Social Desirability Bias and Polling Errors in the 2016 Presidential Election

Abstract

Social scientists have observed that socially desirable responding (SDR) often biases unincentivized surveys. During the 2016 presidential campaign, we conducted three list experiments to test the effect SDR has on polls of agreement with presidential candidates. We elicit a subject's agreement with either Hillary Clinton or Donald Trump using explicit questioning or an implicit elicitation that allows subjects to conceal their individual responses. We find evidence that explicit polling overstates agreement with Clinton relative to Trump. Dividing subjects by party identification, we find that SDR significantly diminishes explicit statements of agreement with the opposing party's candidate. Democrats are significantly less likely to explicitly state agreement with Trump. This threatens the predictive validity of polling, negatively impacts the ability of markets to accurately price assets, and exaggerates disagreements between Democrats and Republicans. We measure economic policy preferences and find no evidence that ideological agreement drives SDR. We find suggestive evidence that SDR correlates with county-level voting patterns.