Home > Reports > September 2017 Archives > Why the U.S. Presidential Election Polls “Missed” the Outcome


September 2017

Why the U.S. Presidential Election Polls “Missed” the Outcome

From an AAPOR Report

Miki Masaki / Akihiko Otaki

The 2016 U.S. presidential election ended with Donald Trump’s victory, beating the dominating predictions of pre-election polls—“Hillary Clinton is leading.” Amid the confidence in public opinion polls wavering, the 2017 Annual Conference of the American Association for Public Opinion Research (AAPOR) was held, where researchers and pollsters involved in the election polls reviewed the polling and the outcome, and a committee formed by AAPOR reported its assessment of the reason why the polls had failed to predict Trump’s victory. The committee has proposed solutions to enhance the accuracy of state-level polls, but it is hard to put them into practice because it is necessary to secure considerable expenses and experienced experts. In the U.S. presidential election system, a candidate receiving a majority of votes by “electors,” who are chosen in each state, wins, not the one who gains the largest number of votes in the national popular vote. To predict the winner, two types of pre-election polls are used: national polls surveying the trend across the nation and state-level polls surveying the trend in each state. The committee pointed out that some state-level polls in the 2016 election had low accuracy, which contributed to the failure in predicting the outcome. The committee’s evaluation focused on the reason why the support for Trump had been underestimated, and several reasons for this, including the following, were suggested: (1) the polls failed to capture the unusual change in voters’ final decision in some swing states, where more than half of late-deciding voters voted for Trump, (2) many polls did not adjust their weights to reflect the voting trend of residents with lower education levels, many of who had shifted their support from the Democratic Party to the Republican Party, (3) impacts of changes in voters’ racial composition, such as a fall in African-American voters’ turnout, may have not been reflected in the analyses of polls. On the other hand, the so-called “Shy Trump” effect, however, has yielded no evidence to support the theory.

The NHK Monthly Report on Broadcast Research


Copyright NHK (Japan Broadcasting Corporation) All rights reserved
Any unauthorized duplication and/or reprinting of data on this Homepage is prohibited.