Why Are Surveys Struggling to Estimate Vote Shares?

Polling in the 2020 U.S. presidential elections significantly underestimated Trump support, calling into question the accuracy of all political surveys. Although many have speculated that this bias is due to Trump supporters refusing to respond to surveys, we have previously lacked the data to directly evaluate this theory. This paper leverages the rich and unique set of data collected alongside the 2020 American National Election Studies (ANES) to identify and quantify sources of error using the Total Survey Error framework. We document evidence of nonignorable unit nonresponse; that is, Trump supporters were less likely to respond and less likely to complete the 2020 ANES if they did respond. We are also able to rule out other potential sources of error. Our analyses identify a fundamental threat to political surveys, and careful diagnosis and quantification of the bias is critical for informing corrections and future survey design.