The Council of Ministers has the final say on the adoption of EU legislation affecting the lives of 400 million EU citizens. Yet for such an important institution comparatively little is known about how its decisions are reached. Decision-making processes are opaque: Council members deliberate behind closed doors, no press conferences are held and no meeting transcripts are released. For citizens and scholars alike, the Council's decision-making process is ‘a secretive and specialized affair’.Footnote 1
However, this situation is changing. EU scholars focusing on the Council have produced an impressive and varied literature, delivering foundational insights into the workings of the Council. In the article we assess the contribution of two approaches, applied theoretical models and voting studies. Acknowledging the strengths and weaknesses of both approaches, we argue that two recent trends may potentially alter the state of the field. First, the Council, bowing to internal and external pressure to increase transparency if not accountability, has begun to increase the scope and scale of the documents that it makes publicly available. Second, EU scholars have started to import varied and increasingly sophisticated data-generation tools from across political science and beyond. For these reasons König, Luetgert and Dannwolf conclude that ‘research on legislative decision-making in the European Union has entered the stage of quantitative analysis,’ with newly available data sources combined with appropriate tools potentially enabling ‘rich and systematic data analyses’.Footnote 2 The aim of this article is to take stock of these developments and to speculate on the next phase of research on the Council.
COUNCIL DECISION-MAKING STUDIES
Before a Commission draft becomes EU legislation, it has to be adopted by the Council. Under the main EU legal procedures to adopt legislation, i.e. the cooperation, consultation and co-decision procedures, the Commission prepares a legislative proposal and submits it to both Council and European Parliament. The type of the legislative procedure then determines the timing and level of support required for a decision on the proposal.Footnote 3 As with other areas of political science, research employing quantitative methods has become more prominent in attempts to explain decision-making outcomes in the Council. Approaches such as formal modelling, though increasingly visible, are still relatively specialized and lack overlap with the broader research community of EU scholars. The following sections therefore provide a non-technical review of the state of the field in voting and formal models of decision-making in the Council.
Applied Theoretical Models
Over the last 15 years, EU scholars have developed and applied a considerable range of theoretical models to decision-making outcomes in the Council of Ministers.Footnote 4 The majority of EU decision-making models are based on rational choice assumptions of human behaviour with solution concepts derived from game theory. All models assume that decision outcomes are the result of interactions between goal-oriented actors operating withininstitutional constraints, but a broad distinction can be made on the basis of the emphasis given to the formal procedural aspects of EU decision-making (e.g. voting weights, decision rules, etc.), or the informal bargaining that takes place prior to the adoption of legislative proposals. This distinction gives rise to procedural and bargaining models.
Procedural models conceive institutions as constraining behavioural opportunities by determining the identity of players, the strategies available to them, the sequence of play and the aggregation rules whereby players' choices are translated into decision outcomes. This class of model is rooted in non-cooperative game theory and spatial voting theory and stresses both the sequential features of the legislative process and the decision-making powers that actors can utilize to effect advantageous outcomes. Actors are differentially empowered in accordance with their voting power or with a first-mover advantage. The structure of a particular procedural model applied to the Council of Ministers depends on the type of legislative procedure under investigation, for example consultation or co-decision.Footnote 5 By contrast, bargaining models privilege behind-the-scenes informal negotiations rather than institutional arrangements, though formal decision-making rules channel actors' interests and partially define actors' evaluations of other stakeholders' capabilities and inform attempts to build coalitions around their own preferred policy positions. In other words, ‘procedures don't determine behaviour but set the boundaries within which the action takes place.’Footnote 6 Bargaining models primarily employ cooperative game theory and do not specify the exact sequence by which decision outcomes are reached. Instead, the bargaining process is conceived as a black box into which actor preferences, the importance they attach to the relevant issues and their capabilities are entered.
In the recent work of Thomson, Stokman, Achen and König,Footnote 7 a wide range of procedural and bargaining models are subjected to empirical tests that reveal a substantial deficit between model predictions and observed outcomes; even the best-performing models fail to predict very accurately. However, bargaining models do consistently and decisively outperform their procedural counterparts and those bargaining models that conceive of actor interaction during the negotiation process in more cooperative terms are the best performers overall.Footnote 8 These results can be interpreted as supporting the conclusions of studies employing more empirical approaches, i.e. that unanimity is ‘a strong norm in the EU’, where the highly iterative nature of day-to-day decision-making, combined with a lack of stable patterns of coalition formation, ‘strongly facilitate the universally inclusive, compromise mode of decision-making’.Footnote 9
Thus, whilst bargaining models can correctly diagnose consensual decision-making as the modus operandi in the Council of Ministers, as well as providing a theoretical explanation for such behaviour, their low predictive accuracy remains a problem.Footnote 10 We speculate that the potential source of this problem lies in the input data that is used to test the models. Most model evaluations rely on data generated by elite interviews, an expedient choice given the precision of data requirementsFootnote 11 but one that suffers from scaling problems and post-dictive bias.Footnote 12 Given high costs in terms of time and financial resources, the majority of data-collection efforts are restricted to a small number of legislative proposals compared to the absolute size of EU legislative output.Footnote 13 Existing data may not, therefore, accurately reflect the EU's day-to-day decision-making process, which in turn may hinder the satisfactory evaluation of applied models.
Council Voting Studies
The decision whether to adopt a proposal under qualified majority voting (QMV) or unanimity depends on the legislative procedure involved. If a proposal reaches the voting stage, roll-call records – indicating whether a Council member voted for or against the proposal, or abstained from the vote – are then issued in the official Council Minutes. The major finding of studies based on the analysis of voting records is that the majority of decisions are made unanimously, even under those legislative procedures where there is legal provision for explicit voting under QMV. Hayes-Renshaw, van Aken and Wallace report that, between 1994 and 2004, 75 per cent to 80 per cent of decisions technically subject to QMV were not contested at the voting stage.Footnote 14
Not surprisingly given this strong finding, one question that has preoccupied scholars is why recourse to voting is so infrequent, or, alternatively, why unanimity is the most frequent method of decision-making in the Council. Heisenberg's explanation emphasizes the ‘culture of consensus’ among member states resulting from decades of frequent negotiations between the same partners and the fast acculturation of new members to a common understanding of the historical importance of the EU project.Footnote 15 She also notes that members of the Council are frequently insulated from domestic constituencies, a particular bone of contention for proponents of greater accountability for this institution, and therefore have the freedom to negotiate the substance of an issue without the pressure of spinning it to an electorate.
Mattila and Lane, by contrast, stress the strategic expedience of issue linkages. Indeed they see vote-trading, or logrolling, as the fundamental factor driving a bargaining process where actors decide on several issues simultaneously and exchange their votes based on the different levels of importance attached to the multiple issues under consideration.Footnote 16 Logrolling works because, given the diversity of member state interests, any one legislative proposal is unlikely to be equally salient across the whole range of actors, who may then be willing to trade support on proposals of lesser importance in exchange for future reciprocal support on issues more salient to them.Footnote 17 Mattila and Lane argue that the iterative nature of negotiations, and the absence of both a government–opposition structure and a coalition that forms a stable majority on most proposals over time, significantly reduce the obstacles to vote-trading in the Council.Footnote 18 In spite of ample research on the extent to which members vote on issues or in terms of underlying dimensions (e.g. integration–independence, left–right, net-contributor–net-benefactor),Footnote 19 studies based on roll-call data demonstrate weak evidence of stable coalition behaviour. In 47 per cent of cases, according to Hayes-Renshaw et al., abstaining or voting against a proposal is limited to a single member state,Footnote 20 an action that may be interpreted as an indication of powerfully held alternative preferences,Footnote 21 future non-compliance, or simply for the consumption of Eurosceptics at home.Footnote 22
Insightful as voting studies in the Council are, they are nevertheless unable to explain exactly how decision outcomes were reached through the exchange process, since they look solely at the final voting stage, when any potential vote trades have already been decided upon. While the voting-study narrative of intense behind-the-scenes bargaining below the ministerial level, with representatives hammering out consensus positions so that decisions taken in the Council are usually a formality, is in line with reports from the field, it is a conclusion that is based on data that are silent on the bargaining process itself. The use of Council roll-call votes is effectively subject to a censoring bias, since only those legislative proposals that have not been withdrawn at an earlier stage of the institutional process are voted on.
NEW APPROACHES TO EU COUNCIL STUDIES; EVOLVING METHODS AND DATA
Formal and voting models have delivered important insights and theoretically driven explanations for decision-making in the Council. Rather than indicating deficiencies in the two approaches, we suggest that some of the weaknesses noted above are the result of difficulties that researchers have faced in generating appropriate data; a situation that may be improved by the increasing availability of primary documents and the application of text analytical methods.
Primary Documents: The EU's Legislative Sources
Until recently, discussion of text-analytical techniques applied to analysis of Council decision-making was largely irrelevant as a result of the scarcity of source materials. However, there are now two main ways of accessing information pertaining to the legislative process, through the Council's Public Register and the Eur-Lex database.Footnote 23
The most useful document available through the register is the Council Minutes, which provide comprehensive information on the passage of a legislative proposal.Footnote 24 Though the Council Minutes are summaries of decisions rather than more revealing transcripts of the discussions, which are currently still held behind closed doors, this documentation at least reveals the timing and content of decisions and includes statements by the individual member states and the Commission, potentially indicating deviation from consensus positions voiced in the formal statements by the Council itself. Further information is available in the form of monthly summaries of Council acts, which include the legislative acts adopted, the voting rule and the results of the voting, in addition to formal statements by member states. Minutes have been made available online covering legislative decisions from 1999 to the present date.
