is negative, this study shows that early agreements are equally likely to involve salient proposals as agreements at later stages of the co-decision procedure.¹

The analysis in this article builds upon this research and seeks to explicitly investigate preference divergence and salience as key indicators for explaining first-reading non-agreement. The choice to single out these indicators is based on veto bargaining models with incomplete information (Cameron and McCarty, 2004). If the actors have complete information, it follows that all legislative dossiers should be decided as early agreements. This is still not the case in the EU, although there has been a notable increase in fast-track legislation. Testing the directly applicable predictions from veto bargaining theory is the main contribution that this article makes to the existing literature. Both indicators are tested in two different ways, and each indicator thus has two different operationalisations. In the first test, each member state’s issue position, the EP position, and the salience the actors assign to their issue positions are measured on a 0–100 scale by policy experts (Thomson et al., 2006, 2012). In the second test, salience is measured as the number of recitals listed as reasons for adopting the relevant legislative act (Häge, 2007; Häge and Naurin, 2013). While Rasmussen (2011) and Reh et al. (2013) identify the EP rapporteur and the EP median or the EP rapporteur and the presiding minister as the key negotiators, this analysis singles out the EP median and the Council pivot under qualified majority voting (QMV) as the key negotiators (Crombez and Hix, 2013). This choice is based on the voting rules in the respective institutions. The EP needs a simple majority to pass acts in the first reading while the Council usually needs a qualified majority. Furthermore, by employing data on EU actor preferences which vary by year and policy area (König and Luig, 2012), the preference distance variable included in this analysis has more variation than the ones included in Rasmussen (2011) and Reh et al. (2013) (which use NOMINATE scores (Hix et al., 2006) and expert survey data (Benoit and Laver, 2006), respectively). The research design presented after the theoretical framework is thus carefully designed to test the empirical implications of veto bargaining models with incomplete information.

Theoretical framework

This theory section seeks to explain why, under the assumption of incomplete information, some co-decision proposals are adopted in the first reading and others are not. The theoretical foundations are based upon standard models of veto bargaining, which generally assume that the actors have Euclidean preferences (Cameron and McCarty, 2004); that is, they prefer the policies closest to their own ideal point to policies farther away. For simplicity, the ideal points of the different actors as well as the location of the existing policy (the status quo) and the new policy are presented in a one-dimensional space.

The standard models show that if all actors are completely informed about the preferences of the other actors, vetoes should not occur. This proposition holds regardless of whether the setting is one-shot veto bargaining or sequential veto bargaining. In the complete information veto game, the agenda setter is completely informed of the veto player’s preferences and thus knows which bills will be successful. The agenda setter can propose its most preferred bill from the set of bills that the veto player will find acceptable (Cameron and McCarty, 2004). Consider Figure 1. Given the ideal point of the agenda setter (A) and the veto player (V), V will veto any bill located to the left of the agenda setter and A will not propose any bill located to the right of the veto player. A successful bill thus has to be located within the interval between the agenda setter and the veto player.

Figure 1.

Veto game with complete information: any successful bill will be located within the interval between the agenda setter (A) and the veto player (V).

Similarly, in the EU setting, if all actors have complete information with regard to the preferences of the other actors, first-reading non-agreement under co-decision should not occur. Under the assumption of complete information, the EP should present the Council with amendments to the Commission proposal that it knows the Council will accept. Yet, first-reading non-agreements do occur. In line with veto bargaining models, relaxing the assumption of complete information is necessary to explain the incidences of non-agreement. If the EP has incomplete information, non-agreement may occur due to the EP overestimating its ability to extract concessions from the Council.

The theoretical assumption of incomplete information in EU bargaining can be defended empirically. While the Council negotiations are traditionally undertaken behind closed doors, the EP negotiations are public (Stasavage, 2004).² Hence, the Council has much more information about the preference configuration within the EP than the EP has about the preferences of the different Council members. Incomplete information games require two conditions to be in place (see Fearon, 1995). The first is uncertainty: at least one of the main actors needs to have doubts with regard to the type of the other actor. The secrecy of Council negotiations ensures that the EP cannot be certain of the positions of the pivotal actors within the Council. The Council therefore has private information.³ The second is incentives to misrepresent its own type to extract concessions from the other actor. If the preference distribution within the Council is significantly different from the preference distribution within the EP, the Council can clearly benefit from signalling a more extreme position in order to obtain a policy closer to its own ideal point. In such situations, the Council will prefer not to reveal its (true) private information to the EP. However, the EP can also benefit from building a more extreme reputation to extract concessions from the Council. This will give the EP more leverage both in trialogues and later rounds of decision making. To simplify the theoretical section, such scenarios, where the Council does not know with certainty the position of the EP, are not considered further. The empirical implications developed in this section hold for both types of scenarios.

