Party and Policy Competition in the European Parliament

Now commonly viewed as the EU-level equivalent of political parties, political groups in the EP fulfil many of the same functions as their national-level counterparts. Seven political groupings existed in 2004, representing over 180 different political parties from the twenty-five member states then part of the EU. Table 1 lists the groups and their names immediately prior to the June 2004 elections, along with their seat shares and the number of national parties affiliated with each grouping.Footnote 4 Over time, party groups have become increasingly cohesive and powerful.Footnote 5 As a consequence, increased emphasis has been placed on maintaining voting cohesion in the party groups, especially the two largest political groups which operate a whipping system. This has occurred in spite of, or perhaps as a response to, increases in the size of the EP and the number of member states.Footnote 6

Table 1

Political Party Groups in the European Parliament just prior to the 2004 Election

Note: These figures represent the standing just prior to the EP elections of 2004. The 785 member total is composed of the 626 members from the fifteen pre-enlargement states and the temporary members representing the accession states since 1 May 2004 prior to the June elections.

Source: European Parliament official website (http://www.europarl.europa.eu/members/archive/term5.do?language=EN).

In elections to the Parliament, EP political groups formulate their own policy positions, and the established groups have issued their own election manifestos since direct elections to the EP began in 1979. Party group policy positions have been mapped in several previous exercises but most clearly by McElroy and Benoit.Footnote 7 Very little is known about the precise manner in which policy platforms are formulated by party groups – a lacuna shared with manifesto studies in almost every other context – but anecdotally the process is driven by the demands and preferences of member parties, who have various ‘demand weights’ or degrees of influence according to delegation size, and also the process is influenced by possibly idiosyncratic factors such as seniority, nationality or the personal influence of well-positioned individuals.

Once formulated, the party groups’ policy positions serve as a guide to their legislative activity in the Parliament. Furthermore, new group members usually must agree officially to follow party policy. The rules of procedure of the ALDE group (successor to the ELDR examined here), for instance, explicitly state that new members must endorse ‘the 10 Point Program for Europe of the ALDE Group’. MEPs elected by national parties are regularly issued voting instructions by party groups and subject to the party whip. Party groups have mechanisms to punish MEPs who vote against the party group instructions, whether on individual initiative or (more likely) because of national party instructions, including demotion or even expulsion.Footnote 8 Strong divergences in policy positions between national member parties and the EP party groups could, therefore, cause persistent and potentially costly conflicts between national party and EP party group voting positions. Hix et al.’s analysis of roll-call votes from the 1999–2004 Parliament, for instance, showed that 6.6 per cent of MEPs disobeyed their party group’s instructions in order to vote with their national party.Footnote 9 MEPs nonetheless vote along party group lines over 90 per cent of the time and this political group cohesion has increased over the years.Footnote 10 Consequently, the high and increasing cohesion of MEP voting behaviour suggests that the policy congruence of national party members’ preferred positions with the positions of EP party groups should also be high. Maintaining cohesion is a constant struggle for party groups, however, because their support depends on constituent national parties whose success and survival come from diverse domestic arenas. When policy preferences between a party group and one of its national party members drift apart, the resulting incongruence creates tensions which may alter the calculus of affiliation.

National Party Dynamism and the Problem of Policy Incongruence

Detailed studies of party platforms over time have documented significant movement in party positions by major parties in the post-war period.Footnote 11 This dynamism follows naturally from the need to respond to shifts in the nature and importance of national issues, the positioning of new parties, and the repositioning of rival parties. Yet this constant motion also creates a large potential for incongruence between a party’s positions and that of the supranational EP party group to which it belongs, since EP party group positions are likely to be much more static given that they are determined by coalitions of different parties. Because membership in a party group is voluntary (and indeed some parties and MEPs may deliberately choose to remain officially unaffiliated), national parties always have the option to leave. When a national party’s position drifts away from the median position of a party group, it will not be the case, following Black’s classic expectation, that majority rule on single dimensions leads to group positions that are equivalent to the median voter’s most preferred point, because an alienated group member will find it easier simply to leave the group for another whose positions are closer to its own.Footnote 12 It is quite feasible, in fact, that a national party’s position may shift far enough away from its party group’s position that a rival party grouping’s position becomes more proximate to the national party’s preferences. When this happens, it is more likely that party groups will ‘unravel’ in the fashion similar to that described by Johnson, as a threshold of tolerable incongruence is exceeded and members withdraw because of unhappiness with the distance between its own changed position and that of their (former) party group.Footnote 13 Parties may also form new groupings when no existing group fits their programmatic needs, provided they meet some threshold requirements (nineteen members from at least five member states in 2004).

