Ground Truth Validation of Survey Estimates of Split-Ticket Voting with Cast Vote Records Data
Alexander Agadjanian & Jonathan Robinson | August 26, 2019
From signaling trends in nationalization and partisanship to clarifying preferences for divided government, split-ticket voting has received copious attention in political science. Important insights often rely on survey data, as they do among practitioners searching for persuadable voters. Yet it is unknown whether surveys accurately capture this behavior. We take advantage of a novel source of data to validate survey-based estimates of split-ticket voting. Cast vote records in South Carolina (2010-18) and Maryland (2016-18) provide anonymized individual level choices in all races on the ballot for every voter in each election, serving as the ground truth. We collect an array of public and private survey data to execute the comparison and calculate survey error. Despite expectations about partisan consistency pressures leading to survey underestimates, we find that surveys generally come close to the true split-ticket voting rates in our set of races. Accuracy varies, but notably is more consistent for split-ticket voting in a given dyad of national races (e.g., President vs. U.S. House) than in one with state races, as the former is often of greater interest in research and practice.
Prepared for presentation at the Annual Meeting of the American Political Science Association, Washington, DC, August 29 – September 1, 2019. An earlier version of this paper was presented at the 2019 American Association for Public Opinion Research Conference (Toronto, Canada) on May 17, 2019. Thanks to Shiro Kuriwaki for making cleaned South Carolina cast vote records available to us, as well as for helpful guidance and discussion.