Freezing Competition: European Integration and the Fragmentation of Party Systems in the EU

by Eric Hines University of Iowa [email protected]

Paper presented at the annual meeting of the Midwest Political Science Association, The Palmer House Hilton, Chicago, Illinois

Hines 2

Introduction The liberal conception of democracy requires political competition between rival candidates and rival programs to be effective. In the modern state, this competition is generally produced by the mechanics of competitive political parties. In theory, competition between parties provides voters with policy alternatives that allow voters to alter the direction of politics, hold politicians accountable for their actions, and give politicians an incentive for policy innovation. However, this conception of democracy assumes that there are no exogenous constraints on political competition. Critics of the “democratic deficit” in the European Union (EU) suggest that European integration has had an indirect effect on political competition in the member states (Hix 2003; Paulson 1997). They believe the creation of the common market and the European Monetary Union (EMU) have constrained democratic political competition by reducing the number of policies alternative available to political parties. For example, the parties of the left have lost their ability to promise higher social benefits and parties of the right have been prevented from making large tax cuts because governmental and opposition parties alike have been forced to advocate the same specific entrenched, centrist EU policies. If integration has constrained political competition it should lead to policy convergence between the parties of the left and right (Hix 2003; Clark et al. 2002). However, policy constraints will not apply to all parties or to voters’ demands. This may have other implications for political competition. Spatial theories of voting behavior suggest that if parties do not present clear and rival programs that meet the prior expectations of voters, their levels of support will decline. Since small parties will be

Hines 3 less constrained by EU policies in their actions, voters on the margins may be more inclined to support them. Thus, if EU policy constraints are sufficiently constraining on the large parties, increased party fragmentation may occur. In the following sections, this paper examines the importance of political competition for a democracy and how the EU may constrain it. It then develops a theory that this constraint would promote party system fragmentation that is consistent with spatial models of voting behavior. Finally, it tests two models of party fragmentation and discusses the results of those models.

The European Union and Political Competition A democratic system implies that the preferences of “the people” should prevail in political competition. In order for the popular will to prevail though, “we require functioning institutions that are designed to, and really do, secure a government responsive to public interest and opinion” (Pitkin 1967, 234). Schumpeter (1942) argues that elections are the principal institutional feature of democracy that provides this responsiveness because they create competition between rival elites for political control. This works because elections produce political competition that functions like a market. This “political market” produces efficient outcomes by providing elites (political parties) with an incentive to present rival policies and candidates that give voters a mechanism for choosing between parties (Besley and Coate 1997). However, just as markets are often imperfect, elections by themselves do not guarantee perfect political competition. Politics can be oligarchic, which is the case when political elites have insufficient incentive to present alternative policies for voters to choose from or when politics ceases to be competitive. Schattschneider (1960)

Hines 4 admonishes that under these circumstances “the people are powerless” (137). Politics can also be fragmented to the point that no clear winner emerges. In a famous case, Sartori (1976) describes the “polarized pluralism” of Italy. The political system was so fragmented that it resulted in permanent capture of the government by the dominant centrist party, Partito della Democrazia Cristiana (DC), and increased support for extremist parties. Anything that constrains political competition can be dangerous to a democratic polity. One argument for the existence of a “democratic deficit” in the EU is that economic integration has proceeded ahead of appropriate political integration (Paulson 1997). Since important economic decisions are made without sufficient input from national governments, national political parties are constrained in the policy alternatives they can offer voters. Hix (2003) summarizes the argument by saying: The EU policy regime has a powerful indirect impact on the process of domestic political competition. The single market rules governing the production, distribution and exchange of goods, services, capital and labour restrict microeconomic policy options. Similarly, Economic and Monetary Union (EMU), the Growth and Stability Pact and the multilateral macro-economic surveillance restrict national budgetary and fiscal policy options. As a result, parties on the left (who accept the EU regime as legitimate) cannot promise high levels of social protection or expansionary budgetary policies. Equally, parties on the right cannot promise further deregulation of labour, product or capital markets or major cuts in public expenditure (9-10).

