Another Endogeneity Problem and More in Electoral Studies: Social Cleavages Revisited

Another Endogeneity Problem and More in Electoral Studies: Social Cleavages Revisited

Prepared for the SCR Graduate Retreat, May 2004

Heather Stoll

Stanford University

What is the relationship between political institutions, social cleavages, and party systems? Party systems influence many other aspects of politics in both established and newly emerging democracies; accordingly, driven by the demands of theory and practice alike, political scientists in the comparative politics sub-field have sought to understand their determinants. At one extreme, for example, scholars have linked a polity’s party system to its type of democracy: the theory that two party systems give rise to a majoritarian democratic process while multi-party ones give rise to a consensual democratic process (Lijphart 1999) in turn has produced institutional prescriptions for achieving certain ends. An impressive amount of research energy has been devoted to both theory building and empirical analysis in the decades following the publication of Duverger’s classic text (1963) that propelled the line of inquiry into prominence. The dependent variable of party system has been characterized both qualitatively and quantitatively, although the most prominent characterization is quantitative. Various independent variables such as social cleavages have been identified and their relationship to the dependent variable investigated.

Sustained scholarly effort has focused primarily upon two political institutional independent variables: electoral systems and regime type. The first systematic theoretical account linking electoral competition at the district and national levels was offered by Cox 1997 in one of the most influential comparative politics texts of the last decade. At the district level, he refined the Duvergerian argument and suggested that different electoral systems promote greater or lesser degrees of strategic coordination; in equilibrium, his ‘M + 1’ rule predicts that the number of candidates or party labels will not exceed an upper bound equal to the district magnitude plus one. The micro-foundation of this district (aggregate) level relationship between electoral and party systems is the strategic behavior of voters (strategic voting) and elites (strategic entry). Further, at the national level, Cox provided the missing link between Duvergerian district-level effects and national party system aggregation: the selection of the national executive, which confronts actors with another coordination problem, encourages the formation of cross-district strategic alliances to capture political power. Social cleavages, or preferences in the parlance of this thesis, enter as a counterpoint to the institutional variables: they explain the number and types of candidates or party labels that naturally compete in elections (i.e., prior to the incentives for coordination that are provided by the electoral system and the need to select an executive). Existing theories operating at the level of abstract concepts accordingly explain cross-national and cross-temporal variance in party systems with the three variables of preferences, regime type, and electoral system, where the latter is by far the best understood.

Despite the importance of the social cleavages variable in the party and electoral systems literature, a brief survey of recent quantitative empirical research (post-1980) that attempts to account for cross-national variation in party systems reveals a seeming lack of consensus about how the variable should be operationalized and its relationship to other variables modeled. In fact, this variable is the most under-theorized and poorly operationalized in the literature. Lijphart (1981; 1984; 1999) tests for a correlation between the number of issue dimensions and the effective number of legislative parties. Powell (1982) utilizes three measures of social heterogeneity in regression models for legislative fractionalization: ethnic fractionalization; an ordinal measure of the agricultural proportion of the population; and an ordinal measure of the Catholic proportion of the population. Taagepera and Grofman (1985) adopt Lijphart’s perspective and utilize his data to develop and test their hypothesis that the number of parties equals the number of issues plus one. Taagepera and Shugart (1989) and Taagepera (1999) also adopt Lijphart’s perspective and operationalization. Ordeshook and Shvetsova (1994) model electoral systems as intervening structures through which the basic characteristics of society work to influence the number of political parties. Their independent variable is most broadly defined as social structure (the homogeneity or heterogeneity of society), which they operationalize as the effective number of ethnic groups. The left-hand side of their models varies from the effective number of electoral parties and the effective number of legislative parties to the number of parties obtaining more than 1% of the vote in two successive elections. Cox (1997) returns to a more traditional definition of the variable, evidenced by his usage of the term ‘social cleavages’ and his equating of social structure with the number and type of cleavages in society; however, empirically he (1997; Amorim Neto and Cox 1997) builds directly upon Ordeshook and Shvetsova’s work, utilizing the same measure of effective number of ethnic groups in his cross-sectional regression models for the effective number of electoral parties. Demonstrating that scholarly work, like institutional development, is path dependent, almost all subsequent studies have ‘hoed the same row’[1]: see, for example, Jones 1997; Filippov, Ordeshook, and Shvetsova 1999; Benoit 2002; Golder and Clark 2003; Chhibber and Kollman 2004. One exception is Jones (N.d.), who employs two operationalizations of the social cleavage variable: the first is the conventional effective number of ethnic groups measure; the second is a set of novel measures of ideological fractionalization, which use public opinion data to generate a measure of the left-right ideological heterogeneity of the populace. Another is Scarritt, Mozaffar and Galaich (2003), who add an additional variable—ethnic group territorial concentration—to the traditional fragmentation measure.

Other quantitative studies recognize the importance of this variable but fail to incorporate it into their models. One explanation for this omission is the daunting theoretical and empirical problems that need to be overcome in order for incorporation to occur. Recent studies in this vein include Lijphart 1990, Nagel 1994, Coppedge 1997, Chhibber and Kollman 1998, and Jones 1999. Qualitative studies such as Mainwaring and Scully 1995 are often similarly characterized by a poorly (and, across studies, variably) defined concept of preferences but their very nature makes the disagreement less noticeable: while both qualitative and quantitative work can be scientific, quantification clearly signals the definition of an abstract concept employed by a researcher, which we too often have to dig to unearth in a qualitative study.

Recent empirical attempts such as these to integrate the institutionalist and sociological perspectives on party systems are to be applauded. However, scholars must ask why we observe such diverse modeling choices in the quantitative empirical party and electoral systems literature. The answer is two-fold: definitional disagreement, often implicit, about the variable of preferences and underdevelopment of theory relating this variable to electoral and party systems. We must also ask about its consequences, along with the appropriateness of many of the modeling choices that have been made: what do we in fact know about the relationship between preferences and party systems?

On the first source of empirical diversity, the previous chapter discussed the different positions that scholars have taken on three definitional issues: the nature of divisions, i.e. whether or not to confine the analysis to conflicts rooted in sociological traits; the persistence of divisions, i.e. whether or not to confine the analysis to long-standing (stable across many elections) conflicts; and the stage in issue evolution, i.e. to which type of preferences—latent, political, or particized[2]—to confine the analysis. Different definitions of the abstract concept are frequently employed. For example, the preferences that Ordeshook and Shvetsova 1994 discuss are latent and sociological; the preferences that appear in Lijphart 1999, conversely, are particized and non-sociological. The other source of empirical diversity is the hypotheses that scholars test, which naturally differ with the different definitions of abstract concepts employed. The most fundamental definitional issue discussed in the previous chapter, the stage in issue evolution, is closely linked to how scholars think preferences should relate to party systems. Some hypothesize that greater latent preference diversity, i.e. more latent cleavages and/or groups generated by latent cleavages, is associated with less electoral coordination and a greater subsequent number of candidates or party labels competing in elections; others put particized diversity on the right-hand side. Further, other differences also abound such as the nature of the relationship between the two: some argue that it is additive and others interactive. As far as consequences are concerned, existing empirical results are called into question by the methodological problems identified by this chapter. It is accordingly difficult to conclude that there is much empirical support for the compelling hypothesis that both social cleavages and political institutions such as electoral systems influence electoral coordination.

