Why the salience of social divisions matters in party systems: Testing the interactive hypothesis in South Africa

Abstract

Scholars have long argued social diversity, and electoral institutions interactively shape party systems: diversity has little effect on the effective number of parties (ENP) in single member plurality (SMP) systems but increases ENP in proportional ones. We argue instead that where diversity is salient enough to generate demand for parties, it also hinders strategic coordination, preventing SMP rules from reducing the number of parties and producing a correlation between diversity and ENP. In contrast, non-salient forms of diversity have little impact regardless of institutional rules. We test this intuition using data from South Africa’s municipal mixed-member system and explore its highly salient racial cleavage and less salient ethnic one. We find racial diversity correlates with ENP in SMP systems while ethnic diversity correlates with ENP in neither SMP nor proportional representation systems. Our study contributes to mounting evidence questioning the interactive hypothesis and points to the importance of the salience of social divisions in shaping party systems.

Keywords

ethnic parties mixed electoral system party systems social cleavages South Africa

Political scientists since Duverger have argued that social diversity and electoral institutions interactively shape party systems. In this view, social diversity has little effect on the district level number of parties in single member plurality (SMP) systems but potentially large effects in proportional ones (Amorim-Neto and Cox, 1997; Clark and Golder, 2006; Duverger, 1952, 1963; Ordeshook and Shvetsova, 1994). The logic appears straightforward: SMP rules induce coordination around two main parties, blocking the translation of all but one social cleavage into the party system. In contrast, by allowing multiple parties to flourish, proportional electoral rules permit party systems to reflect the underlying diversity of society.

The interactive hypothesis builds on the intuition that social diversity generates demand for parties. However, where social diversity is salient enough to produce demand for parties, it may also hinder the strategic coordination underlying Duverger’s law, preventing SMP rules from reducing the number of parties at the district level. Specifically, social diversity may interfere with the conditions that Cox (1997) identifies as critical to strategic coordination by inhibiting short-term instrumentally rational behavior, preventing the emergence of a common set of beliefs about the ranking of parties, and generating significant barriers to elite coalition building. Thus, where social diversity is salient, it should correlate with the size of the district party system even in restrictive SMP systems. Where it is not salient, it is unlikely to shape the district party system in the first place.

Early confirmatory tests of the interactive hypothesis, including Amorim-Neto and Cox (1997), Clark and Golder (2006), Mozaffar et al. (2003), and Ordeshook and Shvetsova (1994), employ cross-national data. Such studies confront several empirical challenges, including causal identification in cross-national designs and the consistent measurement of social diversity across countries with different types of cleavages (Jones, 1997; Potter, 2014; Stoll, 2008, 2013). Furthermore, because the interactive hypothesis pertains to district level party systems, most national tests use an inappropriate level of analysis (Moser and Scheiner, 2012).¹ More recently, scholars have shifted their focus to the district level, with mixed findings (Geys, 2006; Jones, 1997, 2004; Moser and Scheiner, 2012; Moser et al., 2011; Penas, 2004; Potter, 2014; Vatter, 2003).

Our study makes four contributions to this literature. First, we develop a theoretical rationale for why salient social diversity might impede electoral coordination and interfere with the causal logic underlying the interactive hypothesis. Second, we introduce an African case to a literature that has primarily focused on Europe. Third, unlike most previous studies (see Potter, 2014, for an exception), we do not assume that countries have only one type of social diversity. In the case we examine, South Africa, we identify two relevant social cleavages, race and ethnicity, and examine their effects on the party system. Fourth, we employ a research design based on a large number of municipal-level mixed-member electoral systems nested within a single country. Previous work by Moser and Scheiner (2012) pioneered the use of national-level mixed-member systems to study the interactive hypothesis, arguing that these systems offer the possibility of a controlled comparison between SMP and proportional representation (PR) rules for the same unit of analysis—in their case, countries. We also exploit a mixed-member design but use South African municipalities as our unit of analysis. These municipalities have municipal councils with at-large seats elected using PR rules and ward seats elected using SMP. Elections for these councils offer key advantages to the study of electoral systems and social diversity. They allow us to hold constant national-level variables like the salience and nature of social divisions, the degree to which democracy is consolidated, and socioeconomic development. They also provide a larger number of cases than national mixed-member designs. Whereas Moser and Scheiner (2012) compare five national-level mixed-member systems, we explore over 200 municipal-level mixed-member systems, including approximately 4000 wards. Given the size of this sample, we can control for important confounds and more rigorously test the PR component of the interactive hypothesis. To the best of our knowledge, this is the first study to evaluate the interactive hypothesis using a municipal-level mixed-member system.

We find a significant and meaningful correlation between the racial diversity and the effective number of parties (ENP) in South Africa’s SMP-based wards that is robust to a demanding set of controls. Our study therefore contributes to mounting evidence that, contrary to the interactive hypothesis, social diversity matters in SMDs in a variety of different contexts and systems (Moser and Scheiner 2012; Moser et al. In press; Potter, 2014; Stoll, 2013). We also find a significant correlation between the racial diversity and the ENP in municipal-wide PR elections as well as evidence suggesting district magnitude and racial diversity interactively shape ENP in such races. These results, which are consistent with some prior studies, suggest that diversity may have stronger effects in permissive PR systems versus restrictive ones, even if they are not wholly absent in the latter. They are not robust to controlling for racial group size, provincial effects, and the urban/rural composition of a municipality, however, suggesting that previous work finding support for the interactive hypothesis in PR systems may conflate the effects of diversity with other factors like the size of particular groups in the population. Finally, we observe little correlation between ethnic diversity and ENP in most models, at either the ward or municipal level. Our analysis thus casts significant doubt on one component of the interactive hypothesis—the absence of diversity effects in SMP systems. It also reveals that a factor ignored in most previous studies—the salience of a social division—conditions the impact of that division on party systems.