Hines 5 Hix (2003) looks for evidence of this constraint in the electoral manifestos of the major left and right parties in the EU member states. He demonstrates that the ideological distance between the major parties of the left and right in the member states of the EU has indeed grown closer overtime. While the link between integration and this policy convergence is weak, Hix infers at least in the short term, policy convergence has been accelerated by the common market. However, Hix does not look at another possible consequence of limited competition, party fragmentation. If the EU constrains the menu of policy choices available to the leading parties, it has the potential to increase party fragmentation by altering voting behavior. Given the choice between large parties that differ little on their policy positions and smaller parties that are unencumbered by the policy realties of integration, voters may be encouraged to vote for more extreme parties. An examination of the two major spatial models of voting, proximate voting and directional voting, helps to illustrate this point and generate a testable hypothesis about the effects of limited competition on voting behaviour. The classic spatial model of voting is the Downsian proximity model (Downs 1957). As in all spatial models, voters and candidates (or parties) are represented by points in a one-dimensional (or multi-dimensional) space that reflects their respective opinions on the issues. In a single issue, or one-dimensional space, the Downsian model is a formalization of the standard left-right scale. The theory behind this model of voting is that voters will assess the distance between themselves and the candidates and vote accordingly for the candidate who is closest to the voter’s ideal point. This suggests that a voter’s evaluation is based upon utility and that voters maximize their utility by choosing candidates closest to their ideal.

Hines 6 However, Downs noted in his original outline of the model noted that voters would know that no party would be able to do everything they claim they can. Therefore, voters will anticipate what candidates (or parties) will do once in office and vote accordingly. Grofman (1985) offers a modification of the Downsian model that takes into account the “discounting” of candidate claims by voters. This model uses the position of the status quo and an explicit discount of the candidate’s positions to calculate new utilities of vote choice. Voters will vote for candidates whose likely actions will bring the status quo closer to the voter’s ideal point rather then simply taking the candidates promises at face value. If the EU does constrain the policy actions of political parties, this effect will be disproportionate across parties. The larger the party, and therefore the more likely the party is to be the party of the Prime Minister after an election, the more constraint that the party will face. A minor coalition member can blame inaction on its coalition partners. Large parties generally cannot. This will, in the language of Grofman’s model, cause voters to discount the policy statements of the parties differently. Since the EU constrains policy outputs to the center, parties with centrist leanings will be seen as supporting the status quo no matter how much they promise otherwise. Voters are therefore more likely to vote for parties farther towards the edges of the policy space because their discounted policy position is closer to their ideal point than that of the major parties. An alternative spatial model of voting behavior is based on directionality rather than proximity. Reynolds (1974) drew on prior work in psychology to develop a theory of voting based on the aggregate of a voter’s belief about a candidate’s issue positions

Hines 7 and the voter’s evaluation of those issues. For each issue, Reynolds assigned a value ranging from -3 to 3 or very liberal to very conservative on each belief and evaluation. Reynolds then predicted voters’ attitudes towards candidates by summing the products of a voter’s beliefs about a candidate’s position and the voter’s own evaluation of the issue. In this model, choosing a candidate who is perceived to be on the “right side” of the issues maximizes a voter’s utility. Rabinowitz (1978) and Rabinowitz and MacDonald (1989) develop a spatial model of directional voting that uses the same utility function as Reynolds but replaces beliefs and evaluations with intensities. Vote choice is based upon a voter utility determined by the intensity and communality of direction of a position of the voters and the candidates’ positions. This theory assumes that voters have diffuse preferences for certain directions on issues whose importance, or intensity, varies. Candidates are also assumed to vary in intensity with which they campaign on a particular issue position. Voters will vote for candidates whose overall intensity on the issues is closest in direction to themselves. Candidates can be more extreme then the voter, but as long as they are on the “right side” of the issues the voter will vote for them. The model generates several specific hypotheses about voter and party behavior in races with three or more candidates (or parties). The most important is that there is no incentive to move to the center, as is the case with a Downsian model. Since voters will only vote for parties on the their side of the ideological spectrum, the movement of parties to the center does not attract centrist voters from the opposing side. In fact, they risk losing votes to more extreme parties on their ideological flank. If the EU constrains policy competition it will clearly reduce the intensity of the candidates on those policy