In the following, which is based on veto models with (in)complete information (Cameron and McCarty, 2004), the Council and EP are presented as unitary actors. The ideal point of the Council is assumed to be the position of the pivotal government in the Council under co-decision. Since the Council usually decides by QMV under co-decision, the government that is pivotal for a rightward (leftward) move is located to the left (right) of the government with the median vote. In practice, the government with the 5/7th vote is pivotal for a policy move to the left and the government with the 2/7th vote is pivotal for a policy move to the right if the status quo is located as illustrated in Figure 2. The EP votes by simple majority in the first reading under co-decision. Hence, given the assumption of uni-dimensionality, the pivotal Member of the EP (MEP) is the median member of this institution (Crombez and Hix, 2013).⁴ After the Commission has initiated a legislative proposal, the EP can present amendments to the proposal that the Council can choose to accept. According to the logic set out in Figure 1, the EP thus serves as the agenda setter and the Council as the veto player in first reading co-decision bargaining.⁵

Figure 2.

Bargaining game with complete and incomplete information: Council pivot either moderate (M) or extreme (E).

Figure 2 shows EU bargaining with complete and incomplete information. Here, the policy outcome (the adopted bill) is denoted as B and the Council position is either denoted as M or E depending on whether the pivot is moderate (M) or extreme (E). Figure 2(a) and (b) shows possible scenarios when the EP is completely informed about the preference configuration within the Council. If the EP knows that the Council pivot is moderate (Figure 2(a)), the EP will propose amendments to the Commission proposal that reflect its own ideal position. Since the Council pivot prefers such a policy outcome over the status quo, the adopted bill will be identical to the EP position. However, the EP will be worse off if the Council pivot is extreme (Figure 2(b)). An extreme pivot will not accept the EP’s ideal position and the adopted bill will be located between the position of the Council pivot and the EP position. Note that the bill adopted in Figure 2(b) is closer to the moderate Council pivot’s position than the bill adopted in Figure 2(a) when the EP knows that the Council pivot is a moderate type. This provides the Council with an incentive not to reveal its true type in the early stages of inter-institutional bargaining.

Under incomplete information, the EP cannot determine with certainty what type of Council the Council really is; that is, whether the Council pivot has an extreme or moderate position. The set of bills preferred by an extreme pivot is only a subset of the bills preferred by a moderate pivot. Figure 2(c) illustrates this argument. Bill B_m, which is the outcome of bargaining when the Council pivot is moderate, is clearly preferable to the EP than bill B_e, which is the outcome of bargaining when the Council pivot is extreme. However, the EP may falsely believe that the Council pivot is moderate and propose B_m, which the extreme pivot will then refuse to accept.

The veto bargaining models with incomplete information generate several predictions of when vetoes are more likely to occur. Two of these predictions can be directly used to explain why first-reading non-agreements under co-decision happen. The first of these two predictions is that vetoes are more likely to occur the larger the difference between the ideal points of the pivotal actors (Cameron and McCarty, 2004). In the EU setting, this prediction can be transferred to first-reading non-agreement under co-decision. Figure 2 can easily illustrate why this is the case (for the mathematical proof see Cameron and McCarty, 2004). If the distance between the EP pivot and the Council pivot increases, the Council pivot is more likely to be extreme than to be moderate (vis-à-vis the EP). Under the condition of incomplete information, the EP cannot know with certainty whether the Council is moderate or extreme. Still, the EP is most benefited by proposing a bill close to its ideal position. Recognising that later rounds of decision making are costly to the Council (Häge and Kaeding, 2007), the EP knows that there is a probability that the Council will accept its proposal. The probability of the Council accepting the EP proposal is reduced when the distance between the two policy makers increases. For an illustration of this argument, compare Figure 2(c) to (d) where the actors are located closer to each other. Hence, larger preference heterogeneity can be assumed to reduce the probability of first-reading agreements in the EU.

The second prediction that can be borrowed from veto bargaining models with incomplete information is the result that the importance the actors assign to the bill matters for the probability of vetoes occurring or not. Given disagreement between the pivotal actors, more salient bills are more likely to be vetoed than less important ones and more likely to be re-passed if vetoed (Cameron, 2000). This is due to the agenda setter’s incentive to put forward more risky proposals if the agenda setter cares highly about the policy content. Bargaining is assumed to be tougher for important bills. The second prediction is generated from sequential veto bargaining models while the first prediction holds for both one-shot games and sequential bargaining (Cameron and McCarty, 2004).