National parties, whenever possible, will attempt to steer EP group policies on issues of the greatest domestic importance towards the positions most preferred by the national parties. This ability, however – what Hearl calls ‘demand weight’ or political influence over the party group’s median position – may be quite small in the context of EP political groups, especially as the largest ones represent broad coalitions of multiple parties from many different member states.Footnote 14 When demand weight is outweighed by national political considerations, a party is likely to leave (or not join) that grouping. When this incongruence reaches a threshold, a national party will perceive the benefits of switching to a more policy-compatible party group to outweigh the costs of retaining its current less congruent affiliation.

A recent example of this process can be found in the debate within the British Conservative party over whether or not to withdraw from the European People’s Party (EPP), the main centre-right grouping in the EP. Having adopted a stronger and more explicitly anti-federalist policy position on the EU, reinforced in its 2005 manifesto and by Conservative party head David Cameron’s 2006 leadership election, the Conservatives experienced a high degree of tension between their own relatively anti-EU national party position and the pressure exerted by the EPP to vote for pro-federalist measures in the Parliament. The Conservative party eventually formed a new party grouping in 2008 with some like-minded parties from the EPP, such as the Czech Civic Democrats (ODS), whose leader Mirek Topolanek similarly declared in June 2006 that they found ‘it increasingly difficult to accept the way that French and German parties [were] voting in parliament’.Footnote 15

Consistent with this expectation that national party movement implies dynamism in party group affiliation, the empirical record of party group composition in the EP has shown high levels of change. EP groups experience a high frequency of party switching and most of this switching occurs when national political parties decide to change EP group affiliations, bringing all their MEPs along with them.Footnote 16

Realignments of affiliation, in the form of national party switching between EP groups, constitute a dynamic implication of the policy congruence explanation of party group affiliation. At any given time point, however, policy congruence also implies that the national parties affiliated together into party groups will be highly congruent on the most important dimensions of policy, as parties will have placed themselves in party groups according to the match between their policies and those of the party groups. Parties whose policies have become incongruent with those of their party group will exist, but these will be few in number. Moreover, the policy incongruent parties may also be identified as the most likely candidates for switching party groups, from a less to a more congruent parent party.

The Limitations of ‘Party Family’ Explanations

Because it is the conventional wisdom – and because its observable implications are clearly different from the policy congruence explanation set forth above – we briefly discuss the notion that party affiliation in transnational party groupings is driven by party family, determined by historical tradition and a party’s set of core principles. Rooted in cleavage theory, party families categorize parties based on long-standing issue associations and positions that arise from basic political cleavages in which parties are rooted. EP party groups are then the organizational units formed to ‘broadly represent the policy positions of … the classic European “party families” ’.Footnote 17

While capturing the broad spirit of the ideological basis on which party groupings originally formed, the party family explanation is inadequate to explain national party affiliation with EP party groups. First, party family is an essentially static characterization, while national party policy and (most importantly) EP group affiliation are clearly dynamic phenomena. On the one hand, national parties frequently modify their policy positions, sometimes dramatically, and they also switch EP party groups. Party families, on the other hand, are defined by ‘essential and indelible’ issue positions and orientations that do not change between elections.Footnote 18 Because party family is a static conception, while party positions and party affiliation change regularly, party family approaches can never offer more than a very general and indirect explanation of the dynamic process of party group affiliation.

Secondly, many national parties do not belong to clearly identified party families, either because their organizational basis does not correspond to the classic cleavages identified by Lipset and Rokkan or because they are formed primarily around new (single) issues such as European integration or immigration.Footnote 19 This situation characterizes many of the parties from the eight post-communist member states whose young party systems had developed for less than fourteen years at the time of this study. In Hungary, for instance, the Christian Democratic People’s party (KDNP) has been characterized as populist, the populist Hungarian Democratic Forum (MDF) has been classified as Christian Democratic, and the originally liberal Fidesz (Federation of Young Democrats, later Fidesz-Hungarian Civic Party) recast itself as a populist, right-of-centre party. Attempting to classify party families in Poland or Bulgaria – where even keeping track of the latest set of parties and party names is not for the faint-hearted – is even more problematic. In short, party family categories developed from cleavage patterns in West European democracies mean something very different when applied to the party systems of the EU’s post-communist member states.

Finally, while party family does provide a rough measure of party policy positions, within party family categories are parties with quite diverse positions on policy, especially on ‘new politics’ issues such as the environment, immigration or European integration. When direct measures of party policy are available on multiple dimensions, then these will always provide more information than party family categories, which are low information, categorical in nature, and apply poorly to new parties and parties from post-communist states. For these reasons, party family cannot provide a satisfactory explanation of party group affiliation in the EP.

Conditional Logit Model of Party Group Choice

Party group affiliation involves choosing between alternatives, and given the difficulties of establishing new groupings, for all practical purposes this entails choosing from between available alternatives. For national parties considering affiliation, therefore, the driving factor should be the relative distance of alternatives: how close a party group’s position is to a national party’s own position, compared with other party groups’ positions. To test this explanation, we develop and apply a discrete choice model to the question of party group affiliation, using relative policy congruence on multiple dimensions of policy to predict party group choice.

Our model uses conditional logit regression, based on fitting N = 146 national parties for which we observed fully non-missing policy data to the K = 7 political groups in the EP. An attractive feature of the conditional logit model is that it permits both choice-level variables as well as individual-level variables, and can be motivated by a random utility model.Footnote 25 The corresponding utility function associated with the kth choice in this mixed model is given by:

(1)

where z_ik refers to the choice-varying covariates and x_i refers to individual varying covariates. The utility to party i of choice k is linear in some predictors but also contains a random component. An outcome y_ik is observed for each party group category k as:

(2)

The conditional logit model derived by McFadden relates the choice probabilities to u_ik according to:

(3)

when y_i∼multinomial(p_i).Footnote 26

In our model, the key independent variables are choice-related, consisting of the absolute difference between the national party’s mean point (on a given dimension) and the mean policy position of the political group under consideration. The variable thus takes on different values dependent on the response category for each national party. These variables are:

• — Policy distance (a choice-level variable). This variable measures the absolute difference between party group k’s position and the policy position of the national party i on six commonly measured dimensions: Left–Right; Taxes–Spending; EU–Authority; Environment; Social-Liberalism; and Deregulation. Our argument suggests that national parties will seek to minimize these distances in order to achieve policy congruence between themselves and the political groups, and therefore we expect the signs on these coefficients to be negative. The greater the distance between party i and political group k, the less likely it is to choose to affiliate with this group.

• — Decentralization position (individual-level variable). This variable captures the national party position on the decentralization dimension, where 1 indicated that the party favoured decentralized decision making, and 20 indicated that a party opposed decentralization. This variable was re-centred by subtracting it from the scale midpoint of 10.5. This variable could not be included as a choice level variable as there was no measure of decentralization included in the expert survey on the positioning of the EP political groups. Nonetheless, the national party position on decentralization is included because national parties’ positioning on this issue still identifies an important policy dimension around which parties might coalesce into like-minded groups, even if as an individual-level variable it does not capture proximity to the nearest EP group.

• — National party vote share (individual-level variable). This variable measures the percentage of votes that party i received in the most recent national election. This measure is included to capture the concept that large national parties (regardless of the number of MEPs they elect) may be predisposed to association with the larger political groups in the EP for reputational and influence reasons. Key parliamentary positions such as chairmanships and membership of the Bureau are ‘owned’ by the larger groups. So, for instance, when the governing British Conservative party switched to the (large) EPP in 1992, it moved out of a much smaller group in the Parliament, the ED, which had just 6 per cent of the EP seats. Sitting Conservative party MEPs voiced similar concerns in the recent negotiations over the formation of a new group in the EP, one party insider commenting that ‘it’s not really in Britain’s interests if the Tories quit the biggest group in the Parliament’.Footnote 27

Two estimation methods were used to fit the model. The first uses maximum likelihood (estimated via the STATA 9 package). The second fits a general MCMC-based Bayesian model that allows the α parameters to be estimated separately for each party group, k. Using (diffuse) multivariate normal priors, this model was run for 80,000 iterations, discarding the first 50,000 as burn-in, and mixing three chains set with random starting values.Footnote 28 The advantages of fitting the Bayesian model for our context are several. First, it permits us to estimate separate (random) effects from each choice variable for each party group. Because political groups attach different levels of importance to different policy dimensions, we have strong substantive reasons to believe that some policy incongruities will operate differently for certain groups. For instance, the effect of policy distance on the environmental dimension should be different for the Green (Verts) party group versus the anti-European (EDD) party group, for which environmental policy is not a high-salience issue. Secondly, the Bayesian approach permits full characterization of the posterior distribution for the quantities of interest, which includes not only the variable coefficients but also the predicted probabilities for each case, including the ratio of predicted probabilities for a party belonging to one group versus another, something we focus attention on below.

The results of the model estimation are presented in Table 3. The coefficient estimates for the choice variables of policy distance represent the influence of policy incongruence on the probability that a given national party will be a member of a particular group. When the coefficients are negative, this indicates that as the distance between the national party’s mean and that of the European political group in question increases, the less likely the national party is to affiliate with the given party group. The confidence intervals represent the uncertainty around the coefficient estimates, providing evidence of a real effect when the interval does not include zero.

Table 3

Conditional Logit Estimates of Party Group Choice

For the policy distance variables in the first model, the coefficient estimates are very consistent: all are statistically significant (with the exception of Taxes–Spending) and in the expected negative direction. The greater the distance between a national party’s position and the position of a political group in the EP, on a given policy dimension, the less likely the national party is to be a member of this party group. Tax distance fails to reach significance in either of these two models, possibly because it is highly collinear with the Left–Right and Deregulation dimensions. Decentralization is only significant for the Green grouping, which is not altogether surprising given that this party is the only one fully committed to the concept of subsidiarity, with a policy platform centred on a commitment to deepen democracy by decentralization. This commitment makes the group attractive to regional parties lobbying for further autonomy.

The second column (Model 2) reports the results of the more general MCMC model in which the parameters for each group were allowed to vary. These results provide richer detail about the significance of policy distance for the individual groups and their affiliated national parties. For instance, Left–Right distance, as we would expect, is significant for two of the big political groups, though oddly not the PES. Distance on the Social dimension (attitudes to such matters as abortion and homosexuality) is significant for half of the parties (the PES, ELDR, UEN and EDD) and again in the right direction. Environmental distance, as would be expected, is significant for both of the Green groups (Verts and GUE). Attitudes to levels of EU authority are again significant for the two anti-federalist groups, the UEN and EDD. Finally, deregulation is significant and in the expected negative directions for three Groups, the PES, GUE and EDD. The individual level policy variable ‘Decentralization’ fails to reach significance for any political group. While the lack of a baseline category makes it more difficult to detect statistical significance in these coefficients, this more informative model nonetheless forms the basis for generating predicted probabilities of group membership for each party, by drawing from the posterior distribution. On the basis of these predictions, it is then possible to assess, at least from the standpoint of policy congruence, which parties do and do not fit in their existing party groups.

Evaluating Party Group Fit

The Bayesian framework makes it straightforward to generate posterior predictions of our key quantities of interest: the probability that a given party will choose a particular party group. Because these probabilities can be simulated directly from the posterior distribution, they provide an alternative to other methods, such as parametric bootstrapping from maximum likelihood parameter estimates.Footnote 29 By comparing the posterior probabilities of the group predicted as having the highest probability of membership for a given party to the group of which the party is actually a member, we can also directly estimate the ratio of actual to predicted group membership as a posterior quantity, to estimate the degree of ‘fit’ (or ‘misfit’) for each national party. This quantity is of direct substantive interest to any scholar, observer or political actor interested in party group cohesion as well as being relevant to party and coalition formation generally at the European transnational level.

Based on the highest mean posterior probability of being in a party group, computed for each party from the Bayesian estimates in Table 3, we can predict a party group based on the model and compare this to parties’ actual groups. Table 4 presents a cross-tabulation of these predicted groups with actual group membership. The result is impressive: a total of 79 per cent of the national parties’ party groups are correctly predicted by the model, based on 117 of 148 parties correctly predicted by having the mean posterior , where represents the mean posterior probability of party i being in party group k. In other words, simply by minimizing policy distance – without informing the model in any way of the absolute policy positions of the parties or groups – we have predicted nearly four out of five parties correctly. As an indicator of just how impressive this predictive result is, we can compare the prediction to non-informative models. If we simply assumed a uniform probability for each party joining a group, then our baseline predictive rate would be to predict just 0.14 (or one out of seven) correctly. Of course, we could do much better by simply choosing, for every party i, the largest or modal category of party group in our sample (here, the EPP with forty-three national parties from the 148), which would correctly predict 0.29 of the national parties’ group alliances. Measured against this benchmark, our model represents an increase in correct prediction of an additional 50 per cent of the sample. Furthermore, if we exclude as a ‘prediction’ any posterior mean ratios whose highest posterior densities include 1.0, then only five parties are incorrectly predicted, for a percentage correctly predicted of 0.966.Footnote 30

Table 4

Cross-Tabulation of Predicted Group Affiliation versus Actual Group Affiliation

Note: Predicted party group is determined by where represents the mean posterior probability of party i being in party group k, from Table 3.

