Validating Survey Estimates of Split-Ticket Voting with Ballot Image Log Data

• In Event: Polling, Measurement, and Survey Effects

Abstract

The decline of split-ticket voting in American elections has been a salient point of interest for political observers. In documenting the growing over-time correlation between election outcomes across offices, many often interpret this trend as a sign of increasing nationalization and widening of the scope of partisanship’s influence. Individual-level analysis of this phenomenon, however, has entirely been confined to the use of surveys as a data source. In this research, we evaluate the accuracy of survey estimates of the split ticket voting rate by introducing a novel source of data that serves as a ground truth: ballot image logs.

These images represent actual election ballots that contain anonymized individual vote choice for offices across the entire ballot. Major election surveys ask individuals for recalled or intended vote choice for many of the same offices. In approaching the same questions with these two forms of data, ballot images offer a unique opportunity to produce a true estimate of split-ticket voting and check how well surveys capture this and other vote pattern-related dynamics.

Available ballot image data for this analysis comes from two states: South Carolina and Maryland, with other states potentially available with further data collection. To derive our survey estimates, we focus on state data sources like the CCES and state pollsters with publicly released raw data. Surveys like the CCES will be of particular interest, given its much larger sample sizes and breadth of office vote choice questions (President, Senator, Governor, US House, Attorney General, etc) that will expand where split-ticket voting can be examined.

By leveraging a new data source in ballot images, this research will clarify the accuracy of surveys in capturing split-ticket voting and speak to other important topics such as variation in partisanship’s influence on voting for offices at different levels of the ballot.

Authors

• Alexander Agadjanian, MIT

• Jonathan Robinson, Catalist, LLC