The interactive hypothesis: Causal logic

While often categorized as an institutionalist, Duverger also highlights the importance of social forces in shaping party systems, arguing that these create demands for political parties that political institutions subsequently mediate (Clark and Golder, 2006; Duverger, 1952, 1963). Later studies refine this idea as the interactive hypothesis, which has two components. First, social diversity should not strongly correlate with the size of the party system in restrictive electoral systems like SMP; as Ordeshook and Shvetsova (1994: 122) state, “if district magnitude equals one, then the party system is relatively ‘impervious’ to ethnic and linguistic heterogeneity.” Second, social diversity should correlate with the size of the party system in permissive electoral systems, and the more the permissive the system, the more social diversity should matter. Large party systems thus emerge when social diversity creates a high demand for parties and institutions permit it; smaller party systems occur either when low social diversity generates low demand for parties or where institutions are constraining. These expectations should hold primarily at the district level.²

While intuitive on its face, the causal logic linking social diversity to the size of district party system is not particularly clear. Clark and Golder (2006: 681) summarize Duverger’s work as arguing that social cleavages represent “natural constituencies” that generate pressures for multiple parties. Yet, for the interactive hypothesis to hold, it must also be true that SMP rules relegate these pressures to secondary importance, behind the imperatives of strategic coordination. Social cleavages thus must be simultaneously powerful enough to engender demand for parties but not powerful enough to interfere with strategic behavior. This seems unlikely, as a brief review of the conditions necessary for strategic coordination makes clear.

Cox (1997) presents several conditions necessary for strategic coordination: (1) voters cannot be indifferent between the first and second place party; (2) there can be no Condorcet winner; (3) voters must be short-term instrumentally rational; and (4) the ranking of the candidates must be common knowledge. If Cox’s conditions hold, coordination takes place and the number of candidates does not exceed M + 1, where M is district magnitude.

Salient social diversity is social diversity that is relevant in a way that shapes voter and politician behavior, the delivery of social benefits, the formulation of public policies, and other political processes (Posner, 2004). It likely interferes with several of Cox’s conditions. First, as acknowledged by Cox himself, it can affect voters’ preferences. Individuals can vote sincerely to—among other things—signal membership in their social group or capture act-contingent psychic benefits. This inhibits short-term instrumental rationality if such voters support a party with no realistic chance of winning (Horowitz, 1985: 57). Moreover, voters in contexts where social diversity is salient may be indifferent between first- and second-ranked parties not representing their own group. In both situations, low district magnitude may not induce voters to abandon their losing first-choice party, resulting in a correlation between social diversity and number of parties even in restrictive electoral systems.

Second, social diversity can affect the information available to voters. Where a social division is salient and drives vote choice, individuals may use social group size as a shortcut for the estimating size of parties, especially in places where polling data are unavailable (Chandra, 2004; Mozaffar et al., 2003). If social groups are similar in size (which occurs deterministically as social diversity increases), then voters face difficulty generating a common rank ordering of parties. If, for example, there are 10 similarly sized ethnic groups, voters might infer 10 similarly sized parties. Even short-term instrumentally rational voters have little incentive to abandon their first choice in this situation.

Third, even if voters are short-term instrumentally rational and have access to the results of preelection polls, salient social divisions may lead them to believe that such information is politically manufactured. Horowitz and Long (2016) illustrate this dynamic in the 2007 Kenyan presidential election. While polls consistently showed that Kalonzo Musyoka was a distant third in the presidential race, the majority of his Kamba co-ethnics continued to support him. Horowitz and Long speculate that the strength of the Kamba’s ethnic attachment inflated their belief in the electability of their candidate, preventing Duvergerian coordination.

Salient social divisions can also impede the strategic coordination of candidates. Cox (1997) argues that candidates may strategically collude to reduce the number choices faced by individuals on election day. But if candidates anticipate voter loyalty even when they cannot win, they may still run in order to translate electoral support into postelection resources (Mozaffar and Scarritt, 2005). Even if candidates want to coordinate, they must solve a time inconsistency problem: one candidate may promise rewards to persuade another candidate to drop out, but there is no guarantee the payout will happen, impeding coordination (Arriola, 2012). This problem can be exacerbated by social diversity if ethnic differences increase distrust between contenders. Finally, candidates—like voters—may use ethnic group size to predict their success at the polls. If this is the case, ethnic candidates should enter a race when their groups are similarly sized since each will believe he has a good chance of winning.

We remain agnostic about which of these causal mechanisms is most important. Our point is simply that where social divisions are salient enough to create demand for unique parties, they are also likely to undermine political actors’ desire and ability to engage in the strategic behavior underlying the interactive hypothesis. The interactive hypothesis is therefore founded on flawed logic. Where social diversity is salient, it is likely to influence the size of the district party system regardless of rules. Diversity should correlate with party system outcomes even in restrictive SMP systems. Where it is not salient, it is unlikely to have an impact even under permissive rules because it is unlikely to generate demand for parties.

Our argument complements work that finds new, less institutionalized democracies often lack the conditions necessary for electoral coordination (Cox, 1997; Moser, 1999; Moser et al., 2011; Moser and Scheiner, 2012). Social diversity is a contextual factor that, like a democracy’s age and institutionalization, shapes the impact of institutions (Ferree et al., 2014). While different contextual factors may interact and reinforce one another (Clark and Golder, 2006), their effects may be quite distinct and warrant separate consideration.

Testing the interactive hypothesis: Prior work

Initial empirical work employed cross-national research designs and found empirical support for the interactive hypothesis (Amorim-Neto and Cox, 1997; Brambor et al., 2006; Clark and Golder, 2006; Mozaffar et al., 2003; Ordeshook and Shvetsova, 1994). Yet, these studies are subject to the well-known weaknesses of the cross-national research design. National measures of social diversity are unreliable. Studies frequently use ethnic diversity as a measure of social diversity, but a universally accepted definition of ethnicity does not exist, and its meaning, salience, and measurement vary greatly across countries (Jones, 1997; Lublin, 2015; Potter 2014; Stoll, 2008, 2013). Cross-national designs also face causal identification problems. Choices over electoral institutions are not independent of existing party systems. More likely, the causal relationship is bidirectional (Bawn, 1993; Boix, 1999; Lijphart, 1994; Ordeshook and Shvetsova, 1994). Omitted variable bias also threatens conclusions made about the effects of social diversity. The salience of social divisions, for example, could correlate with a country’s degree of diversity, the choice of institutions, and/or the nature of the party system (Stoll, 2013).

