Who leads, who follows? Re-examining the party–electorate linkages on European integration

Abstract

This article re-examines and evaluates the link between electorates' opinions and national political parties' positions on European integration, i.e. the extent to which political parties lead and/or follow public opinion on this issue. Applying a method for causal modelling (Granger causality tests) to panel data concerning political parties' positions and voters' opinions in 15 countries from 1973 to 2003, I find (contrary to previous investigations of this relationship) that there is little empirical support for an electoral connection or reciprocal causation between party positions and electorates' opinion regarding European integration. Parties have an influence on voter opinions, but are at the same time unresponsive to changes in voter opinion.

Keywords:

• European integration
• European Union
• Granger causality
• panel data
• political parties
• public opinion

Notes

The second-order nature of European Parliament elections, which are mainly fought over national, rather than European, issues (Van der Eijk and Franklin 1991, 1996, 2004), could also be seen as confirmation of the permissive consensus hypothesis.

Responsiveness can be regarded as the degree to which political parties attempt to translate voters' preferences into public policies, and accountability refers to the extent that voters' stated policy choices are really carried out by political parties (and implemented by governments). In this study I do not address the accountability of political parties, since it is better assessed through a comparison of electoral promises and the following legislative behaviour of the political parties.

Since information is rarely transparent and easily accessible for voters, their mere possibility making reasonable decisions is limited without external information shortcuts. For a ‘bounded rational’ citizen (Simon 1982, 1985) it makes sense to use available ‘information shortcuts’ to make reasonable decisions with minimal cognitive effort. This argument also applies to ‘rational voters’, since it is reasonable for the individual voter to use cues or information disseminated by the parties when taking decisions to minimize their information costs (Downs 1957: 98–100, 220–30). In other words, rational and reasoning voters do not need to be highly informed in order to make rational choices; with the help of cognitive heuristics, information shortcuts, or cues, they can act as if fully informed. Thus, according to this argument, low information levels do not necessarily have major adverse consequences for the functioning of democracy (Lupia and McCubbins 1998).

Causation and causality concepts should be used with caution, since aspects of causality have been intensively debated by philosophers for millennia, and in contexts such as this the concepts should clearly be distinguished from any more comprehensive philosophical notion of causality. In addition, mere correlation is not sufficient to establish a causal connection between events A and B. One could argue that the minimum requirements for this are: a firm statistical correlation, a solid theory or well-founded reason to believe that A causes B, and justifiable reasons for excluding alternative hypotheses. However, when dealing with time-series of data it can always be argued that if event A precedes event B, it is possible that A has caused B (but not vice versa). Previous events may affect the present. The future, however, cannot affect the past. For an extensive discussion of the concept of causality in social science, see Pearl (2000).

More details regarding the data used, full specifications of all estimated models and a more comprehensive methodological discussion (including information about unit root tests, heteroskedasticity tests and the selection of optimal lag length for the Granger causality tests) are available in the ‘Methodological Appendix’, which can be downloaded from: http://www.pol.umu.se/papers/JEPP2008_appendix.pdf

In addition, I evaluated an alternative strategy for selecting the lag length involving initially testing a long lag and incrementally reducing it, testing the results each length yielded. This approach indicated that the results are generally robust across different lag lengths.

Alternatively, the practical significance, or to be more precise the effect size, of voter opinions may be too small to be statistically significant in relation to the sample size. In other words, even if voters are able to influence the positioning of parties with regard to EU policies, their influence is minor.

I also examined the possibility that the relationship between voters' and parties' opinions may have changed over time (e.g. pre- and post-Maastricht Treaty). When running period-specific estimations this yielded rather ambiguous results, depending on the estimation technique used (not shown here). With voter opinion aggregated at the national level using within-group estimation I found some support for the hypothesis that parties adapt to voter opinions, but not with other estimation techniques used in this study, or when using voter opinion aggregated at the party level.