In the EU setting, the two predictions from the veto bargaining models can be transformed into a theoretical framework that seeks to explain the occurrence of first-reading non-agreement under the co-decision procedure:

H1: First-reading agreement is less likely to occur the greater the distance between the ideal points of the Council pivot and the EP median.

H2: First-reading agreement is less likely to occur the greater the importance attached to the legislative proposal.

The statistical analysis in this paper tests whether these two predictions can contribute to explaining why some proposals can be agreed upon at the first-reading stage of the co-decision procedure while others cannot. Figure 3 shows how the theoretical framework relates to the research design.

Figure 3.

Co-decision bargaining under incomplete information.

Data and methodology

The analysis is based on two different tests of the hypothesised relationships between the key independent variables and the dependent variable First-reading agreement. The first test is based on positional data taken from the ‘Decision Making in the European Union (DEUI/II)’ project (Thomson et al., 2006, 2012). In this dataset, the issue-specific positions of the different member states, the Commission and the EP are measured by policy experts on a 0–100 scale reflecting the policy distances between the different actors. The reference point (similar to the concept of status quo), the policy outcome and the salience that the actors attach to the issues are also identified with this scale. The above data were collected through 349 semi-structured interviews with key informants. Altogether, the dataset contains 331 controversial issues nested in 125 legislative proposals that were introduced between 1996 and 2008 (Thomson et al., 2012). For the purpose of this paper, only co-decision proposals are relevant. This reduces the dataset to 183 issues raised by 64 legislative proposals.

The second test utilises the European Union Policy-Making (EUPOL) dataset. The EUPOL dataset covers all information contained in the European Commission PreLex database for the years between 1975 and 2009 (Häge, 2011). The second test also employs data taken from the EUropean Legislation and Ideological positions in policy Spaces (EULIS) dataset (see König and Luig, 2012), EUR-Lex and the monthly summaries of Council acts. The sample of proposals from the EUPOL dataset consists of all binding co-decision legislation adopted under QMV-rule between 1999 and 2009 that could be sufficiently merged across the different data sources.

The dependent variable, First-reading agreement, is used in both tests and is taken from the EUPOL dataset. This binary variable is coded 0 if agreement was reached at later stages than the first reading and coded 1 if only one reading was needed to adopt the relevant legislation. While Rasmussen (2011) also codes first-reading agreement as early agreement, Reh et al. (2013) measure early agreements as first and early second-reading agreements that include an informal compromise between the EP and the Council. While the latter measurement is very efficient in the sense that it removes uncontroversial first-reading agreements from the early agreements category, it is not suitable for the theoretical framework employed in this paper that focuses on why not all proposals are adopted at the first-reading stage. However, redoing the analysis with early second-reading agreements (where the EP has accepted the Council’s common position without amendments) as part of the early agreement category yields similar results to those presented in this paper. To control for non-controversial first-reading agreements, the second test includes a variable that measures whether the EP made any amendments to the proposal in question.

The independent variable employed in the first test is a combination of the two key variables Salience and Distance between EP median-Council pivot. Each position is weighted by the salience the actor attached to the issue. In practice, all issue positions were multiplied by the respective salience estimates and then divided by 100. This ensures that the results are not driven by extreme positions that are not prioritised by the actors that hold them (Aksoy, 2012; Cross, 2013). Another approach would be to include salience and positional distance as separate variables and an interaction of these in the model to assess the theoretical expectations. However, the Council pivot could then potentially be a government that does not care about the proposal in question. This government is thus not the actual pivot in the negotiations, which would bias the results accordingly.

The Council pivot is identified by taking the 5/7th percentile of the salience-weighted positional distribution vis-à-vis the reference point. This is the pivotal actor when the Council decides by QMV, which it usually does under co-decision. If the reference point is 0, the Council pivot is located to the left hand side of the scale. If the reference point is 100, the Council pivot is located to the right hand side of the scale. After this, the absolute salience-weighted distance between the Council pivot and the EP position is calculated.

There are some missing positions in the DEUI/II dataset. Failure to account for this may bias the results (Honaker and King, 2010; Little, 1992). In the DEU sub-sample employed in this paper, 39 of 75 pre-enlargement and 42 of 108 post-enlargement issues have complete positional information. Even though the level of missingness is not very high, the analysis employs the iterative imputation method, Multiple imputation using chained equations (MICE) for robustness purposes (see Van Buuren et al., 1999 for a detailed account of MICE). Here, the missing values are imputed on an issue-by-issue basis where predictive mean matching selects the predicted values closest to the observed ones. The other observed positional values on one issue are thus used to predict the missing values on that particular issue. The analysis is run with both MICE and list-wise deletion of missing values.⁶ Descriptive statistics for both the imputed and unimputed datasets are presented in the Online Appendix. Since the independent variable is on the issue-level, control variables are not included. While proposal-level controls could potentially be included, the small number of observations is an argument in favour of a simple model. The second test includes more observations as well as control variables at the proposal level.

