Individual Submission Summary
Share...

Direct link:

Lexicographic Preferences in Candidate Choice

Sun, October 3, 6:00 to 7:30am PDT (6:00 to 7:30am PDT), TBA

Abstract

The question of which political candidates are elected for office is fundamental to democracy and political science. Whereas there is much agreement that party affiliation is one of the most important candidate characteristics to voters, evidence regarding other attributes such as gender and race of the candidate is much more debated (King and Matland 2003; McGraw 2011, p. 191; Christensen, Rosa, and Grönlund 2020; Ono and Yamada 2020; Schwarz and Coppock 2020). We suggest that voters have lexicographic preferences, which means that they rank their preferences in a hierarchy in which at least one candidate attribute is placed over all others and this attribute is sought satisfied at (nearly) all costs. Party affiliation may be such a first-order preference for voters since it has been found to be of prime importance for candidate choice (Popkin 1991; Rahn 1993). However, some second-order candidate attributes, for instance gender and race, may become more important if there is a tie on the first-order attribute.

This alternative structuring of preferences challenges the usual substitutability-between-alternatives assumption in classical choice theory (Dutter 1981; Drakopoulos 1994; Scott 2002; Rosenberger et al. 2003) as well as most theories on how voters weigh different attributes of candidates. This means, on the one hand, that if party affiliation is a first-order lexicographic preference, a black, female Republican candidate may not make democratic voters swing to the Republican candidate. On the other hand, second-order lexicographic preferences may be more important than recognized when voters choose between candidates from the same party. Such political choices happen not only in primary elections in the US. Voters also choose between candidates from the same party in parliamentary elections with multiple candidates from each party, and they are forced to focus on other attributes in non-partisan elections as well (Kirkland and Coppock 2018). Lexicographic preferences may therefore explain some of the mixed results regarding second-order attributes like gender and race in the literature (Matland and King 2002; McGraw 2011; Schwarz and Coppock 2020). We propose a unified theoretical framework that is capable of explaining these results.

In the presence of lexicographic partisan preferences, analyzing voter preferences without distinguishing between situations where candidates are from the same party and from different parties, generates an estimate of average marginal causal effects that may not be very informative. This is because the average depends on the relative weight of situations with and without party variation (Hainmueller, Hopkins, and Yamamoto 2014). Yet, this is how voter preferences have hitherto been analyzed in the vast literature using conjoint experiments to study candidate choice.

We show how the conjoint method can be used to test for lexicographic preferences. In randomized conjoint experiments, respondents are often asked to choose between two candidates from the same party (about half the cases with two parties and two candidates to choose from (Hainmueller, Hopkins, and Yamamoto 2014). Whereas many studies have examined interactions between conjoint attributes, no one has to our knowledge examined whether variation vs. tie on one dominating attribute affects the importance of other attributes. Within this conjoint setup, we propose a simple test of lexicographic preferences, which can be carried out in all randomized conjoint experiments: We split the data in choice scenarios in which there is a tie on the proposed first-order lexicographic preference (in our case party affiliation) and choice scenarios where there is variation on the first-order preference. Then we test for differences in the marginal causal effect of the proposed second-order preferences in the two types of scenarios.

To examine our framework empirically, we first reanalyze conjoint data from Kirkland and Coppock (2018) and find evidence that gender and race are indeed second-order attributes. The effect of gender increases when there is a tie on the lexicographic first-order attribute, party affiliation. Furthermore, when comparing Democratic and Republican respondents, we see that Republican voters tend to prefer white to black candidates when the candidates are from the same party. At the same time, the second-order effect of gender is primarily driven by democratic respondents who tend to prefer female from male candidates. These results suggest that in the US, gender and race are second-order, ideological attributes whose effects differ more between democratic and republican voters when there is a tie on party affiliation. To demonstrate that these are not just chance findings, we pre-registered and tested the hypothesis in a new data collection using Amazon's Mechanical Turk. Results confirm the hypothesis.

Authors