Finally, theories of electoral systems—including the interactive hypotheses—apply to the district, not national, level. Studies seeking to test the interactive hypothesis should therefore use the district as the appropriate unit of analysis (Jones, 1997; Moser and Scheiner, 2012). Shifting to district-level analysis also mitigates many identification problems associated with national-level studies. If chosen by central governments, electoral institutions are exogenous to district party systems, and the salience and nature of social divisions is more uniform within countries than across them.

A growing number of studies exploit the advantages of district-level data to examine the interactive hypothesis, finding mixed support (Geys, 2006; Jones, 1997, 2004; Penas, 2004; Potter, 2014; Moser and Scheiner, 2012; Moser et al., 2011; Singer and Stephenson, 2009; Vatter, 2003). Geys (2006) discovers a significant interactive effect in a study of Belgian municipalities. Singer and Stephenson (2009), examining district-level outcomes across a large number of countries, also find support for the interactive hypothesis, although it is important to note that they measure ethnic diversity at the national level. Jones (1997, 2004) shows that social diversity shapes the number of effective candidates in majority-runoff elections for the Louisiana State House of Representative and in presidential elections, as the interactive hypothesis would predict. Conversely, Penas (2004) and Vatter (2003) find more support for additive rather than multiplicative models of the effects of electoral institutions and social diversity in their study of European elections using PR. Moser et al. (2011) and Moser and Scheiner (2012) show that ethnic diversity strongly correlates with district-level party outcomes in a wide variety of single member district elections, a result sharply at odds with the interactive hypothesis. Stoll (2013), studying presidential elections (a country-wide single member district), also finds effects for social diversity. Potter (2014) provides evidence that the interactive effect emerges only in certain national contexts; in particular, those with limited cross-district diversity.

We contribute to this growing literature using the approach of Moser and Scheiner (2012), who examine mixed-member electoral systems to draw direct comparisons between SMP and PR institutions. Unlike Moser and Scheiner (2012), however, we study a municipal rather than national mixed-member electoral system. The municipal mixed-member system design offers important advantages, which we discuss below.

Research design: South Africa’s municipal mixed-member electoral system

Mixed-member electoral systems have become increasingly common since the early 1990s. These systems, typically used for national legislatures, combine single member districts elected through majority or plurality rules, with an at-large multi-member district elected via PR. Moser and Scheiner (2012) suggest that mixed-member systems offer an ideal laboratory for testing the interactive hypothesis because they permit a “controlled comparison” of party system outcomes under different electoral rules holding national context constant. If the interactive hypothesis holds, argue Moser and Scheiner, then social diversity should shape outcomes in the PR component of mixed electoral systems, but not the SMP one. As discussed above, they find little evidence supporting the interactive hypothesis: in five countries with mixed-member systems, social diversity shapes party systems equally under both rules. We adapt the mixed-member design and apply it to the municipal-level mixed-member electoral systems of South Africa.

South Africans have voted in five postapartheid municipal elections: 1995, 2000, 2006, 2011, and 2016. The 1995 elections used jurisdictional boundaries from apartheid. After 1995, the government embarked on an ambitious plan to redraw local government boundaries. It also allotted municipalities a completely new and important set of political and economic rights and responsibilities (Republic of South Africa, 1998, 2004). The result was 284 newly empowered municipalities containing 3774 wards for the 2000 elections. While South Africa has a dominant single party at the national level—the ruling African National Congress (ANC)—municipal elections display greater political heterogeneity. Opposition parties win 20–30% of wards (approximately 800) in each election.

In South Africa’s mixed-member municipal system, voters cast two ballots: one for a ward councilor under SMP rules and one in a municipality-wide PR contest. Municipal councils have an equal number of PR and ward seats.³ The most important institutional feature of the electoral system varying across municipalities is the size of the council, which is twice the magnitude of the PR tier. Given highly similar political borders in 2000, 2006, and 2011, we use electoral data from these elections and combine them with data from the 2001 census.⁴

Our research design offers three advantages over previous work. First, the salience of a social division is more consistent and comparable across units within one country than across many. Second, unlike most national-level designs that study just one type of diversity,⁵ we examine two with differing degrees of salience, race, and ethnicity. South Africa has four recognized racial groups: Blacks, Whites, coloreds, and Indians. Racial divisions have a long and contentious history in South Africa and race plays a well-documented role in shaping voting behavior and the party system (Ferree, 2006, 2011). South Africa also has ethnolinguistic divisions. Whites subdivide into English and Afrikaans speakers, and the Black population consists of nine different linguistic subgroups: Xhosa, Zulu, Ndebele, Pedi, Sotho, Tswana, Venda, Tsonga, and Swazi. While ethnolinguistic divisions have historical relevance in South Africa and continue to have social salience, they lack the deep divisiveness of race and play a subtler role in current democratic politics.⁶ We thus study race and ethnicity to explore whether the salience of a division affects the performance of the interactive hypothesis.

Third, the municipal-level mixed-member design offers a larger number of observations (over 200) than national-level mixed-member systems, which are limited by the number of these systems in the world. This is particularly important for the PR component of the analysis. While there are many single member districts in each national mixed-member system, each system has only one national PR tier. Hence, these designs provide little variation across PR systems. In contrast, each of our municipal systems yields a PR observation.

Our design has two potential limitations. The first concerns generalizability. Municipal elections are sometimes considered low-stakes affairs, possibly reducing the applicability of institutional theories (Hicken and Stoll, 2008). If voters view the election as insignificant, they may be more likely to vote expressively, based on social identity considerations, thereby reducing the applicability of the interactive hypothesis and institutional theories in general. Institutional theories may also be less applicable to relatively young democracies like South Africa’s (Clark and Golder, 2006; Cox, 1997). If either scenario holds, then our findings may not generalize to the full set of cases—particularly to high stakes elections in long-term consolidated democracies. In essence, we could be working with a least likely case for the interactive theory; if so, then our analysis might be seen as establishing the scope conditions for the general theory rather than evaluating it for all contexts. The second limitation is possible contamination between the SMP and PR elections. If candidates run in SMP elections only to increase their chances of winning in the PR election, then any test seeking to compare social diversity’s impact on each election type will generate misleading results.

Neither critique stands up to critical evaluation. First, South Africa’s municipal elections are not low stakes affairs. Municipal councils lead the “developmental crusade” in South Africa and, in addition to setting and collecting taxes, provide important local public goods (water, electricity, health, schools, and housing) that assume high relevance in a developing country where most people had no access to basic services until recently (Siddle, 2011). They are the most visible and immediate form of government to many South Africans and have become the focal point for extensive protest (Atkinson, 2007; Siddle, 2011). Local elections also offer a rare opportunity for opposition parties to chip away at ANC dominance (Ferree, 2011) and local seats provide a means for parties to reward activists (Sparks, 2011). Turnout figures provide an indicator of the importance of local elections. Observers of the 2006 local election estimated turnout at around 50% (South Africa Civil Society Observation Coalition, 2006). In 2011, the Independent Electoral Commission recorded turnout of 58% (South African Press Association, 2011). In comparison, turnout in national elections in 2004 and 2009 was around 57%. These characteristics suggest that municipal elections are in fact consequential and thus appropriate for testing institutional theories.

Furthermore, South Africa has a highly institutionalized party system. The ANC formed a century ago, and the primary opposition parties also boast long political histories. South Africa’s high district magnitude in national elections and parliamentary system heightens party discipline and strengthens party labels. Electoral volatility is low: a handful of parties have controlled most votes in elections to date.⁷ Independent candidates run in ward elections but rarely win (Ndletyana, 2007).⁸ The high degree of institutionalization of the South African party system should moderate the effects of its youth (Moser, 1999). In addition, neither the stakes of municipal elections nor the youth of South African democracy should preclude the applicability of institutional theories. However, to the degree that our findings apply only to lower stakes elections in consolidating democracies, this is still a large category of elections that deserves greater scrutiny.

A second design limitation is possible contamination between ward and municipal elections (Cox and Schoppa, 2002; Ferrara, 2004; Herron and Nishikawa, 2001). Some studies find that parties running candidates in the lower tier SMP races receive an electoral bonus in the form of higher votes in the upper tier PR race (Cox and Schoppa, 2002; Ferrara, 2004; Herron and Nishikawa, 2001). If true, then Duverger’s logic may be violated in the lower tier races if parties run uncompetitive candidates to influence the PR election. Moser and Scheiner (2012) argue the risks of contamination have been overstated and, if anything, should bias effects toward zero. Moreover, even if we accept contamination as a problem, its effects should be weakest in systems that, like South Africa’s, employ dual ballots, equally balance PR and SMP seats, and use plurality rules in the single member district elections (Ferrara and Herron, 2005). We nevertheless take the possibility of contagion seriously and pursue multiple strategies to evaluate it.

Hypotheses and data

Our first two hypotheses directly test the interactive hypothesis. Following Moser and Scheiner (2012), hypothesis 1 (H1) is a straightforward operationalization of the interactive hypothesis: we should see no relationship between diversity and the ENP in SMP (ward) elections. Hypothesis 2 (H2) tests the interactive logic in PR races: the effect of diversity should increase as district magnitude increases. Hypothesis 1 is a stronger statement of the interactive hypothesis than hypothesis 2, which simply stipulates that effects should be larger in more permissive PR systems. Given the logic developed earlier, we expect neither hypothesis to hold. Hypothesis 3 (H3) evaluates our argument that salient social divisions impede strategic coordination. We expect highly salient racial diversity to correlate with ENP under both sets of electoral rules, whereas less salient ethnic diversity should have little relationship with ENP under either system.

The dependent variable for our tests is ENP, an inverse Herfindahl index that captures the number of parties and their concentration of votes (Laakso and Taagepera, 1979).⁹ We calculate ENP for wards (SMP) and municipalities (PR) over three elections: 2000, 2006, and 2011 (sample statistics are in Online Appendix Table 1A).¹⁰ In each year, ENP is slightly higher for municipal elections (1.8–2.0 parties) than for ward elections (1.7–1.8 parties).

We generate our key independent variables—racial and ethnic fractionalization at the ward and municipal levels—using 2001 South African census data. Fractionalization is a Herfindahl index of the fraction of each group in the total ward or municipal population and indicates the probability that two individuals picked at random will be from different groups.¹¹ Higher values indicate more diversity; a score of zero occurs when one group comprises the entire population. For wards, racial fractionalization has a mean of 0.13, a minimum value of 0, and a maximum value of 0.73; ethnic fractionalization ranges from 0 to 1 with a mean of 0.33. For municipalities, racial fractionalization ranges from 0 to 0.66 with a mean of 0.23; ethnic fractionalization ranges from 0 to 0.8 with a mean of 0.24.

To test the interactive hypothesis in municipal elections, we measure the number of seats in the PR election, which ranges from 3 to 108. We log to smooth the distribution. We then interact this seats variable with each diversity index.

We control for several potential confounds. Following McLaughlin (2007), we include the percentage of the population represented by each group comprising each diversity index. This is a conservative specification that ensures that any effects attributed to fractionalization are a function of fractionalization, not the size of particular groups in the index. Previous work on the interactive hypothesis did not control for the size of particular groups, a potential problem. If homogenous areas tend to be black (true in South Africa), then a correlation between diversity and ENP could reflect the effect of diversity or it could reflect the effect of fewer blacks. These represent different causal mechanisms, one focused on the effects of diversity in general, the other reflecting the impact of a specific group. To sort them out, data must have observations of homogeneous districts in which the dominant group varies, a requirement that is more likely to be met in large datasets. We also control for municipal urbanization level in municipal models. Urbanized municipalities are more socially heterogeneous than rural ones and have characteristics like greater availability of information and resources that affect the ability of voters and parties to coordinate electoral choices. We use municipality fixed effects in ward models to control for municipal-level confounds and provincial fixed effects in municipal models. Our models thus have a fairly restrictive set of controls, including some, like group size, not employed in previous tests of the interactive hypothesis

Results

Table 1 displays Ordinary Least Squares (OLS) models of the relationship between racial and ethnic fractionalization and ENP in ward elections in 2000, 2006, and 2011. We show simple bivariate models for each type of fractionalization (first and third column) and models with controls for group composition and municipal fixed effects (second and fourth column). The final model (fifth column) includes both types of fractionalization as well as all controls.¹² Across all models, and contrary to hypothesis 1, racial fractionalization robustly correlates with ward ENP. Figure 1, which shows the simple bivariate relationship for 2000, demonstrates this correlation and suggests a linear relationship. Scatterplots for the other years show similar patterns. Coefficient estimates range from 0.344 (for the 2011 model including ethnic fractionalization as well as all controls) to 1.232 (for the 2006 bivariate model) and are significant at the 0.05 level or higher. With the highest estimate, a change of two standard deviations in racial fractionalization (0.23) implies approximately 0.28 additional effective parties. Given the small average ENP in these elections (around 1.6–1.7), this is a sizeable effect.¹³ In the fully controlled model for the last time period, the effect size is smaller (0.08 additional effective parties) but still significant at the 0.05 level. In contrast, ethnic fractionalization is only significant in one of three bivariate models (column 3). When we control for the size of groups and municipal fixed effects, we do see significant although small correlations (column 4), but these disappear when we also control for racial fragmentation (column 5).

Table 1.

Ward level SMP models.

	1	2	3	4	5
2000
Racial fragmentation	1.000*** (0.0748)	1.036*** (0.0784)			1.119*** (0.096)
Ethnic fragmentation			0.092 (0.069)	0.413*** (0.0818)	−0.130* (0.075)
Constant	1.521*** (0.0255)	1.126*** (0.0763)	1.623*** (0.036)	1.613*** (0.104)	1.679*** (0.103)
Observations	3761	3761	3761	3761	3761
R ²	0.150	0.535	0.003	0.379	0.538
2006
Racial fragmentation	1.232*** (0.0932)	0.533*** (0.126)			0.557*** (0.112)
Ethnic fragmentation			0.331*** (0.108)	0.625*** (0.144)	−0.062 (0.081)
Constant	1.552*** (0.0308)	1.769*** (0.0856)	1.613*** (0.041)	1.471*** (0.147)	1.592*** (0.0765)
Observations	3406	3406	3406	3406	3406
R ²	0.160	0.511	0.029	0.433	0.514
2011
Racial fragmentation	0.593*** (0.125)	0.419*** (0.127)			0.344** (0.140)
Ethnic fragmentation			−0.0678 (0.0998)	0.278*** (0.091)	−0.045 (0.090)
Constant	1.702*** (0.0399)	1.691*** (0.0886)	1.794*** (0.0507)	1.568*** (0.0840)	1.598*** (0.0823)
Observations	2863	2863	2863	2863	2863
R²	0.034	0.526	0.001	0.503	0.528
Racial group controls		×			×
Ethnic group controls				×	×
Municipal fixed effects		×		×	×

Note: SMP: single member plurality. Robust standard errors in parentheses clustered by municipality.

***p < 0.01; *p < 0.1; **p < 0.05.

Figure 1.

ENP and racial diversity in 2000 South African ward races. ENP: effective number of party.

Table 1 reveals attenuation in the coefficient on racial diversity over time, both in the bivariate models (column 1) and model with controls (column 5). This result might support theories that institutional effects build over time as learning occurs. However, an alternative explanation is that the 2000 census became an increasingly less precise measure of diversity. Unfortunately, we do not have a way to parse these explanations in this data and therefore refrain from drawing strong conclusions.

Table 2 displays OLS models of the relationship between racial and ethnic fractionalization and ENP in municipal at-large (PR) elections in 2000, 2006, and 2011. Columns 1 and 5 show bivariate correlations for each type of fractionalization; we then add controls for urbanization, group percentages, and provincial effects in columns 2 and 6. Columns 3 and 7 include seats (logged) and the seats–fractionalization interaction terms without controls, which we include in the final models in columns 4 and 8.

Table 2.

Municipal level PR models.

	1	2	3	4	5	6	7	8
2000
Racial Fragmentation	1.130*** (0.128)	0.233 (0.268)	−0.206 (0.429)	0.145 (0.418)
Seats (logged)			−0.122** (0.0537)	−0.00962 (0.0493)			−0.0389 (0.0560)	0.0683 (0.0460)
Racial Fragmentation × Seats (logged)			0.539*** (0.162)	0.0428 (0.154)
Ethnic Fragmentation					−0.0481 (0.121)	0.0209 (0.159)	−0.0347 (0.374)	0.443 (0.316)
Ethnic Fragmentation × Seats (logged)							0.00111 (0.135)	−0.177 (0.115)
Constant	1.584*** (0.0378)	2.732*** (0.637)	1.906*** (0.146)	2.750*** (0.642)	1.852*** (0.0410)	1.499*** (0.189)	1.944*** (0.143)	1.339*** (0.220)
Observations	233	233	233	233	228	228	228	228
R ²	0.254	0.536	0.290	0.537	0.001	0.426	0.006	0.433
2006
Racial Fragmentation	1.517*** (0.153)	0.0230 (0.286)	−0.466 (0.518)	−0.446 (0.446)
Seats (logged)			−0.246*** (0.0674)	−0.058 (0.051)			−0.170** (0.0690)	−0.056 (0.043)
Racial Fragmentation × Seats (logged)			0.785*** (0.198)	0.220 (0.161)
Ethnic Fragmentation					−0.160 (0.149)	0.055 (0.151)	−0.692 (0.462)	−0.120 (0.309)
Ethnic Fragmentation × Seats (logged)							0.219 (0.169)	0.071 (0.119)
Constant	1.504*** (0.0453)	1.839*** (0.653)	2.145*** (0.182)	1.938*** (0.658)	1.883*** (0.0506)	1.326*** (0.203)	2.293*** (0.175)	1.152** (0.286)*
Observations	230	209	230	230	229	229	229	229
R²	0.302	0.735	0.348	0.721	0.005	0.635	0.034	0.640
2011
Racial fragmentation	0.758*** (0.159)	−0.22 (0.300)1	−0.548 (0.536)	0.145 (0.468)
Seats (logged)			−0.249*** (0.0699)	0.0503 (0.0547)			−0.172** (0.0628)*	−0.04 (0.04569)
Racial Fragmentation × Seats (logged)			0.485** (0.206)	−0.171 (0.169)
Ethnic fragmentation					−0.341** (0.140)	0.176 (0.163)	−0.55 (0.4267)	0.23 (0.3318)
Ethnic Fragmentation × Seats (logged)							0.10 (0.1542)	−0.03 (0.1268)
Constant	1.771*** (0.0472)	1.897*** (0.684)	2.419*** (0.188)	1.822*** (0.690)	2.020*** (0.0466)	1.438*** (0.291)	2.431*** (0.160)	1.622*** (0.319)
Observations	232	207	226	207	225	207	225	207
R²	0.09	0.655	0.348	0.738	0.007	0.720	0.039	0.722
Controls (percent urban, ethnic group percentages, and provincial fixed effects)		×		×		×		×

Note: PR: proportional representation. Standard errors in parentheses.

***p < 0.01, **p < 0.05, *p < 0.1.

Several patterns emerge in Table 2, some surprising. First, while racial fractionalization strongly correlates with municipal ENP in bivariate models, these correlations disappear when we add simultaneous controls for group size, percent urban, and province fixed effects. The model with the full set of controls is particularly demanding and may decrease the data’s variation to a point where we cannot differentiate the effects of diversity from the other factors. Group size—especially the size of the African population—particularly weakens the correlation between diversity and ENP, largely because homogenous municipalities tend to be Black. Homogeneous White, colored, or Indian municipalities (as opposed to wards) are rare. Thus, although it is clear that racial demographics influence municipal party systems, we cannot parse the specific mechanism—whether it is diversity or the size of particular groups.

Second, and contrary to our expectations, we find some support for the interactive hypothesis in the municipal data. When we run the model with racial fractionalization, seats, and their interaction, the interaction is significant and we see the relationship predicted by the interactive hypothesis. Significant effects persist when we add controls one by one. Once we include the full set of controls (column 4), however, the relationship disappears. This could indicate a spurious result for the interactive hypothesis or it could again reflect insufficient data to parse all variables in the model.

Third, ethnic fractionalization is significant in only one bivariate municipal model (column 5) and only at the 0.10 level. Interactions between ethnic fractionalization and seats are insignificant in all models.

Figure 2 shows the interactive hypothesis results for 2006.¹⁴ The upper left panel, based on the unconstrained model of racial fractionalization (Table 2, column 1), demonstrates the interactive relationship. At low district magnitude of three seats (a logged value of approximately 1), racially diverse areas do not produce more parties than racially homogeneous areas. Both have around two effective parties. As district magnitude increases, however, racially diverse areas develop bigger party systems whereas racially homogeneous areas do not. When district magnitude is around 148 (a logged value of approximately 5), racially diverse areas have nearly two more effective parties than racially homogeneous areas. However, as shown in the upper right panel, this relationship disappears after we control for percent urban, province fixed effects, and racial group size. Additionally, the relationship does not emerge at all for ethnic fractionalization, as shown in the bottom two panels of Figure 2.

Figure 2.

The interactive hypothesis in South African municipalities (2006), with racial and ethnic fragmentation.

Altogether, our analysis leads us to reject hypothesis 1, as there is no effect of social diversity in SMP systems. We find robust effects for racial diversity in South Africa’s ward elections and therefore join Moser and Scheiner (2012) and Stoll (2013) in questioning the SMP component of the interactive hypothesis. We also find support for hypothesis 3, suggesting that the salience of a social division shapes its impact on the party system.

Our findings for hypothesis 2—that social diversity and district magnitude should interactively shape ENP in PR systems—are less conclusive. When we include a full set of controls, we find no interactive effect, evidence against hypothesis 2. However, when we relax the model, including controls in a piecemeal fashion, support for the interactive hypothesis emerges. These results suggest support for hypothesis 2, inducing us to moderate our earlier argument: salient social diversity may impede electoral coordination and generate effects in highly restrictive SMP systems, but its effects may be even stronger in permissive high magnitude PR systems. This represents a refinement of the interactive hypothesis rather than a full dismissal of it and suggests that salient social divisions inhibit but do not completely eliminate electoral coordination. However, an alternative interpretation of our data is that bare bones empirical models of the interactive hypothesis that do not control for the size of groups conflate the effects of homogeneity with the effects of particular groups. This interpretation would suggest rejection rather than refinement of hypothesis 2. As we cannot adjudicate between these two possibilities with the data we have, we surmise that support for the interactive hypothesis in PR systems is partial at best.

A closer look at contamination

Prior studies on mixed-member electoral systems highlight the potential for contamination between upper and lower tiers. While contagion should be minimal in mixed-member systems with the constellation of institutions found in South Africa (Ferrara and Herron, 2005), we nevertheless consider the possibility that ward races are not fully independent of municipal races. Municipal fixed effects neutralize many sources of contamination, and the robustness of our ward results to their inclusion gives us confidence that contamination does not drive our findings. Our results are also robust to controlling directly for the number of parties in the municipal race (with no municipal fixed effects), and the coefficient on municipal number of parties is generally insignificant and inconsistent across models. We nonetheless pursue two additional strategies to explore the robustness of our finding to contagion effects.

First, we examine the relationship between racial fractionalization and party system outcomes in a sample of nearly 600 by-elections held between 2001 and 2008.¹⁵ By-elections occur at the ward level, without simultaneous municipal elections. There should be, therefore, little risk of contagion in these elections. If our basic results hold, then it is unlikely that contamination generates them. This is indeed the case in bivariate models (Table 2, column 1): racial fractionalization has a positive, well-estimated relationship with the effective number of candidates. When we include racial and municipal fixed effects, the fractionalization result disappears. This attenuation likely reflects the challenges of trying to estimate fractionalization and racial group effects, which are strongly correlated, in a relatively small sample, especially when also neutralizing cross-municipal variation.

As a second cut, we split our 2006 ward sample according to whether the ward was in a majority marginal municipality (MMM) in 2000 (any municipality where the largest party had less than 52% of the vote). Contagion effects should be strongest in wards in municipalities that were competitive in 2000 because these are races parties should prioritize if they enter ward races to influence municipal outcomes. Table 3 (columns 3–6) shows the results for 2006. The coefficients on racial fractionalization for the pooled, MMM, and non-MMM samples are all positive, well-estimated, and very similar in size and exhibit little evidence of contagion; they also remain robust to racial controls and municipal fixed effects.

Table 3.

Contagion in South African ward elections?

	By-elections (no concurrent PR race)		Competitive municipalities		Uncompetitive municipalities
Racial diversity	0.614 (4.38)***	−0.003 (0.02)	0.977 (3.58)***	0.609 (3.60)***	1.060 (12.41)***	0.530 (7.34)***
Observations	580	580	470	470	2936	2936
R ²	0.05	0.48	0.10	0.39	0.14	0.45
Racial controls		×		×		×
Municipal fixed effects		×		×		×

Note: PR: proportional representation. Robust t-statistics in parentheses. All models include (but do not show) constants.

***Significant at 1% level.

In addition, contamination appears to have little impact on our estimates of the relationship between racial fractionalization and the ENP, providing greater confidence for our ward-level results.

Conclusion

Where social diversity is politically salient enough to increase demand for parties, it also likely shapes the ability and desire of voters and candidates to engage in strategic coordination. Contrary to the consensus in the comparative politics literature that social diversity and electoral rules interactively shape party systems, we suggest that salient social diversity should correlate with the number of parties even in places that use restrictive SMP rules. In contrast, where social diversity is not salient, it should have little impact on the party system.

We test this intuition using a case that allows us to simultaneously examine the effects of social diversity under SMP and PR rules for a large sample: South Africa’s mixed municipal-level electoral system. We considered two types of social diversity with varying salience levels: racial diversity (high salience) and ethnolinguistic diversity (lower salience). In South Africa’s ward elections, which operate according to SMP rules, we find that racial diversity strongly correlates with the ENP. Effects are small and diminish somewhat over time (perhaps evidence of learning) but nonetheless are quite robust. Racial diversity also correlates with ENP in at-large municipal PR races, and its effects appear to increase with district magnitude, suggesting some support for the interactive hypothesis in PR systems. However, these relationships are sensitive to the inclusion of controls for group size and may simply reflect spurious correlations. In contrast, we observe that ethnic diversity has little to no relationship with ENP in either ward or municipal races.

Our results thus add to accumulating evidence questioning the interactive hypothesis and point to the importance of examining not just the effects of social diversity but also salience of social divisions, when exploring the social origins of party systems.

Supplemental material

Supplemental Material, PPQ_Appendix_Table_1_(1) - Why the salience of social divisions matters in party systems: Testing the interactive hypothesis in South Africa

Supplemental Material, PPQ_Appendix_Table_1_(1) for Why the salience of social divisions matters in party systems: Testing the interactive hypothesis in South Africa by Karen Ferree, Clark Gibson, and Barak Hoffman in Party Politics

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Supplemental material

Supplementary material for this article is available online.

References

Amorim-Neto

Cox

(1997) Electoral institutions, cleavage structures, and the number of parties. American Journal of Political Science 41 (January): 149–174.

Arriola

(2012) Multiethnic Coalitions in Africa: Business Financing of Opposition Election Campaigns. New York: Cambridge University Press.

Atkinson

(2007) ‘Taking to the streets: has developmental local government failed in South Africa?’ In: Buhlungu

Daniel

Southall

Lutchman

(eds) State of the Nation South Africa 2007. Cape Town: Human Sciences Research Council, pp. 53–77.

Bawn

(1993) The logic of institutional preferences: German electoral law as a social choice outcome. American Journal of Political Science 37 (November): 965–989.

Boix

(1999) Setting the rules of the game: the choice of electoral systems in new democracies. American Political Science Review 93 (September): 609–624.

Brambor

Clark

Golder

(2006) Are African party systems different? Electoral Studies 26 (June): 315–323.

Chandra

(2004) Why Ethnic Parties Succeed: Patronage and Ethnic Head Counts in India. New York: Cambridge University Press.

Clark

Golder

(2006) Rehabilitating Duverger’s theory: testing the mechanical and strategic modifying effects of electoral laws. Comparative Political Studies 39 (August): 679–708.

Cox

(1997) Making Votes Count. Cambridge: Cambridge University Press.

10.

Cox

Schoppa

(2002) Interaction effects in mixed-member electoral systems: theory and evidence from Germany, Japan, and Italy. Comparative Political Studies 35 (November): 1027–1053.

11.

Duverger

(1952) ‘Public opinion and political parties in France. American Political Science Review 46 (December): 1069–78.

12.

Duverger

(1963) Political Parties. New York: Wiley.

13.

Ferrara

(2004) Electoral coordination and the strategic desertion of strong parties in compensatory mixed systems with negative vote transfers. Electoral Studies 23 (September): 391–413.

14.

Ferrara

Herron

(2005) Going it alone? Strategic entry under mixed electoral rules. American Journal of Political Science 49 (January): 16–31.

15.

Ferree

(2011) Framing the Race in South Africa: The Political Origins of Racial Census Elections. Cambridge: Cambridge University Press.

16.

Ferree

Powell

Scheiner

(2014) Context, electoral rules, and party systems. Annual Review of Political Science 17: 421–439.

17.

Ferree

(2006) Explaining South Africa’s racial census. Journal of Politics, 68 (November): 803–815.

18.

Geys

(2006) District magnitude, social heterogeneity and local party system fragmentation. Party Politics 12 (March): 281–297.

19.

Herron

Nishikawa

(2001) Contamination effects and the number of parties in mixed superposition electoral systems. Electoral Studies 20 (March): 63–86.

20.

Hicken

Stoll

(2008) Electoral rules and the size of the prize: how political institutions shape presidential party systems. Journal of Politics 70 (September): 1109–1127.

21.

Horowitz

(1985) Ethnic Groups in Conflict. Berkeley: University of California Press.

22.

Horowitz

Long

(2012) Does Ethnicity Reduce Strategic Voting? New York: Mimeo.

23.

Horowitz

Long

(2016) Strategic voting, information, and ethnicity in emerging democracies: Evidence from Kenya. Electoral Studies 44: 351–361.

24.

Independent Electoral Commission (IEC) of South Africa. Election data. http://www.elections.org.za/content/Elections/Municipal-elections-results/ (accessed 12 April 2017).

25.

Jones

(1997) Racial heterogeneity and the effective number of candidates in majority runoff elections: evidence from Louisiana. Electoral Studies 16 (September): 349–358.

26.

Jones

(2004) Electoral institutions, social cleavages, and candidate competition in presidential elections. Electoral Studies 23 (March): 73–106.

27.

Laakso

Taagepera

(1979) “Effective” number of parties: a measure with application to west Europe. Comparative Political Studies 12 (April): 3–27.

28.

Lijphart

(1994) Electoral Systems and Party Systems. Oxford: Oxford University Press.

29.

Lublin

(2015) Electoral systems, ethnic heterogeneity and party system fragmentation. British Journal of Political Science 47: 373–389.

30.

McLaughlin

(2007) Beyond the racial census: the political salience of ethnolinguistic cleavages in south Africa. Comparative Political Studies 40 (April): 435–456.

31.

Moser

(1999) Electoral systems and the number of parties in post-communist states. World Politics 51 (April): 359–384.

32.

Moser

Scheiner

(2012) Electoral Systems and Political Context: How the Effects of Rules Vary Across New and Established Democracies. Cambridge: Cambridge University Press.

33.

Milazzo

Moser

Ethan

(In press) Social Diversity Affects the Number of Parties Even under First-Past-the-Post Rules. Comparative Political Studies 51 (July): 938–974.

34.

Mozaffar

Scarritt

(2005) The puzzle of African party systems. Party Politics 11 (July): 399–421.

35.

Mozaffar

Scarritt

Galaich

(2003) Electoral institutions, ethnopolitical cleavages, and party systems in Africa’s emerging democracies. American Political Science Review 97 (August): 379–390.

36.

Ndletyana

(2007) Municipal elections 2006: protests, independent candidates and cross-border municipalities. In: Buhlungu

Daniel

Southall

Lutchman

(eds) State of the Nation South Africa 2007. Cape Town: Human Sciences Research Council, pp. 146–170.

37.

Ordeshook

Shvetsova

(1994) Ethnic heterogeneity, district magnitude, and the number of parties. American Journal of Political Science 38 (February): 100–123.

38.

Penas

(2004) Cleavages and thresholds: the political consequences of electoral law in the Spanish autonomous communities, 1980–2000. Electoral Studies 23 (March): 23–43.

39.

Posner

(2004) Measuring ethnic fractionalization in Africa. American Journal of Political Science, 48 (October): 849–863.

40.

Potter

(2014) Demographic diversity and district-level party systems. Comparative Political Studies 47(13): 1801–1829.

41.

Republic of South Africa (RSA) (1998) The White Paper on Local Government. Ministry for Local Government and Provincial Affairs, Government of the Republic of South Africa, Pretoria.

42.

Republic of South Africa (RSA) (2004) Local Government Municipal Rates Act 2004. Cape Town: Parliament of the Republic of South Africa.

43.

Sartori

(1976) Parties and Party Systems. Cambridge: Cambridge University Press.

44.

Siddle

(2011) Decentralisation in South African local government: a critical evaluation. Doctoral Thesis. Graduate School of Business, University of Cape Town, South Africa.

45.

Singer

Stephenson

(2009) The political context and Duverger’s theory: evidence at the district level. Electoral Studies 28 (September): 480–491.

46.

South Africa Civil Society Observation Coalition (SACSOC) (2006) ‘SACSOC Commends Free and Fair Municipal Elections. 2 March. Available at: http://www.eisa.org.za/PDF/SACSOC06lg.pdf (accessed 12 April 2017.)

47.

South African Press Association (SAPA) (2011) Voter Turnout Highest Ever – IEC. News 24, 21 May. Available at: http://www.news24.com/SouthAfrica/Local-Elections-2011/Voter-turnout-highest-ever-IEC-20110521# (accessed 12 April 2017).

48.

Sparks

(2011) “The Swing’s the Thing” That Shows the Political Trend. Business Day, 25 May.

49.

Statistics South Africa. The 2001 Census Data. Census Bureau of South Africa (Statistics South Africa): http://www.statssa.gov.za/?page_id=4082 (accessed 12 April 2017).

50.

Stoll

(2008) Social cleavages and the number of parties: how the measures you choose affect the answers you get. Comparative Political Studies 41 (November): 1439–1465.

51.

Stoll

(2013) Changing Societies, Changing Party Systems. New York: Cambridge University Press.

52.

Vatter

(2003) Legislative party fragmentation in Swiss cantons: a function of cleavage structures or electoral institutions? Party Politics 9 (July): 445–461.

Supplementary Material

Please find the following supplemental material available below.

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

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

0.00 MB

0.11 MB

0.39 MB