Hines 8 issues. If voters vote directionally, minor parties will benefit from the declining intensity of the major parties. Even voters who favor centrist outcomes, but who have a passion about a certain issue, will vote for outlying parties because of the benefit of voting the right direction on an issue. Both spatial models of voting behavior suggest that if the political competition of major parties is constrained to the ideological center, more extreme parties will benefit from greater support. If the EU does in fact cause these constraints, we should see an increase in party fragmentation as each additional constraint of European integration, like completion of the common market and the EMU, is placed on political competition in the member states. This leads to the central hypothesis of this paper: As European integration places constraints on the menu of policies available for political parties to compete on, policy convergence amongst the large governmental parties in the member states will lead to an increasing fragmentation of the party system.1

Data and Methods The dependent variable in this study is the effective number of electoral parties (ENEP) in all-national elections held in the EU15 from 1960-1998. If European integration has pushed voters to support new parties, controlling for other factors that contribute to volatility in the number of political parties, the ENEP would increase. If voters have simply shifted support between existing parties, the ENEP would remain steady. Further, if integration had reduced the number of political parties by concentrating the vote, the ENEP would decline. 1

I realize that the causal argument behind this hypothesis needs to be unpacked. The policy convergence that leads to greater party fragmentation could be caused by many other factors then the EU. At best, integration can contribute to party fragmentation.

Hines 9 This particular measure of the number of political parties deals with several practical issues faced while studying party systems. Specifically, when counting parties a researcher will be faced with counting parties of different sizes and particularly challenged on the issue of small parties. A general consensus exists for using a weighted measure of political parties (Lijphart 1994). Rae (1971) developed the first generally accepted weighted measure of party fractionalization. However, this measure is not as intuitive as the subsequent measure, ENEP, developed by Laakso and Taagepera (1979). Their measure can be calculated for both vote share and seat share. The formula for calculating the ENEP in an election, where vi is the vote share for each party in the election, is simply:

ENEP =

1 " vi 2

(1)

! While the effective number of parties can be based on either vote share or seat share, the effect of European integration on the number of parties is more likely to be captured by the measure at the electoral level. The ENEP and the effective number of parliamentary parties (ENPP) are affected differently by Durverger’s (1954) “mechanical” and “psychological” effects. The mechanical effects are simply the translation of votes into seats after an election. Therefore, ENPP is a function of both the mechanical functioning of electoral rules and the broader psychological effects underlying vote choice. On the other hand, ENEP is a function of only the psychological effects or the expectations of how votes will be translated into seats. Since European integration only changes the psychological calculation of voters and not the mechanical rules for calculating seats, ENEP is a more appropriate measure of political competition

Hines 10 than ENPP for this study. The data used comes from a dataset on democratic elections compiled by Golder (2004). To measure the potential effects European integration has on electoral fragmentation, I use a series of six measures. Throughout the period covered in this study, three major changes occurred in integration in each of the member states. First, the number of members expanded from six to fifteen. The EU should have no effect on states before they join the EU. To capture any change in the ENEP caused by membership, I include a dummy coded 1 for each election-year in which a country was a member of the EU (or its predecessors). Further, because membership becomes more constraining over time, I include an interaction of membership and time. A second change to European integration that potentially constrained political competition is the push towards completion of the common market that began in 1986. To capture any change in the ENEP caused by this dramatic increase in the power of the EU, I include a dummy coded 1 for each election-year after 1986. Because the common market was also phased in over time, I include an interaction of the common market indicator and time. The third change in the depth of European integration, and the most constraining for political competition, is the introduction of the EMU and the convergence criteria in 1992 as Europe prepared for the adoption of a common currency. To capture any change in the ENEP caused the growth and stability pact, I include a dummy coded 1 for each election-year after and including 1992 in which a member state was a member of the EMU.

