Abstract
Political science has strong theoretical expectations about the conditions in which people vote “correctly,” that is, in a way consistent with their true preferences and all information known, and the extent to which they display what Converse (1964) defined as “constraint,” i.e., the ability of people to make choices that fit together in a clear ideological way. Yet, outside of surveys, we have little evidence about how people actually behave in the voting booth, because voting is a private act.
State election audit systems that provide electronic records of each cast vote (CVRs) offer a novel way to overcome this limitation. We exploit the availability of CVRs in South Carolina and San Francisco, California to conduct novel tests of correct voting and ideological constraint among the mass public. In both jurisdictions, we merge available cfscores to candidates on the ballot in a given election cycle. We then utilize several million individual cast votes in each election to infer voters’ partisan preferences, and then to determine the extent to which people vote for the “correct” (or most ideologically proximate) candidate in non-partisan races (where there is no party signal on the ballot).
We further employ the San Francisco data to study the extent to which voters show constraint. California elections are ideal for this effort because of the large number of ballot initiatives that voters must decide on in any given election. By coding the voter’s partisanship and capturing their choices on initiatives, we can gain unprecedented insight into voting behavior. Our paper will result in substantially improved insight into both whether people vote correctly in “real” elections, and whether they can reach constrained decisions on ballot questions, absent strong partisan signals.