The second test is conducted on 656 co-decision proposals adopted between 1999 and 2009 (Häge, 2011). The independent variables in this test are coded as follows. The Distance between EP median-Council pivot is based on the EULIS dataset, which combines issue-specific Comparative Manifesto Project data with the EU legislative agenda. While political parties are relatively constant in their ideological positions on each issue, the relative importance of each issue on the EU legislative agenda is not fixed. By weighting issues according to their yearly prominence on the legislative agenda for the different EU policy sectors, more time and area variation is secured. The issue-specific party positions are then aggregated to a left/right and national/supranational dimension (König and Luig, 2012).

On the basis of this information, the upper and lower (5/7th and 2/7th percentile) Council pivots are identified for each of the policy sectors, on the left–right dimension, with regard to the applicable years in the dataset. Since day-to-day EU politics can be better measured by the left–right dimension than the more or less integration dimension, the paper rests on the former dimension of conflict in the analysis (e.g. Crombez and Hix, 2013; Hertz and Leuffen, 2011; Hix et al., 2006). The Council pivot is calculated from the year and sector-specific distribution of mean government positions within the Council. Data on the composition of national governments are taken from the ParlGov dataset (Döring, 2013). The EP median is identified by the same procedure as Crombez and Hix (2013). Each MEP is assigned the same position as its national party. To combine MEP party information with the EULIS dataset, EP5 and EP6 background data from the dataset used by Hix et al. (2006) are utilised. After merging the data, the median within each political group is calculated. Finally, the EP median is identified as the median political group on each dimension for the different sectors and years in the EULIS data. The Distance between EP median-Council pivot variable is then created as the absolute distance between these actors within the two institutions. Since it is logical to assume that the Council pivot is the upper or lower (5/7th or 2/7th) actor that generates the largest distance to the EP median, the variable is further operationalised as the largest absolute distance between the actors.

The Salience of a dossier is measured as the number of recitals in the original Commission proposal. Information on recitals is obtained from the EU database EUR-Lex. Recitals are listed reasons for adopting the particular piece of legislation and the importance of a piece is assumed to be positively correlated with the number of reasons provided for its adoption (Häge, 2007; Häge and Naurin, 2013).⁷

The variables More than one policy area, Any EP amendment, First-reading Council consensus, Ministers involved up to first Council decision and dummy variables for some of the most important policy sectors are included as controls in the second test of the hypothesised relationships. These variables take the value of 1 if the proposal is subjected to the labelled characteristic and 0 otherwise. All the control variables except for First-reading Council consensus are taken from the EUPOL dataset (Häge, 2011) and are measured in a similar manner as the variables employed in Häge (2007). These variables are selected as control variables in line with previous studies in the literature (Häge and Naurin, 2013; Rasmussen, 2011; Reh et al., 2013).

The control variables from EUPOL are constructed as follows. More than one policy area measures whether the proposal was treated by more than one Council configuration and/or EP committee. Any EP amendment contains information on whether the EP initiated any amendments to the proposal. Ministers involved up to first Council decision indicates whether ministers were personally involved in discussions leading up to the first Council decision on a proposal (i.e. the Council common position at the first reading) (Häge, 2007; Häge and Naurin, 2013).

First-reading Council consensus is based on voting information from the monthly summaries of Council acts. The variable is coded as 1 if the proposal was decided by consensus and as 0 if one or more member states voted against or abstained. While information concerning voting on final acts is enough for first-reading agreements, records from the first-reading (non-final) voting stage are needed for acts adopted at later stages. The latter information is scarcer for the time frame in this paper, so this variable contains some measurement bias (Høyland and Wøien Hansen, 2014). Even so, it may be useful to control for Council consensus since this indicator can measure controversy variation from proposal to proposal and not only between year and sector as captured by the Distance between EP median-Council pivot variable.⁸ The latter argument rests on the assumption that negative Council votes are synonymous with controversy in the sense that preferences had to be particularly compromised in the bargaining.

The descriptive statistics of the variables included in the second test are shown in the Online Appendix.⁹

Both tests employ a logistic regression model due to the binary nature of the dependent variable First-reading agreement. Test diagnostics confirm that the logit models are appropriately specified and fitted to the data. Since the inverse-logistic function is curved, the expected difference in y corresponding to a fixed difference in x is not a constant (Gelman and Hill, 2007). The probability that First-reading agreement equals 1 is modelled as

\Pr (Y_{i} = 1) = {logit}^{- 1} (β x_{i})

(1)

Here, the assumption is that the outcomes y_i are independent given these probabilities (Gelman and Hill, 2007).