Hines 11 Alternative covariates of the ENEP are also controlled for in the statistical model. The most important mechanical determinant of ENEP is district magnitude because it determines the permissiveness of an electoral system (Rae 1971, Taagepera & Shugart 1989, Cox 1997). The greater the average magnitude of an electoral system’s districts, the more permissive the system become for small parties. I log the measure of average district magnitude because the marginal effect of adding an additional seat is smaller as the magnitude increases. The increasing fragmentation of European party system may simply be a function of time. This is in part due to the argument that economic prosperity leads to the creation of new demands on government and political parties advocating those demands. There is also a social learning component in which parties will require time to learn the ways to maximize their electoral performance.2 To control for these possibilities, I include a trend variable that takes on a value equal to the year of the observation minus the start year, 1960. The variable ranges from a value of 0 to 38 for elections in the final year of the study. This variable is also used in the interactive variables discussed above. Finally, I control for a variety of economic factors that may increase (or decrease) electoral volatility during an election. One should expect political competition to more fierce during times of economic hardship. However, political competition should be constrained by greater exposure to international markets. Therefore, I include a measure of the percent of the member state’s GNP that is from the exportation of good to capture the effects of global trade. Greater exposure should increase the number of political 2

For example, the French Greens eventually gained sufficient support to enter government (and parliament) by forming electoral coalitions with the Socialists in districts where their combined support could win against a divided right. However, it took nearly two decades for the French Greens to do this.

Hines 12 parties. To capture the impact of economic hardship, I construct a “misery index” that is the logged annual inflation rate times the unemployment rate. I also include the two components of the index separately. This data comes from the World Bank’s World Development Indicators Database.3 Data is missing for unemployment for election-years prior to 1980. Therefore I estimate a model that includes data from 1980-1998 and 19601998. This gives the final model the following specification: ENEPt = " 0 + "1 Membership + " 2 (Membership * Trend) + " 3CommonMarket +" 4 (CommonMarket * Trend) + " 5 EMU + " 6Trend + " 7 ln(Magnitude) + " 8 ln(Trade) + " 9 Misery + "10 ln(Inflation) + "11Unemployment

The data and the model present some specific methodological issues. The model

!

uses pooled cross-sectional times series data. Cross-sectional times series data has two types of information, the cross-sectional data between subjects and the times-series information within subjects that changes over time. Standard OLS regression can be applied to cross-sectional times-series data, but it is not the most efficient method. The reason is that the estimates derived from standard regression may suffer from omitted variable bias. The correction for this problem changes depending on whether the bias is in the cross-section or the time-series portion of the data. The most common type of omitted variable bias occurs when the omitted variables differ between subjects but are constant over time. The correction for this is to use fixed effects regression. However, panel data can also suffer from omitted variable bias when an omitted variable varies over time but is constant between subjects. The solution for this is between effects regression. When both types of bias are present in the 3

Available on-line at http://devdata.worldbank.org/dataonline/.

Hines 13 data, a common solution is to use random effects regression that weights fixed and between effects. Since almost any panel data of the sort analyzed in this paper will suffer omitted variable bias due to fixed effects, that correction is necessary. However, if between effects bias exists as well, it is necessary to use a random effects model. A fixed effects model provides more consistent estimates. This provides a way to test which of the two models, fixed effects or random effects, is appropriate. This can be accomplished with a Hausman test (Stock and Watson 2003, Chapter 8). The Hausman test tests the null hypothesis that the estimates from a random effects regression model are the same as the consistent estimates obtained from the fixed effect regression. If they are, it is all right to proceed with the random effects regression. After running the random effects and the fixed effects regression, I obtained a Hausman test statistic that indicates that random effects are acceptable.4 Therefore, I use a random effects generalized least squares regression in the following analysis.

Results Table 1 shows the results of two random effects GLS regressions. The first model includes all eleven variables and therefore covers eighty-one elections between 19801998. The second model excludes the misery index and unemployment to cover 153 elections from 1960-1998. The performance of both models is mixed. The first model performs best at explaining the variation within countries, but barely accounts for any variation between the countries. The shorter time-span of the 4

A Hausman test statistic is distributed Chi-square with the degrees of freedom equal to the number of estimated coefficients. In this study, I estimate eleven independent variables and the Hausman returns a test statistic of 1.55. This fails to achieve statistical significance.

Hines 14 first model means that there is less within country variation being caused by omitted variables. This produces the better fit for the fixed effects or cross-sectional data. The low model fit for the between effects, or the time-series, does not have an easy interpretation. All that can be concluded is that it does well with the explaining the variation inside countries, but not between them. Of the five independent variables, two are significant. It appears that membership in the EU and the time trend associated with membership is not significant. Nor is the time trend associated with the membership in the common market. However, entry into the common market in 1986 and entry into the EMU in 1992 both have positive and significant effects on the ENEP. Marginally, joining the common market added .8 more political parties to the party systems of the member states. The EMU had a slightly bigger effect, adding an additional .84 political parties to the party systems of the member states. These findings are consistent with the expectations of my theory that integration has lead to some increased fragmentation of party systems. Of the six control variables, only two are significant. The logged average magnitude of an electoral system actually decreases the number of ENEP. At first, this seemed to be an anomalous result. Decades of research on electoral systems suggest the exact opposite is true. However, this appears to be a function of the fixed effects portion of the regression model.5 The average magnitude seems to be negatively correlated with the omitted variable bias corrected for by fixed effects. In short, the bias becomes greater as the number of districts increases and therefore the model must reduce the marginal effect. A second way of looking at this is that average magnitude is such a powerful

5

I examined the data using a between-effects regression and a standard OLS without country dummies. In these cases, as theoretically expected, the coefficient for magnitude was positive and highly significant.

Hines 15 predictor of the average ENEP in a country that it establishes the intercept, or starting level, for each country. The remainder of the information included in the term simply controls for differing effects between countries. Another significant control variable was the misery index. This suggests, not surprisingly, that as the overall economic hardship increases voters begin to vote for smaller parties in protest of the policies of the government. However, unemployment and inflation taken alone do not have an independent effect on the ENEP. Global trade is the only the significant variable in the model. This suggests countries become more electorally permissive as their interdependence on other countries increases. The second model is much poorer at explaining variation within countries, but does a comparatively better job explaining variation between countries. The longer timespan of the data in the second model increases the likelihood of misspecification error and reduces the explanatory power of the model for fixed effects. I do not know why the between country fit becomes stronger. Both measures of fit indicate that I have an incomplete model. This is also indicated by the substantive findings of the second model, where only average magnitude is significant. This does not add any support for the theory that integration has increased party fragmentation, but it does not really detract from it either. The model is underspecified and the effects of the EU may simply be too difficult to separate from the background without refining the model and hopefully eliminating missing data.

Conclusion Overall, it is difficult to make any conclusive inferences from the results. The first model’s findings are suggest are consistent with my theoretical expectations, but the

Hines 16 relationship disappears in the longer time span of the second model. Without better data and some deep reflection on the likely sources of specification bias in the models, it is impossible to conclusively say that integration has affected political competition undertaken in the member states by increasing party fragmentation. There are some important normative reasons for considering the claims of the critics of the “democratic deficit” in the European Union (EU) when they suggest that European integration alters political competition in the member states. This paper has explored one implication of that alteration, party system fragmentation. Increased fragmentation of the party system could have consequences for government formation and stability, for parliamentary effectiveness and responsiveness, and most importantly, on the levels of support for extreme parties. This is especially important if it leads to support for the most extreme parties, especially of the right wing. Looking for these consequences is one possible “next step” for this project. Unfortunately, there are two important steps that need to be taken before I get to that point. First, there needs to be a more substantial test of the relationship between European integration and the policy convergence of the major political parties in the member states. This paper’s analysis rests upon the rather weak assumption that integration has something to do with this trend. This paper does not test this assumption and the literature is divided on what impact integration has had. Second, this paper rests upon an assumption that integration has altered voting behavior. My theory requires that voters be more willing to support a greater number of political parties. While the evidence is not inconsistent with the assumption, my analysis suffers from a general problem of formal modeling — I am looking at macro level

Hines 17 evidence of micro level behavior. It is equally plausible that supporters of the major parties have stopped voting so the supporters of smaller parties have more weight. Further, there is no way to say conclusively that the common market, the European Monetary Union (EMU) and its convergence criteria is in anyway responsible for changing voter behavior. The liberal conception of democracy requires political competition between rival candidates and rival programs to be effective, but it assumes that there are no exogenous constraints on political competition. This assumption may not hold in the member states of the European Union. This paper has looked at one implication of the potential constraint of political competition in the member states, party fragmentation. While the evidence is consistent with the theory, it is far from conclusive. Breaking the theory into its component parts and testing those parts is the likely next step in providing a more satisfactory conclusion.

Hines 18

References Besley, Timonthy J, and Stephen Coate. 1997. An Economic Model of Representative Democracy. Quarterly Journal of Economics 112 (1):85-114. Clark, Willam, Matthew Golder, and Sona Nadenichek. 2002. Fiscal Policy and Democratic Process in the European Union. European Union Politics 2 (2):205230. Cox, Gary W. 1997. Making votes count : strategic coordination in the world's electoral systems, Political economy of institutions and decisions. Cambridge, U.K. ; New York: Cambridge University Press. Downs, Anthony. 1957. An Economic Theory of Voting. New York: Harper and Row. Duverger, Maurice. 1954. Political parties, their organization and activity in the modern state. New York: Wiley. Golder, Matthew. 2004. Democratic Electoral Systems Around the World, 1946-2000. Electoral Studies forthcoming. Grofman, Bernard. 1985. The Neglected Role of the Status Quo in Models of Issue Voting. Journal of Politics 47:230-47. Hix, Simon. 2003. The End of Democracy: How the European Union (As Currently Designed) Restricts Political Competition. London School of Economics and Political Science. Laakso, Markku, and Rein Taagepera. 1979. Effective Number of Parties: A meausre with Application to West Europe. Comparative Political Studies 12:2-27. Lijphart, Arend, and Don Aitkin. 1994. Electoral systems and party systems : a study of twenty-seven democracies, 1945-1990. New York: Oxford University Press. Paulson, Andrew. 1997. Cumulative Regional Decline, Institutional Inadequacy and the 'Democratic Deficit': Is European Monetary Union Economically and Politically Sustainable? In Economic Working Paper Archive at WUSTL. Pitkin, Hanna Fenichel. 1967. The concept of representation. Berkeley,: University of California Press. Rabinowitz, George. 1978. On the Naure of Political Issues: Insights from a Spatial Analysis. American Journal of Political Science 22:793-817. Rabinowitz, George, and Stuart Elaine Macdonald. 1989. A Directional Theory of

Hines 19 Voting. American Political Science Review 83:93-121. Rae, Douglas W. 1971. The political consequences of electoral laws. Rev. ed. New Haven: Yale University Press. Reynolds, HT. 1974. HT Reynolds. Journal of Politics 36 (4):983-1005. Sartori, Giovanni. 1976. Parties and party systems: a framework for analysis. Cambridge [Eng.] ; New York: Cambridge University Press. Schattschneider, E. E. 1960. The Semisovereign People : a realist's view of democracy in America. New York: Holt, Rinehart and Winston. Schumpeter, Joseph Alois. 1942. Capitalism, socialism, and democracy. New York, London,: Harper & Brothers. Stock, James H., and Mark W. Watson. 2003. Introduction to econometrics. Boston, MA: Addison Wesley. Taagepera, Rein, and Matthew Soberg Shugart. 1989. Seats and votes: the effects and determinants of electoral systems. New Haven: Yale University Press.

Hines 20 Table 1: Fixed Effects Regression for the Effective Number of Electoral Parties in the EU15, 1980-1998 and1960-1998 (1) 1980-1998 (2) 1960-1998 Membership -0.076 -0.375 (1.033) (0.402) Membership * Trend -0.009 0.009 (0.035) (0.018) Common Market 0.803** -0.208 (0.406) (0.513) Common Market * Trend 0.003 0.004 (0.015) (0.018) EMU 0.841*** 0.257 (0.274) (0.307) Average District Magnitude (ln) -0.369*** -0.354** Trend Global Trade (ln) Misery Index Unemployment Inflation (ln) Constant

(0.143) -0.037 (0.045) 1.209** (0.555) 0.006** (0.003) -0.029 (0.032) -0.036 (0.029) 1.308 (2.158)

(0.173) 0.024 (0.019) 0.392 (0.531) — — 0.014 (0.038) 3.107* (1.673)

Observations 81 153 Number of countries 15 15 R-squared within .45 .26 R-squared between 0.01 0.10 Note: Dependent variable is the effective number of electoral parties. Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%