Eur-Lex is a legislative database covering inter-institutional procedures and as such contains information on the inputs and outputs of the Council as part of a time line for Commission proposals.Footnote 25 König et al. conclude that the database ‘enables [researchers] to follow the major stages of the decision making process between the Commission and the other institutions’, and allows researchers to track ‘changes in the application of particular procedures or voting rules in specific sectors and identify general trends in the quantity and type of legislative decisions over time’.Footnote 26 Such results are of relevance to theoretical questions posed by practitioners of various approaches, such as delimiting the influence of the European Parliament and uncovering factors influencing decision outcomes such as the effect of specific procedural rules, institutional involvement in different policy sectors and instances of difficult or controversial proposals.
Data-Generation Techniques
The availability of primary documents increases the need for appropriate data-generation techniques. While content analysis approaches have delivered valid and reliable quantitative data for numerous applications across political science, application, and even discussion, has been largely absent in research on Council decision-making. Consideration of text-analytical methods as a source of information on Council proceedings is now a timely exercise.
Content analysis is a flexible method of data generation employed throughout the social sciences,Footnote 27 increasing in visibility with the development and successful application of computer-assisted content analysis techniques. While acknowledging Benoit and Laver's warning that ‘it would be silly to claim that text analysis is a panacea for all research problems,’Footnote 28 it would be equally foolish to dismiss an approach that has delivered reliable data at low cost for a range of diverse ends. In the case of Council decision-making, deriving information on actor positions from political texts would be very useful. One such fully automated technique, wordscores,Footnote 29 produces estimates of actor positions based on the comparison of relative word frequencies in ‘reference texts’ that are agreed to indicate positions on a given policy dimension and in ‘virgin texts’ about which nothing is known beforehand. The technique is not language specific, it handles large quantities of text and has various published applications.Footnote 30 Wordscores is not without its problems,Footnote 31 but its application to the analysis of political texts at the domestic level is worth exploring, given the importance of two-level dynamics on legislative behaviour suggested by formal and voting studies.Footnote 32
Alternative approaches to quantitative content analysis include parsing techniques that employ software to break sentences down into their semantic units.Footnote 33 This approach yields more information on the interaction of actors and issues than the wordscoring approach although applications are at an early stage. Nonetheless van Atteveldt et al. were able to determine the degree of political conflict and issue positions from newspaper articles during Dutch parliamentary elections in 2006.Footnote 34 A further application in this tradition, latent semantic analysis, uses a kind of factor analysis to measure the semantic similarity of words and text passages, approximating human performance in such tasks.Footnote 35
NEW STUDIES ON THE COUNCIL
The convergent trends of newly available sources of primary data and the adaptation of appropriate data-generation techniques is already starting to pay dividends in the research literature on the Council. For instance, Hagemann addresses concerns about the reliance of voting studies on final roll-call records by collecting information on the determinants for voting behaviour from various stages of the decision-making process.Footnote 36 Using Council documents recently made available online through the Public Register, Hagemann compares the ideal point estimations produced by NOMINATE and Bayesian Markov Chain Monte Carlo (MCMC) models. Not only does the model produce estimates of ideal points (with the advantage of reporting standard errors), it also does so for specific issues and can incorporate ‘more complex behavioural assumptions such as the number and nature of underlying dimensions, apparent party coalitions, determinants of legislator preferences and the evolution of the legislative agenda’.Footnote 37 Moreover, we support Hagemann's injunction that EU scholars should ‘make more use of the data that are already available from the Council’.Footnote 38
Another innovative approach to generating data on actor preferences appears in Franchino's study on delegation,Footnote 39 in which estimates of the preferences of Council members and the Commission are reconstructed from measures of national government positions collected as part of the Comparative Manifesto Project (CMP).Footnote 40 If, as Franchino argues, the CMP codes can be reconstructed to provide reliable estimates of member state preferences, then the CMP datasets represent an underemployed resource for Council studies. Yet another innovative approach pioneered by GolubFootnote 41 and Schulz and KönigFootnote 42 yields considerable supplementary information on Council decision-making, employing the event history approach to analyse the extent to which institutional rules increase or decrease the speed of Council negotiations. And though methodological questions remain,Footnote 43 the approach has demonstrated some facility in evaluating decision-making models. For instance, Golub shows that spatial models and coalition theory perform comparatively well and that formal voting rules do matter in the Council.Footnote 44
CONCLUSION
Attempting to explain how decisions are reached in the Council has generated an impressive literature. Summarizing the strengths and weaknesses of the major quantitative approaches, voting studies and applied formal models, we suggest that valuable insights have been gained into the opaque decision-making processes of the Council, but much work remains to be done. One area where improvements may be possible is in the generation of appropriate data; an issue that has not previously garnered much attention, presumably because the scarcity of materials released by the Council did not demand it. However, the innovative application of new data-generation techniques in combination with the increasing abundance of documentary materials made available online demonstrate the timeliness of considering this issue.