Results

The theoretical expectations of the analysis are that (1) greater distance between the pivotal actors and (2) greater salience reduce the probability of first-reading agreements. Tables 1 and 2 present the results of the two tests. The odds ratios of the logit models are larger (smaller) than 1 to indicate a positive (negative) effect. Both tests confirm the two theoretical expectations: the greater the salience of a dossier and the greater the distance between the EP median and the Council pivot, the lower the odds of first-reading agreement. These findings also hold when relevant controls are added to the second test. The Marine Strategy Directive (2005/0211/cod) and the legislation related to the programme of community action in the field of public health (2000/0119/cod) can serve as illustrative cases for the hypothesised relationships. The first piece of legislation has high values on the variables measuring salience and distance between the pivotal actors in both datasets and was approved by the Council at the second reading after the EP had issued amendments to the Council’s common position. The second illustrative case is listed with high salience (45 reasons provided for its adoption) and large distance between the pivotal actors across the Council and the EP (1.11 on the 0.28–1.12 scale) in the proposal-specific dataset. This proposal was adopted after more than two years of negotiations and had to be settled in the conciliation committee (at the third reading).

Table 1.

Effect of salience-weighted positional difference between EP median and Council pivot on first-reading agreement.

	(1)	(2)	(3)	(4)
	First reading	First reading	First reading	First reading
	List-wise deletion		MICE
Salience-weighted positional distance between EP and Council Constant	0.988**	0.988**	0.989**	0.989**
	(−1.97)	(−1.97)	(−1.99)	(−2.20)
	0.683	0.683	0.642	0.642
	(−1.43)	(−1.00)	(−1.75)	(−1.22)

Observations (issues)	162	162	173	173
Proposals	62	62	64	64
Log likelihood	−98.07	−98.07	−103.7	−103.7

p < 0.10, **p < 0.05, ***p < 0.01.

Exponentiated coefficients (odds ratios); t statistics in parentheses. Models 2 and 4 clustered at proposal level. Extreme outliers are removed from the analysis.

Table 2.

Determinants of first-reading agreement.

	(1)	(2)	(3)	(4)
	First reading	First reading	First reading	First reading
Distance EP median and Council pivot	0.160***	0.309**	0.232**	0.239**
	(−4.08)	(−2.33)	(−2.32)	(−2.05)
Salience	0.946***	0.946***	0.972**	0.965***
	(−6.48)	(−5.71)	(−2.35)	(−2.81)
More than one policy area		6.985***	5.489***	4.555***
		(9.61)	(6.63)	(5.72)
Any EP amendment		0.197***	0.439**	0.550
		(−4.67)	(−2.02)	(−1.43)
First-reading Council consensus		0.831	0.706	0.727
		(−0.77)	(−1.11)	(−0.98)
Ministers decided 1st decision			0.0421***	0.0446***
			(−12.39)	(−11.70)
Transport-telecommunications				0.381***
				(−2.90)
Competitiveness				0.830
				(−0.60)
Economic and financial affairs				1.286
				(0.51)
Agriculture and fisheries				9.109***
				(3.21)
Constant	13.16***	12.74***	23.75***	24.94***
	(7.98)	(4.70)	(4.75)	(4.53)
Observations	656	656	656	656
Log likelihood	−396.8	−328.6	−229.5	−216.4

p < 0.10, **p < 0.05, ***p < 0.01.

Exponentiated coefficients (odds ratios); t statistics in parentheses.

Observations that have a disproportionately high influence on the results (i.e. observations with high Cook’s Distance values) are removed in the first test. A closer inspection of the DEU data reveals that these outliers are issues in multi-issue proposals (i.e. proposals with more than two issues) where the salience-weighted positional distance between the EP and the Council pivot shows the opposite tendency of the other issues contained in the proposal.

With regard to the first test, four different models are estimated (see Table 1). Models 1–2 delete issues with missing values while Models 3–4 impute missing issue positions. Since the issues are nested within proposals, Models 2 and 4 are clustered at the proposal level. The results are similar across the four models. Consider Models 1–2, a one unit (on the 0–100 scale) increase in the salience-weighted positional distance between the EP median and the Council pivot reduces the odds of first-reading agreement by 1.2% ((1−0.988)100). This effect is significant at the 5% level and is plotted on the basis of predicted values in Figure 4. Figure 4(a) shows the result when list-wise deletion is used while Figure 4(b) shows the result when missing positions are imputed. Both figures show a clear negative tendency; the predicted probability of first-reading agreement decreases as the salience-weighted positional distance between the EP median and the Council pivot increases.