In addition to merely predicting the most likely category, our model estimates also provide direct information on the degree of ‘fit’ for each national party according to the model’s posterior predictions. Figure 2 summarizes the fit for sixty parties by plotting the estimated ‘odds ratios’ that a party would be in its observed group: the mean posterior probability of being in this group divided by the probability of not being in this group. To reduce skew and improve interpretation, we have logged the scale using base 10 to provide the ‘logged odds ratios’. A value of 1.0 – such as that scored by the Cypriot DISI party at the top of the plot – therefore means a party was estimated at being ten times more likely to be in its actual party group than in another group. Conversely, a party like the British Labour party with a value close to −1.0 means it was ten times less likely to be in its observed group than not. In the most extreme cases, such as the five parties at the bottom of the plot whose logged odds ratios were less than −2.0, these were more than 100 times less likely to be in their observed party than not, according to the model. Finally, the dashed line through the zero point on the x-axis indicates the point at which the odds are the same for being in the observed group as any other group (since log(1) = 0), and hence this line demarcates what we have called the ‘mis’- versus ‘correctly’-placed parties. For all parties whose mean posterior (logged fit) was less than 1.0, Figure 2 thus summarizes at a glance the ‘fit’ of each national party, along with estimates of uncertainty represented by the width of the lines.

Fig. 2

Logged odds ratios of probability that a party belongs to its actual party group, from posterior distributions Note: Plots the posterior distribution of base-10 logged odds ratios of the probability of a party belonging to its actual group, according to the (Bayesian) model from Table 3. Solid circles represent posterior means, and the bars represent the 90 per cent highest posterior density region.

More insight on ‘correct’ versus ‘mis’-placement of parties into party groups can be gained by focusing on the posterior predictions of individual parties. In Figure 3, we focus the spotlight on the three main British parties, directly plotting the posterior densities of the probability of group membership for several candidate party groups. In each panel, the solid line represents the posterior density for the party group in which the party is observed, while the dashed lines represent posterior densities of the probability of this party being a member in some other party group. In the first panel, for instance, we see that the model clearly predicts Labour as better fitting in the liberal ELDR grouping – because its mass is concentrated around the (0.8, 1.0) probability region – no doubt because under Blair its well-known ‘New Labour’ policies assumed many centrist or even liberal characteristics. The Conservative party is also ‘misplaced’ since the greatest mass of its posterior density is in the anti-federalist UEN group rather than its observed group, the EPP. As we have already pointed out, this misplacement is not surprising given the party’s pledge under leader David Cameron (elected in 2006) to quit the EPP and their subsequent formation in 2009 of a new political group, the European Conservatives and Reformists Group. Finally, the widely distributed posterior mass for the probability of the Liberal Democrats being in their actual group, the liberal ELDR, shows that while this is the ‘best’ fitting group for this party, it is nonetheless not a great fit. The Liberal Democrats could also have fitted into the PES, or even – although considerably less likely according to the model – the Greens.

Fig. 3

Posterior probability densities for British parties Note: Plots individual posterior densities for the probability of each party being in the given party group, from a Bayesian model shown in Table 3.

As a final insight into the model’s ability to evaluate party group fit, we examine a case where French MEPs from two national parties each split into two different party groups. From the Union pour la Démocratie Française (UDF), seven deputies moved from the ELDR to join the EPP, while one deputy refused and remained affiliated with the former party group. Similarly, in 2004, French MEPs from the Rassemblement pour la France (RPF) were split between two party groups, three joining the EDD and two affiliating with the UEN. By examining the posterior densities of the probabilities of each party’s membership in the different groups, we can assess which party faction is in the ‘wrong’ party group. Figure 4 plots these densities for each party group for the UDF and RPF. In the case of the UDF, this national party is clearly much more likely to belong to the EPP, and hence the stubborn MEP remaining in the ELDR either does not belong or has very different policy preferences from his national party. This is something which our data and model cannot determine, of course, since our examination focuses on parties rather than individuals, but it does suggest that this lone member is in one way or another ‘out of sync’ with his national party. In the case of the RPF, the posterior densities have an almost identical mass over the middle regions of probability, suggesting that one group is basically as good as the other from the standpoint of policy congruence.

Fig. 4

Posterior probability densities for two French parties, each split between two EP groups Note: The UDF (Union pour la Démocratie Française) was split with seven members in the EPP and one in the ELDR; the RPF (Rassemblement pour la France) was split with three members in the EDD and two in the UEN.