4.3 Empirics
4.3.4 Robustness checks
I checked the robustness of the results when using an alternative model specification, a different measurement for policy platforms and inconsistency. Furthermore, I checked robustness for potential unobserved confounders.
First of all, as interaction effect models are the common way to specify mediating effects, I also run the analysis using interaction effects models. The results are ba-sically unchanged: The marginal effect of policy-distance decreases with increasing inconsistency. A detailed description of the results can be found in Appendix A.5.1.
Second of all, I rely on a second measurement of policy-distance, using responses on a seven-point liberal-conservative scale. From this I calculate the quadratic distance of a respondents self-placement to perceived candidate platform on the same scale, as well as the distance to the average placement of the two candidates. For the decreased weight specification, as well as, an interaction effect model, I find similar results for the distance to the average candidates platforms: Inconsistency moderates the effect of policy-distance. For the perceived platforms, I find no statistically significant effect. This does not directly allow to conclude that policy-distance in this case is not moderated, but hints in direction that employing mean platforms reveals a different pattern. One
4.4. SUMMARY
apparent reason for the result when using perceived platforms is perception biases.
Respondents who I classify to posses inconsistent policy preferences, still place the candidate they intend to vote for closer to their self-placement. Perception bias is a well-known problem when using self-placements and perceived positions in the specification of policy-based voting rules (Grynaviski, 2006). The results are reported in Appendix A.5.2.
Thirdly, I run the analysis using a different measurement of inconsistency. An alternative implication of the policy belief model (in combination with randomness of survey response) is that respondents with wider policy beliefs (more inconsistent) have a higher variation in self-placements over different panel waves. Therefore, as a second measurement of inconsistency, I employ the standard deviation of self-placements in different panel waves. Appendix A.5.3 reports the results from interaction effect models.
For all combinations of the different policy-distance measurements, the interaction effects are negative, indicating decreasing marginal effects. However, only for the self-placements the interaction effects are statistically significant.
Finally, it might be that the results are due to unobserved confounders. To minimize this possibility, I re-run the analysis on five three-quarter subsets of the sample, combing the estimates of inconsistency using methods from multiple imputation (King, 2002).
The results do not change over the different subset (see Appendix A.5.4)
Taken everything together the evidence presented in this analysis supports the expectation from the learning perspective. Voters with inconsistent policy preference, put less weight on policy-distance.
4.4 Summary
Citizens with inconsistent policy preference rely less on a candidate’s policy platforms, than citizens with well-defined policy preference. This chapter offers a theoretical explanation for this hypothesis. The way that citizens form expected utility from political programs depends on each citizen’s own policy beliefs regarding what political outcomes they would actually like to see. Because these beliefs not only vary direction (some think that a liberal platform is better than a conservative one), but also differ in certainty about the ideal policy platform. Candidates’ communicated platforms are informative to a different degree to citizens. In seeking a representative match, citizens place different emphasis on policy considerations.
While many previous empirical studies tested a related political sophistication hypoth-esis about spatial voting decisions (Jessee, 2010; Lavine and Gschwend, 2007; Wyckoff, 1980; Knight, 1985), in which less informed or aware voters rely less on policy-distance, this chapter is the first one to put these suppositions on a solid theoretical basis. I am able to show that these kinds of argument cannot be directly derived from standard ex-pected utility models regarding voting choice. The argumentation highlights two things.
First, the ongoing interaction between candidates and voters and how voters learn what they can expect from different candidates, is the key to understanding heterogeneity in a voter’s decision calculus. Second, standard measurements of information are only an indirect approximation of the factors that condition voting decisions. Although political information might matter for the structure of preference, more important are the aspects about policy beliefs themselves.
All in all, this chapter provides empirical evidence for the importance of a broader conceptualization of policy preference. Building a theoretical argumentation on policy beliefs led to a hypothesis that a standard depiction of policy preference would not have been able to devise. This theoretical perspective is better suited in explaining patterns as to how voters’ choose representatives based on their policy views. The next chapter will build on the vital role of this perspective, by examining how it can be used to explain the effect of arguments on voting decisions.
4.4. SUMMARY
Do arguments matter? The effect of political arguments on voting decisions
5.1 Introduction
Theories of representative democracy reserve different roles for political arguments.
Deliberative and discourse theories praise the exchange of arguments as the building block of a democratic society (Ackerman and Fishkin, 2004; Chambers, 2003; Habermas, 1994; Cohen, 1997). They are seen as the essential mean by which citizens are able to find out which policies are best for them and the society. Liberal conceptions are in general more skeptical about the importance of political arguments. They rather highlight the voting act as the crucial mechanism of assuring representation in democracy (Key, 1966; Dahl, 1972). From this perspective political arguments are dispensable for understanding the political process: The starting point are citizens’
policy preferences and their election of representatives based thereon.
A similar dichotomy of viewpoints has developed over the past 60 years of empirical and theoretical research on the topic. Spatial voting theories neglect aspects of prefer-ence formation and analyze in what way citizens’ policy views affect voting decisions (Downs, 1957; Davis, Hinich and Ordeshook, 1970; Enelow and Hinich, 1984). The question of how opinions are formed and what role political arguments play in this formation process are treated as exogenous. By contrast, research in political psychol-ogy and public opinion is interested in the effect of political arguments on preference formation (Lazarsfeld, Berelson and Gaudet, 1948; Zaller, 1992; Taber and Lodge, 2006;
Lodge, McGraw and Stroh, 1989). In many of these models, citizens evaluate political arguments and learn which policy platforms they actually prefer1. However, these models generally pay less attention to vote choice. This shows that the two perspectives
1E.g. what is the effect of elite discussions about policy issues on public opinion (Bartels, 1993; Zaller, 1996; Nelson, Clawson and Oxley, 1997)? Or which political arguments are persuasive and which are not (Cobb and Kuklinski, 1997; Druckman, 2004; Jerit, 2009; Arceneaux, 2012)?
5.1. INTRODUCTION
focus on different but connected aspects: While one side studies the effect of exogenous political preferences on vote choice, the other examines how political preferences are formed by means of political arguments.
Integrating these two perspectives can enhance our understanding of electoral democ-racy in manifold ways. While spatial models have a competitive advantage in generating hypotheses about how citizens’ policy preferences affect voting decisions (see e.g Kedar, 2005) and how candidates position themselves to gain citzins’ votes (see e.g. Adams, Merrill and Grofman, 2005), theories of preference formation prevail in explaining where political preferences originate from (Zaller, 1996; Nelson, Clawson and Ox-ley, 1997), how they change (Mutz, Brody and Sniderman, 1996; Bartels, 1993) and are framed, shaped and persuaded (Druckman, 2004; Jerit, 2009; Arceneaux, 2012).
The missing link between the two perspectives is a theoretical framework that allows political arguments to shape policy beliefs and thereby affect voting decisions.
The theoretical model of this dissertation offers an extension to the spatial voting theory that allows to study the effect of political arguments on voting decisions. In this chapter, I show that this is important for the studying of electoral behavior as one argument can directly influence a voter’s decision. Putting expectations from this theory to a test using an experimental study, I am able to show that arguments about a policy proposal (e.g. whether gay marriage should be legalized) changes the state in which a voter is otherwise indifferent. While in the classic model a voter is indifferent between two equidistant policy platforms, in the model I present, this state changes after evaluating an argument. Voters learn: If a voter finds an argument about a specific policy proposal persuasive, she will prefer the platform that is more likely to be in favor of this policy. Therefore, the outlined model forges the link between findings in political psychology and the spatial analysis of the electoral process.