Figure 4.

Predicted probabilities: salience-weighted positional distance EP median-Council pivot. (a) List-wise deletion of missing issues and (b) imputation of missing issues.

The second test includes relevant controls and assesses the effects of the two key indicators by separate measures instead of one combined measure. The results shown in Table 2 confirm the finding from the first test. The greater the distance between the EP median and the Council pivot and the greater the salience of a proposal, the lower the odds of first-reading agreement across the institutions. Recall that the variable Any EP amendment controls for uncontroversial first-reading agreements (the ones where the EP did not make any amendments to the Commission proposal). The results also hold when this variable is included (Models 2–4).

Consider Model 4 in Table 2 where all controls are included: a one unit increase in the distance between the EP median and the Council pivot on the left–right dimension reduces the odds of first-reading agreement by a factor of 0.239. Furthermore, a one unit increase in salience (meaning in the number of reasons provided for its adoption by the initial Commission proposal) reduces the odds of first-reading agreement by 3.5% when controlling for the other variables. In other words, the odds of first-reading agreement are reduced by a factor of 0.965.

The substantial effects are robust across the four different models. With regard to the control variables, the findings are in line with previous research. EP amendments and ministerial involvement are negatively associated with first-reading agreement (Häge and Naurin, 2013; Rasmussen, 2011). Interestingly, if a proposal is on the agenda in more than one policy area, the odds of first-reading agreement increases. This may be explained by the notion that inter-policy area proposals can be more subjected to package deals and thus adopted more urgently than other proposals (Kardasheva, 2013). There are also some differences across the policy areas. While proposals on transport issues reduce the odds of first-reading agreement, proposals belonging to agriculture and fisheries increase the odds of early agreement.

Figure 5 plots the substantial effects on the basis of the predicted values from Model 3 in Table 2 (the other variables are fixed at the mean). Figure 5(a) illustrates that preference divergence between the pivotal actors in the two institutions is negatively associated with first-reading agreement. However, the negative relationship with the dependent variable is more uncertain for high levels of preference divergence. Similarly, Figure 5(b) shows that salient dossiers are negatively related to first-reading agreement. As the level of salience reaches its maximum value, the probability of early agreement drops.

Figure 5.

Predicted probabilities: EP median-Council pivot distance and salience: (a) EP median-Council pivot distance and (b) salience.

The substantial effects from Model 3 (Table 2) are further explored in Figure 6. Although theoretically expected, there is no significant interaction between Salience and Distance between EP median and Council pivot (see Table 5 in the Online Appendix).¹⁰ Still, Figure 6(a) shows that the effect of preference divergence does vary with salience when low and high levels of salience are considered. Non-salient legislation with low preference divergence is predicted to be around 0.3 times more likely to be adopted at the first reading than highly salient legislation with low preference divergence. Levels of salience also matter when the preference divergence between the pivotal actors increases. Even so, both slopes show a negative trend regardless of salience levels. Figure 6(b) illustrates that the effect of preference divergence also varies by ministerial involvement. If ministers are not involved in the decision making and the preference divergence is low, the probability of first-reading agreement is substantially higher than when ministers are involved and the preference divergence is low. The difference in predicted probability also holds when the distance between the EP median and the Council pivot increases. This analysis thus confirms the negative relationship between ministerial involvement and first-reading agreements identified by Häge and Naurin (2013).

Figure 6.

Predicted probabilities: Distance EP median-Council pivot by salience and ministerial involvement: (a) by salience and (b) by ministers decided first decision.

In Model 4 (Table 2), the odds of first-reading agreement are reduced by a factor of 0.0446 when ministers are involved. Häge and Naurin (2013) find that first-reading agreements have a negative effect on the likelihood of ministerial involvement, and argue that this reduces the democratic legitimacy of these agreements. While this claim is supported by the analysis in this article, robust evidence is also found for the notion that final decisions on highly salient and controversial proposals are not taken in the first reading. On this type of proposal, ministers are more likely to be involved. Taken together, this shows that salient and controversial cases are associated with transparent decision making.

Conclusion

Standard models of veto bargaining show that if the actors in EU bargaining have complete information with regard to the positions of the other actors, agreement at stages after the first reading should not be observed. Although the number of first-reading agreements under the ordinary legislative procedure (co-decision) has steadily increased since the introduction of the Amsterdam Treaty, agreement across the EP and Council is frequently reached at later stages than the first reading. The assumption of complete information can thus be relaxed in the EU setting.

