Toward a Unified Model of Turnout and Vote Choice

• In Event: (Compulsory) Civic Duty and Turnout

Abstract

Due to Donald Trump’s polarizing effect on the electorate, American presidential voter turnout jumped dramatically from 58.6% in 2012 to 66.7% in 2016 (McDonald 2020). Franklin Roosevelt, Huey Long, and Adolf Hitler also raised turnout and rode it to victory, vividly demonstrating what scholars have known for a century: Charismatic and divisive leaders who can make the citizenry care about an election outcome mobilize voters to the polls. Thus, turnout and vote choice are endogenously related.

In practice, however, turnout and choice are nearly always analyzed separately. At most, abstention is treated as just another vote choice in a multivariate logit or probit specification with uncorrelated errors. This may work fairly well much of the time, but it makes no sense theoretically, and it is particularly likely to perform poorly when one party is much more mobilized than the other, as in the 2016 U.S. election. In such cases, turnout rises disproportionately among the supporters of one party and may drop among partisans of other parties, a violation of the statistical model. Allowing for correlated errors is an ad hoc fix, but how well it works is not known.

Second, when survey data are available, “caring about the outcome” has long been employed as an explanatory factor for turnout, a tradition that dates to The American Voter. But how exactly does caring relate to the intensity of candidate preferences in a typical electorate? For example, when African-American turnout rose during the candidacies of Barack Obama, was their enthusiasm picked up by the “caring” variable? What about the case of Trump’s white non-college voters? To the best of our knowledge, no one knows. And when the available data are from election returns or voter files, attitudinal variables are unmeasured, and so the intensity of voter concern is unknown.

Third, civic duty is widely recognized as the theoretical key to why anyone votes at all, and it has a gigantic effect, as researchers have known since The American Voter. But many election surveys omit it entirely, and it plays no role in many turnout studies even when it is available. How should the ongoing project about civic duty carried out by Andre’ Blais and his coauthors be extended to incorporate the concerns of the previous two paragraphs?

Finally, researching turnout has grown increasingly difficult as survey refusal rates have skyrocketed. Respondents have always over-reported turnout, but now the selection effect of who responds is an even larger problem. For example, the reported turnout rate in the sophisticated 2016 American CCES exceeds 90%, no matter which of the supplied survey weights is used. Overestimation of turnout is rampant in recent surveys in new and old democracies alike, making it an almost impossible task to provide a serious account of contemporary voters. In response, researchers and practitioners have increasingly turned to official election returns and voter files to get validated turnout data, but then the theoretically-relevant attitudes go unmeasured, as do many demographic factors. Scholars have been tempted to use imputation to estimate the unmeasured variables, but experience has been quite mixed. And of course, in most countries the voter files are not public information available to researchers.

In sum, the study of turnout has come upon hard times, both theoretically and empirically. In this paper, we take two steps. First, we propose an integrated theoretical model of turnout and vote choice that incorporates Riker and Ordeshook’s widely cited decision model for turnout and subsequent developments in the rational choice tradition. We extend that framework to cope with the endogeneity of turnout and the recent developments in the non-additive character of civic duty. Then we study countries in both the West and the East, showing how to use available data from different sources in those countries to estimate this unified model.

Authors

• Won-ho Park, Seoul National University

• Christopher H. Achen, Princeton University