Keywords:

2024 should have been a good year for pollsters in the U.S. The final national polls correctly indicated a tight race between Harris and Trump. Battleground state polls also predicted a close race, while offering hints of Trump’s upcoming Electoral College victory. And yet, the polling of Trump’s third and final presidential run was, again, not without controversy. The final Selzer poll in Iowa appeared to indicate a monumental last-minute swing to Harris. That single poll attracted unprecedented media attention in the final days of the campaign and fueled speculations about Harris’s decisive breakthrough among women concerned with abortion restrictions. Yet, on election night, instead of losing it by 3 points, Trump won Iowa by 13 points and cruised to victory by winning all seven battleground states.
The Selzer controversy is symptomatic of the enduring problem of the systematic underestimation of Trump by pre-election polls, which now extends to the third consecutive election cycle. In 2024 again, the final national polls in the RealClearPolitics (RCP) average underestimated Trump’s margin over Harris by 1.7 points. Of these 17 national polls, Trump was only ahead in 5. Similarly, Trump’s lead was underestimated, albeit modestly, in all seven battleground states. However, the average signed error by state (based on polls listed in the RCP database) was -4.0, on par with the dismal performance of 2020 and 2016 state polls.
Therefore, despite eight years of experience polling Trump, pollsters have failed to address the systematic polling error. The proposed paper argues that future analysis should reconsider the social-desirability induced Shy-Trump effect, which was dismissed by multiple postmortem efforts after the 2016 and 2020 elections, including two extensive AAPOR reports. The Shy-Trump hypothesis was previously dismissed by these inquiries on questionable grounds both theoretically and empirically.
While recognizing that any evidence for the Shy-Trump hypothesis is necessarily circumstantial, the paper demonstrates that such evidence can be uncovered through secondary analysis of 2016 and 2020 survey datasets, including the American National Election Studies (ANES), and publicly released surveys by the Pew Research Center. The findings suggest that among survey respondents who did not disclose a candidate preference, the number of individuals with pro-Trump attitudinal predispositions was disproportionately high. Additional evidence points to a possible social desirability-induced mechanism behind the pattern.
As 2024 survey datasets by the aforementioned sources become available, the analysis will be extended to a study of the unprecedented systemic polling error that seems to occur every time Trump is on the presidential ticket.