Veto bargaining models with incomplete information infer two propositions that can be applicable to bargaining scenarios with the possibility of more than one round: (1) the greater the distance between the ideal points of the pivotal actors and (2) the greater the salience, the greater the probability of more than one round of bargaining.

The analysis in this paper confirms both expectations and shows that preference divergence and salience are negatively associated with first-reading agreements in the EU. When legislative proposals are highly salient and the distance between the EP median and the Council pivot is large, agreement is significantly less likely to be reached at the first-reading stage. These findings are robust across two different empirical tests. The first test employs positional data at the issue level while the second utilises proposal-specific data and introduces relevant controls.

Whereas previous research on the determinants of early agreements in the EU has mainly emphasised the negative consequences for democratic legitimacy, this analysis contributes to the positive side of the debate by showing that highly salient and controversial cases are more likely to be resolved in a formal and transparent manner with the involvement of accountable actors. Less salient and less controversial acts are, however, associated with informal trialogue bargaining at the first-reading stage. This analysis is based on 656 proposals adopted in the fifth and sixth term of the EP (1999–2009) and on a small sample of 64 proposals adopted before and after the 2004 enlargement.

The issue of incomplete information in EU decision making is well acknowledged. Even so, few studies have attempted to specifically model the uncertainty. This article shows that predictions generated from veto bargaining models with incomplete information can contribute to explain first-reading non-agreements in the EU. Future studies may thus benefit from investigating the theoretical and empirical implications of incomplete information and the effect of these implications on EU decision making.

However, the increasing number of first-reading agreements can serve as an argument against this path of possible future research. While first-reading decisions accounted for only about 25–30% of all legislation made under the co-decision procedure in 2000 and 2001, around 80% of all co-decision proposals were adopted at first reading in the period from 2009 to 2011 (Hix and Høyland, 2013). There could well be several explanations for this trend. Enlargements and a large legislative workload demand efficient decision making. Furthermore, legislators wish to finalise as much as possible before the term ends. Another explanation would be a learning effect from working closely together over a longer time frame. The actors involved may thus have more information with regard to each other’s types. This may again result in less incomplete information and higher levels of trust.

Even so, lessons from veto bargaining models with incomplete information can contribute to explain the incidences where first-reading agreement cannot be established across the institutions. In the second test of this paper, first-reading agreements accounted for 64% of the legislation. Hence, the empirical analysis has accounted for the tendency of first-reading agreement occurring more frequently than first-reading non-agreement. The task of exploring the leverage of the theoretical explanations for the seventh term of the EP is left for future research. Despite the data limitations of this analysis, applying veto bargaining models with incomplete information to the EU bargaining setting under the co-decision procedure is shown to be a more than useful application.

Footnotes

Acknowledgements

An earlier version of this article was presented at a departmental research seminar in October 2013 at the Department of Political Science, University of Oslo. I thank the participants, in particular Øivind Bratberg, Carl Henrik Knutsen, Tatjana Stankovic and Håvard Strand for their useful comments. I would also like to thank Bjørn Høyland and four anonymous reviewers for their very constructive and helpful comments.

Funding

The Research Programme on Democracy at the University of Oslo funded the author’s research.

Notes

References

Aksoy

(2012) Institutional arrangements and logrolling: Evidence from the European Union. American Journal of Political Science 56(3): 538–552.

Benoit

Laver

(2006) Party Policy in Modern Democracies, London: Routledge.

Cameron

(2000) Veto Bargaining: Presidents and the Politics of Negative Power, Cambridge: Cambridge University Press.

Cameron

McCarty

(2004) Models of vetoes and veto bargaining. Annual Review of Political Science 7(1): 409–435.

Crombez

(1996) Legislative procedures in the European Community. British Journal of Political Science 26(2): 199–218.

Crombez

(1997) The co-decision procedure in the European Union. Legislative Studies Quarterly 22(1): 97–119.

Crombez C and Hix S (2013) Legislative activity and gridlock in the European Union. Working paper. Available at: http://personal.lse.ac.uk/hix/Working_Papers/Crombez_Hix_Gridlock-5May2013.pdf (accessed 17 June 2014).

Cross

(2013) Everyone’s a winner (almost): Bargaining success in the Council of Ministers of the European Union. European Union Politics 14(1): 70–94.

Döring

(2013) The collective action of data collection: A data infrastructure on parties, elections and cabinets. European Union Politics 14(1): 161–178.

10.

Farrell

Heritier

(2003) Formal and informal institutions under codecision: Continuous constitution-building in Europe. Governance: An International Journal of Policy, Administration and Institutions 16(4): 577–600.

11.

Farrell

Heritier

(2004) Interorganizational negotiation and intraorganizational power in shared decision making: Early agreements under codecision and their impact on the European Parliament and Council. Comparative Political Studies 37(10): 1184–1212.

12.

Fearon

(1995) Rationalist explanations for war. International Organization 49(3): 379–414.

13.

Gelman

Hill

(2007) Data Analysis Using Regression and Multi-level/Hierarchical Models, New York: Cambridge University Press.

14.

Häge

(2007) Committee decision-making in the Council of the European Union. European Union Politics 8(3): 299–328.

15.

Häge

(2011) The European Union policy-making dataset. European Union Politics 12(3): 455–477.

16.

Häge

Kaeding

(2007) Reconsidering the European Parliament’s legislative influence: Formal vs. informal procedures. Journal of European Integration 29(3): 341–361.

17.

Häge

Naurin

(2013) The effect of codecision on Council decision-making: Informalization, politicization and power. Journal of European Public Policy 20(7): 953–971.

18.

Hertz

Leuffen

(2011) Too big to run? Analysing the impact of enlargement on the speed of EU decision-making. European Union Politics 12(2): 193–215.

19.

Hix

Høyland

(2013) Empowering the European Parliament. Annual Review of Political Science 16: 171–189.

20.

Hix

Noury

Roland

(2006) Dimensions of politics in the European Parliament. American Journal of Political Science 50(2): 494–511.

21.

Honaker

King

(2010) What to do about missing values in time-series cross-section data. American Journal of Political Science 54(2): 561–581.

22.

Høyland

Wøien Hansen

(2014) Issue-specific policy-positions and voting in the Council. European Union Politics 15(1): 59–81.

23.

Hug

(2010) Selection effects in roll call votes. British Journal of Political Science 40(1): 225–235.

24.

Kardasheva

(2013) Package deals in EU legislative politics. American Journal of Political Science 57(4): 858–874.

25.

König

Luig

(2012) Party ideology and legislative agendas: Estimating contextual policy positions for the study of EU decision-making. European Union Politics 13(4): 604–625.

26.

Little

RJA

(1992) Regression with missing X’s: A review. Journal of American Statistical Association 87(420): 1227–1237.

27.

Moser

(1996) The European Parliament as a conditional agenda setter: What are the conditions? A critique of Tsebelis (1994). American Political Science Review 90(4): 834–838.

28.

Rasmussen

(2011) Early conclusion in bicameral bargaining: Evidence from the co-decision legislative procedure of the European Union. European Union Politics 12(1): 41–64.

29.

Rasmussen

Reh

(2013) The consequences of concluding co-decision early: Trilogues and intra-institutional bargaining success. Journal of European Public Policy 20(7): 1006–1024.

30.

Reh

Heritier

Bressanelli

(2013) The informal politics of legislation. Explaining secluded decision making in the European Union. Comparative Political Studies 46(9): 1112–1142.

31.

Schneider

Finke

Bailer

(2010) Bargaining power in the European Union: An evaluation of competing game-theoretic models. Political Studies 58(1): 85–103.

32.

Scully

(1997) The EP and the co-decision procedure: A reassessment. Journal of Legislative Studies 3(3): 58–73.

33.

Shackleton

Raunio

(2003) Codecision since Amsterdam: A laboratory for institutional innovation and change. Journal of European Public Policy 10(2): 171–187.

34.

Stasavage

(2004) Open-door or closed-door? Transparency in domestic and international bargaining. International Organization 58(4): 667–704.

35.

Steunenberg

(1994) Decision making under different institutional arrangements: Legislation by the European Community. Journal of Institutional and Theoretical Economics 150(4): 642–669.

36.

Thomson

(2011) Resolving Controversy in the European Union: Inputs, Processes and Outputs in Legislative Decision-Making before and after Enlargement, Cambridge: Cambridge University Press.

37.

Thomson

Arregui

Leuffen

(2012) A new dataset on decision-making in the European Union before and after the 2004 and 2007 enlargements. Journal of European Public Policy 19(4): 604–622.

38.

Thomson

Stokman

Achen

(2006) The European Union Decides, Cambridge: Cambridge University Press.

39.

Toshkov

Rasmussen

(2012) Time to decide: The effect of early agreements on legislative duration in the EU. European Integration Online Papers (ELoP). 16(111–20.

40.

Tsebelis

(1994) The power of the European Parliament as a conditional agenda setter. American Political Science Review 88(1): 128–142.

41.

Tsebelis

Garrett

(2000) Legislative politics in the European Union. European Union Politics 1(1): 9–36.

42.

Van Buuren

Boshuizen

Knook

(1999) Multiple imputation of missing blood pressure covariates in survival analysis. Statistics in Medicine 18(6): 681–694.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB