Socioeconomic heterogeneity and party system fragmentation

ABSTRACT

Considerable evidence suggests that social diversity increases the number of political parties. While a vast literature examines this relationship by measuring social diversity as ethnic fragmentation, the convenience of capturing the level of social heterogeneity by using alternative indicators has been traditionally overlooked. In this article, I offer new evidence that documents the impact of social diversity on party system size by proxying the former as the number of social classes. Using both objective class positions and subjective class identifications and employing two different datasets that cover democratic legislative elections around the world between 1981 and 2015, I find that the effective number of social classes has a positive effect on party system fragmentation when the average district magnitude in a democracy is high. Given the difficulty to test these effects in a cross-country environment net of unobserved heterogeneity, I examine the robustness of the findings in Spain between 2000 and 2016, an ideal case due to large differences in district magnitude and a changing party system. Overall, the results suggest that high levels of social diversity are likely to increase party system fragmentation when the electoral system is permissive enough.

KEYWORDS:

• Party systems
• electoral rules
• social cleavages
• Spain
• cross-sectional time series

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 All data from Bormann and Golder’s dataset have been updated with information from Gallagher’s online dataset (www.tcd.ie/Political_Science/Staff/Michael.Gallagher/ElSystems/index.php) and the electoral commissions’ websites for each country. Information for the Spanish part is available at: Available at http://www.infoelectoral.mir.es.

2 Table A6 in the Appendix shows that results remain almost identical when I adopt an alternative definition of democracy and restrict the analyses to the observations that attained a Polity score of at least 8 in the 2008 edition of the Polity IV dataset. Polity IV Project: Political Regime Characteristics and Transitions, 1800–2010. Available at http://www.systemicpeace.org/polity/polity4.htm [last accessed 19th of November of 2017]. Results are also substantively identical when I exclude each country or district consecutively (see Figures A2-A4 in the Appendix).

3 The single-member seats in Ceuta and Melilla are filled by the plurality winner.

4 Seats in each multi-member constituency are allocated according to the D’Hondt method. In order to participate in the apportionment of seats, a list must receive at least three percent of all valid votes cast in the district.

5 There are more elections than data on the distribution of social class from the World Values Survey. This is a consequence of using the same value of this variable for more than one observation. To be more specific, the socioeconomic stratification variable is used for all elections until a new wave of World Values Survey is conducted. For example, the first available observation in time for Argentina is from the survey conducted in that country in 1985. This value is used as independent variable for the 1985, 1987 and 1989 elections in Argentina because the next survey is conducted there in 1991.

6 Unlike in the other analyses, I cannot use respondents’ occupation and build an objective measure of social class because the information relevant to this end is missing.

7 The post-electoral surveys are the following: CIS 2384 for the 2000 elections (N = 5,283), CIS 2559 for the 2004 elections (N = 5,377); CIS 2757 for the 2008 elections (N = 6,083); CIS 2920 for the 2011 elections (N = 6,082); CIS 3126 for the 2015 elections (N = 6,242); and CIS 3145 for the 2016 elections (N = 6,175).

8 Oesch’s (2006) class schema distinguishes between the following 17: large employers, self-employed professionals, petit bourgeoisie with employees, petit bourgeoisie without employees, technical experts, technicians, skilled crafts, routine operatives, routine agricultural, higher-grade managers and administrators, associate managers and administrators, skilled office, routine office, sociocultural professionals, sociocultural semi-professionals, skilled service, and routine service. In order to transform the respondents’ occupations into social classes, I follow the code included in the ESS website and the Bernardi and Ares’s (2017) paper. Routine agricultural are missing in both, whereas we lack large employers in the latter. Hence, only 16 and 15 social classes can be identified in the cross-national and the Spanish parts, respectively.

9 Figure A1 if the Appendix displays some data on the distribution of these variables across countries and over time. As shown by the Figure, I do not find big differences in this respect between wealthier and developing countries. Likewise, time does not seem to affect the level of socioeconomic heterogeneity registered in one society. Hence, level of development is not likely to explain the cross-sectional variation in party system fragmentation.

10 I further test my argument by slightly departing from previous operationalizations of mean district magnitude (e.g., Clark and Golder 2006; Golder 2006) and calculating a weighted measure for multi-tier and mixed-member systems that multiplies the average district magnitude in the tiers by the seat percentages allocated in each of them. Main results are displayed in Table A11 of the Appendix and remain robust to this operation. Moreover, Table A10 of the Appendix includes some robustness analyses in which, following Cox (1997) and others, I include as additional independent variables the number of seats allocated in the upper tier on its own and interacted with the measures of social diversity. This operation does not change the main results.

11 To test the robustness of the results, in Table A7 of the Appendix I replicate Models 3 and 4 of Table 1 including alternative measures of ethnic, linguistic and religious heterogeneity (Alesina et al. 2003). The results largely confirm the second hypothesis, i.e. the interaction between the effective number of socioeconomic groups and the logged district magnitude is always positive and statistically significant at least at the 5% level.

12 For the 2000 elections, I use the 2002 Institutions and Autonomies Study (CIS 2455, N = 10,476); for the 2004 elections, I use the 2005 Autonomic Barometer (CIS 2610, N = 10,371); for the 2008 elections, I use the 2010 Autonomic Barometer (CIS 2829, N = 10,409); and for the 2011 elections, I use the 2012 Autonomic Barometer (CIS 2956, N = 11,181). For the 2015 and 2016 elections, I use data from the CIS post-electoral surveys.

13 As robustness check, in Table A9 of the Appendix I replicate the main analyses by excluding presidential coattails because, considering the work by Elgie et al. (2014), the interaction between proximity and ENPRES produces problematic baselines of comparison. Likewise, in Table A8 of the Appendix I do not use ethnic fragmentation as independent variable. Main results are robust to the exclusion of these controls.

14 I replicate the specifications of Tables 1 and 3 (see Tables A12 and A17 of the Appendix) with panel-corrected standard errors (PCSE) to correct for panel heteroskedasticity and temporally correlated errors (Beck and Katz 1995). Main results are robust to the use of this type of standard errors.

15 As robustness checks, in Tables A13 and A14 of the Appendix I replicate the main analyses by not including WVS fixed effects and the lagged dependent variable as explanatory factors, respectively. The latter is the estimation strategy also followed for the ESS part in Table A16. Main results are robust to these alternative estimations.

16 Tables A15 and A18 of the Appendix address the serial correlation between the observations of the same country in a slightly different way by specifying a series of hierarchical linear models (Luke 2004). Main results are robust to the use of this alternative specification.

Additional information

Funding

This work produced with the support of a 2019 Leonardo Grant for Researchers and Cultural Creators, BBVA Foundation.