More Money, Fewer Problems? Cross-Level Effects of Economic Deprivation on Political Representation

Abstract

While equal political representation of all citizens is a fundamental democratic goal, it is hampered empirically in a multitude of ways. This study examines how the societal level of economic inequality affects the representation of relatively poor citizens by parties and governments. Using CSES survey data for citizens’ policy preferences and expert placements of political parties, empirical evidence is found that in economically more unequal societies, the party system represents the preferences of relatively poor citizens worse than in more equal societies. This moderating effect of economic equality is also found for policy congruence between citizens and governments, albeit slightly less clear-cut.

Acknowledgements

All authors contributed equally; the principle of rotating order of authors applies. The authors gratefully acknowledge the financial support provided by the EUROCORES Programme of the European Science Foundation. Julian Bernauer has received support from the Heinrich Böll Foundation. An earlier version of this manuscript has been presented at the 3-Länder-Tagung in Basel, 13--14 January 2011 and at the Humvib Final Conference in Berlin, 8--10 September 2011. We would like to thank the workshop participants and, in particular, Thomas Bräuninger, Guillem Rico, Robert Rohrschneider, Didier Ruedin, Lisa Schädel, Jacques Thomassen, Jale Tosun, Hanna Wass and the anonymous reviewers for their helpful comments as well as Anna Walsdorff for excellent research assistance.

Notes

1. See, however, Flavin (2011), who introduces income inequality as a control variable to explain variation in political inequality across American states.

2. Please note that while studies focusing on the responsiveness of political elites are most interested in a dynamic perspective (i.e. do elites respond to changes in citizens preferences?), this study aims at documenting and explaining the actual state of closeness between the preferences of the elites and citizens, i.e. in the question of congruence between the two. Both perspectives are considered equally relevant and have their legitimation in democratic theory (for a discussion see e.g. Lax and Phillips 2012).

3. It is worth noting that the relationship between income disparities and political influence is likely to be mitigated by the laws that govern party and campaign finance (see Flavin 2011).

4. This is notably reflected in the fact that corruption tends to be higher in more unequal societies (e.g. You and Khagram 2005).

5. Our sample comprises elections from a period of 10 years. Controlling for possible temporal effects does not alter the results presented here and has very little substantial influence on the dependent variable.

6. A general justification for using a left–right policy dimension is given by Powell (2000: 162), who argues that it is ‘not only the most widely available single measure of the preferences of citizens in different countries but seems to meet reasonably well our need to capture comparably the general stances of citizens and the general policy orientations of the parties that compete for policy-making positions’. Other policy positions of voters are not available in the CSES. Arguably, there is a theoretical connection between the preferences towards redistribution and the general, (economic) left–right scale (e.g. Kumlin 2007).

7. Alternatively, the party positions can also be estimated using the Benoit–Laver expert survey data (Benoit and Laver 2006). These two measures of party positions correlate very highly (Pearson’s correlation coefficient Module 1 (1996–2001): 0.89; Module 2 (2001–2006): 0.92). The CSES Expert data carry the advantage that they are measured on the same scale as the citizen preferences and that they are time variant. To avoid additional (problematic) rescaling, we decided to use the CSES Expert data. The results are robust against the use of Benoit–Laver expert party positions.

8. We note that other options to calculate the positions of parties exist. First, we can rely on party manifesto data as collected by the Comparative Manifesto Project (CMP) (Budge et al. 2001; Klingemann et al. 2006). The use of this type of data carries the large disadvantage that by focusing on the percentage of left–right statements, we capture more the salience a party puts on left–right issues rather than a party’s substantive left–right position. Another option would be to estimate the party placement as the mean voters’ evaluation of a party, available in the CSES survey. However, it is likely that the individual idiosyncratic factors do not cancel out completely, as for example the position of the own party is often evaluated differently than the rest of the parties (see Aldrich and McKelvey 1977). In our opinion, the remedy just to use the judgements of more educated persons to calculate party positions (e.g. Alvarez and Nagler 2004) does not solve this problem entirely.

9. Arguably, proximity measures share the disadvantage that scales for citizens and elites can carry differing meanings and thus the resulting scores might be biased. However, as our research focus is largely in relative differences only, all we have to assume is that the measures are equally flawed across societal groups. Therefore, if we find that one societal group, in the present study economically disadvantaged citizens, has a larger distance to the party system compared to the rest of the population, we can at least say that they are relatively worse represented.

10. The findings presented in this study remain unchanged if one uses only the highest/lowest quintiles to construct the groups of the rich/poor.

11. The young democracies in our sample are all post-communist states. One could argue that because of their history the meaning of left and right is different in the post-communist countries (McAllister and White 2007). This would make it harder for citizens to place themselves on this scale and, most importantly, a left position would not necessarily be connected to a preference for a more equal society. Therefore, we also estimated the same models with a sample excluding post-communist countries. The results obtained for old democracies only do not differ substantially from those presented in the next section of the study.

12. A caveat of the empirical analyses is the possible endogeneity problem we are facing when looking at the link between inequality and underrepresentation. Bartels (2008: 286) posits that economic and political inequalities might be reinforcing one another. As he puts it: ‘disparities in representation are especially troubling because they suggest the potential for a debilitating feedback cycle linking the economic and political realms: increasing economic inequality may produce increasing inequality in political responsiveness, which in turn produces public policies that are increasingly detrimental to the interests of poor citizens, which in turn produces even greater inequality, and so on’. Hence, as underrepresentation could cause economic inequality as well as inequality cause underrepresentation, the effects of inequality might be overestimated in our current design. Extensions could incorporate an instrumental variable approach to isolate the endogenous part of the effect of inequality on representation, for example in a TSCS (time-series cross-section) design. Hurdles for the implementation of such an approach are to find suitable instruments which are correlated with inequality, but exogenous to representation, and to combine the random slopes multilevel model with the instrumental variable approach.

13. We opted for elections as the context level in this study. The primary reason is that the level of economic inequality (our focal independent level) is measured at this level (time variant across elections). Other options included to add the country level as level 3 in the multilevel models. This does not change the results presented here.

14. Effects in the range of −0.1 to −0.2 are still considerable given the mean value of −0.58 and the standard deviation of 0.78 on the party policy congruence scale.

15. In Models 3 and 4, the identifier for high income is excluded, as we now want to focus on the underrepresentation of the poor compared to all richer citizens after having shown in Models 1 and 2 that the higher the income, the higher the quality of representation.

16. One likely explanation is the fact that the average distances between voters and the government are larger than the distance between voters and their parties. This also influences the precision with which the interaction effect is estimated.