Author Summary by Mert Moral and Andrei Zhirnov
Do all voters consider same parties when they make their vote choices? And if not, how does such variation in their “choice sets” matter for the electoral process?
Although these questions are of crucial importance for our understanding of voting behavior, spatial competition, and democratic representation, previous research on issue voting –the empirical investigation of how voters respond to party policies in distinct issue domains– rarely tackles with them and, instead, assumes a common and fixed choice set composition.
We suggest that voters’ effective choice sets vary widely, and many of them are different from the ones imposed by researchers. Some voters may disregard certain parties. Others may take into consideration larger numbers of party alternatives when making their vote decision. If this is the case, conventional empirical models of issue voting might produce biased estimates of the effects of voters’ policy considerations on their vote choice.
To address this problem, we present the so-called “Constrained Choice Conditional Logistic Regression” (CCCL) — a random utility model that builds on the conditional logistic regression and the constrained multinomial logistic regression. CCCL treats the observed vote choice as a product of two interrelated but distinct processes –the formation of individuals’ choice sets and the choice among the parties in such choice sets. We assume that party identification, and parties’ electoral viability and policy extremity determine choice set formation. Along with the prominent spatial theories of issue voting, we inform our theoretical expectations from the literatures on strategic voting and party identification.
The main text presents an application of the CCCL, as well as competing models, to the data from the 1989 parliamentary election in Norway (in online appendices we present several additional cases). The CCCL has a better fit to the data than the conditional logistic regressions even when the latter includes the same set of non-spatial covariates (CCCL corrects 20 to 25% of its wrong predictions).
The examination of parameter estimates confirms that voters do consider different sets of parties in their choice sets. It also shows that non-spatial factors do not simply compete for influence with policy considerations. Rather, they shape individuals’ choice sets, and thus condition the effects of their policy considerations on vote choice. Under certain conditions, these factors may render the policy positions of political parties irrelevant for the vote choice of particular voters and political outcomes in general. This result also suggests that, as parties choose their policies, they do not always need to pay attention to all voters’ policy preferences but only to those whose choice sets include them.
Our results offer a number of methodological and theoretical takeaways. First, one should not simply ignore the variation in voters’ choice set compositions. Second, arbitrary exclusion of parties from empirical analyses of voting, as often done in previous literature, may harm the validity of our inferences regarding the weights of particular policies in individuals’ voting calculus and parties’ position-taking incentives. Finally, the use of the CCCL model is not limited to the literature on voting behavior. Along with its flexibility to employ chooser-, choice-, and chooser-choice dyad-level variables, this constrained decision-making model, which we present in our forthcoming article in the American Journal of Political Science, can be of use to researchers who seek to separate the process of choice set-formation from the choice process.
About the Authors: Mert Moral is an Assistant Professor of Political Science at Sabancı University and Andrei Zhirnov is a Ph.D. candidate at Binghamton University. Their article, “Issue Voting as a Constrained Choice Problem” is now available in Volume 62, Issue 2 of the Amerian Journal of Political Science.
Speak Your Mind