Economic Integration, Party Polarisation and Electoral Turnout

Abstract

Recent research provides evidence that economic integration has a negative effect on electoral turnout. Taking up these recent findings, this article explores the causal chain in more detail. Specifically, it argues that one way by which economic integration affects the calculus of voting is through the positioning of political parties. The expectation is that the polarisation between parties on an economic left–right scale is lower the more integrated an economy is. Consequently, electoral turnout should be lower with less polarisation in the party system. The article employs aggregate-level data from legislative elections in 24 developed democracies. Using data from the Comparative Manifestos Project, evidence is found not only that economic integration has a negative effect on party polarisation as measured on an economic left–right dimension, but also that this in turn exerts a negative effect on electoral turnout.

Acknowledgements

Previous versions of this article have been presented at the 68th Annual Meeting of the Midwest Political Science Association 2010, at the 2010 Spring Meeting of the methodology section of the German Political Science Association (DVPW), and in a seminar at the Johannes Gutenberg University of Mainz. We wish to thank the participants, and especially Robert Lowry and Edeltraud Roller, for their critical and constructive discussions of our article. We gratefully acknowledge the insightful comments of two anonymous reviewers as well as editorial remarks that helped us refine our arguments. Sara Ceyhan and Thomas Speth provided excellent research assistance.

Notes

1. Some of the more recent contributions include, on tax policy, Plümper et al. (2009) and, on public spending, Busemeyer (2009).

2. For a more thorough discussion on this causal model, see Steiner (2010); for yet another causal model operating through political attitudes and specifically feelings of identity, see Rahn (2008). Further evidence on the attitude mechanism is discussed by Vowles (2008) and Marshall and Fisher (2010).

3. We draw here on an analogous argument that has been put forward with regard to the effects of European integration (Nanou and Dorussen 2011) and has been kindly brought to our attention by an anonymous reviewer.

4. If Ward et al. (2011) are correct in their conditional hypothesis, we should observe more variance between countries with respect to parties' reaction to globalisation influence than would be the case if an unconditional hypothesis were to adequately capture the data generating process. However, this should not influence the relation between economic integration and the choices offered to voters, but only render estimation somewhat more difficult.

5. These countries include: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Israel, Italy, Japan, Luxembourg, the Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, the United Kingdom and the United States. Only elections that are reasonably free and fair are included. The first observation for Portugal is thus the parliamentary election of 1976, for Spain the one in 1977, and for Greece the one in 1974.

6. Due to missing CMP data, midterm elections in the US had to be excluded, however.

7. An alternative weighting procedure with vote shares gives next to identical values for dispersion.

8. As Pontusson and Rueda (2010) remind us, economic inequality might be another important determinant of parties' economic policy positions. We do not deny the potential significance of this variable on a theoretical level. However, given the scope of our sample we have to omit this variable from our empirical estimations for reasons of data availability.

9. These issues are amplified to the extent that the measurement problems associated with party polarisation result in a measure that is better at identifying true variation between countries than within countries at different elections (recall that measurement errors lead us to employ moving averages of initial positions in the first place). While this is, of course, something we cannot prove, so that we ultimately have to leave the judgment to the readers, we strongly believe this to be the case.

10. Note that the models we present here all exclude Israel due to missing data on the occupation variables.

11. Standard deviations, maxima and minima refer to a sample including only the observations contained in models 1 to 5.

12. When adding the KOF index of economic integration to the specification of models 3 and 4, we obtain a strong and highly significant negative effect of integration on turnout as in previous research. This negative effect gets even larger when omitting the dispersion measure (but keeping the same set of observations). The coefficient on the dispersion measure itself is smaller (coefficients are 0.85 for the voting age and 0.80 for the models based on registered eligible voters) than in the presented models and reaches statistical significance in only one of the models, i.e. for the voting age measure of turnout. (Note that this is clearly not driven by the fact that we lose some observations when entering economic integration.) The observed pattern is in line with our theoretical expectations as we argue that only part of the effect of economic integration operates through party dispersion such that the negative effect of economic integration should be reduced, but still exist when controlling for dispersion. As our model assumes the effects of economic integration and economic policy dispersion to be intertwined to some extent, we expect at the same time that the effect of dispersion can be more efficiently identified in models not including economic integration. It is clearly not possible to reliably estimate the full effect of both variables in one regression model. Hence, economic integration is not included in the models presented.

13. We tried to estimate the otherwise exact same specifications as in models 1 and 2, but had to drop two variables from the turnout models in order to avoid collinearity problems which made the estimation through GLS impossible. We dropped voting age and population since these two variables clearly failed to reach statistical significance in all of the initial models.

14. An analogous bivariate test for dispersion's turnout effect seemed infeasible to us, because of the far greater number of identified strong covariates.

15. We have also checked a measure of the distance between the two most extreme party positions recorded. This results in a clear nil finding which we take not as evidence against our hypothesis but rather for the poor validity of this measure as compared to the concept we intend to capture. Note that this measure might also be strongly affected by decisions of the Manifesto researchers on which parties to include.

16. For instance as an anonymous reviewer suggested, it is plausible that globalisation processes could trigger the emergence of right-wing populist parties advocating nationalistic immigration policies. To the extent that established parties do not follow suit, this could mitigate the turnout depressing effect of economic integration by increasing polarisation.