This chapter reviews and evaluates the existing theoretical and empirical electoral and party systems literature. It attempts to improve upon existing work by identifying the underlying sources of the diversity of modeling choices; formalizing the bias that results from both measurement error and the modeling of social cleavages as exogenous to electoral systems; assessing the robustness of existing results; and identifying potential theoretical and empirical targets for improvement. It initially delves deeper into the theory that relates preferences to party systems. The different abstract and testable hypotheses that scholars have developed contribute to the empirical diversity discussed above; the usefulness of these hypotheses for advancing the research goals of the literature is assessed. The chapter then turns to the empirical literature: it summarizes existing results and conclusions drawn; provides a methodological critique of the approaches employed; and assess the robustness of results to different operationalizations of the preference diversity variable. It concludes

What Lies Beneath: Theory and Testable Hypotheses

A party system, the abstract concept of interest to be explained, is the parties in a polity and, most importantly, the patterns of interaction between them (Sartori 1976; Ware 1996; Broughton and Donovan 1998). A reasonable consensus has emerged in the comparative politics literature that the best way to define a party system is by the number of parties (Lijphart 1994). Such a definition captures the crucial distinction between two and multi-party systems as well as the general degree of party system fragmentation. It does, however, leave out information about some aspects of competition, e.g. whether competition is centripetal or centrifugal (see Sartori 1976 and Lijphart 1994) and the nature of the ideological conflicts about which political battles are waged. The quantitative operationalization most commonly used is Laakso and Taagepera’s effective number of parties (1979), which can be calculated with respect to either vote or seat shares to measure either the elective or legislative party systems, respectively.

From Lipset and Rokkan to Cox, scholars in the electoral and party systems literature have theorized at the most general level that the more ‘social cleavages’ there are in a polity, the more political parties that polity will have. In other words, polities with similar degrees of preference diversity should have similar numbers of political parties, after—according to some—controlling for institutions such as the electoral system. The generation of political parties, in this view, is essentially demand-driven: the groups that comprise the cleavage structure of a polity will each demand that a party represent them and parties will accordingly be supplied. The resulting number of parties is the ‘natural number’ in a polity.

However, the theoretical development of the relationship between preference diversity and party systems has not proceeded apace with that between institutions and party systems. As Jones (1999, 174) argues[3], “unfortunately, there exists little theoretical guidance as to how and why these factors influence the number of… candidates… [A] focus on institutional variables for which the hypotheses are supported by strong theory is very important…” The process by which groups demand and subsequently receive organizational representation by political parties is most developed by Cox 1997. Instead of the simple additive relationship sketched above, he argues for an interactive one. For example, Cox writes, “the correct understanding of the institutionalist model implies that the number of parties in a system ought to be an interactive function of electoral and social structure” (1997, 9). He elaborates on this thesis at the district level later in the text:

“If we adopt the simple notion that the more cleavages there are in a society, the more parties it will have, but modify it by appeal to the institutionally imposed upper bound articulated by the M + 1 rule, we should expect that the number of competitors, N, will be an interactive function of electoral and social structure: N will be low if either the electoral system is strong or social diversity is low; N will be high only if the electoral system is permissive and social diversity is high” (Ibid, 141-42).

Further, he goes on to argue, “just as at the district level, [so too at the national level] one might argue that a large number of separate parties will arise only if there is both social diversity and electoral proportionality” (Ibid, 193).

Nevertheless, even this account leaves many questions unanswered. What type of preferences generates the natural number of parties that electoral systems constrain given sufficient restrictiveness? For Taagepera and Grofman (1985), it is particized preferences. For Cox, it is latent preferences, although this choice seems driven more by practicality—methodological concerns—than by theoretical conviction. Further, should focus be on the groups generated by cleavages or upon the cleavages themselves? Again, scholars such as Taagepera and Grofman and Cox part ways on this issue. What is the model of elite and mass behavior underlying group demands for representation? How can we think about the process spatially and what, if anything, does the vast spatial literature have to contribute to the theory? Does strategic entry occur in two stages, creating the issue space (the particized cleavages) and later in reaction to a given issue space? Where in the process do other variables, such as Cox’s presidential party system (1997) and Chhibber and Kollman’s (2004) government centralization, play a role? These theoretical questions need to be resolved before the predicted relationships can be satisfactorily subjected to empirical analysis and conclusions drawn about causality.

Despite these theoretical ambiguities, testable hypotheses have been developed. Three one-sided, additive hypotheses are suggested in light of the most fundamental outstanding theoretical issue discussed briefly above and at length in the prior chapter—how preferences should be defined: