According to a FiveThirtyEight.com analysis, state polls in key battlegrounds like Florida, Michigan, and Pennsylvania predicted Hillary Clinton would win by an average of 0.6, 4.0, and 3.7 points, respectively. Instead, Mr. Trump won Florida by 1.19 points, Michigan by 0.22 points, and Pennsylvania by 0.72 points. While the state level polls missed the mark, national polls were closer to the national popular vote results.
The Washington Post highlighted Fordham University political scientist Costas Panagopoulos’ study on the accuracy of the top 14 national polling outlets. Panagopoulos found that eleven polls (McClatchy/Marist, ABC/Washington Post, CBS News/New York Times, Bloomberg, Economist/YouGov, Fox News, CBS News, NBC News/Wall Street Journal, Reuters/Ipsos, Monmouth, NBC News/SurveyMonkey) overestimated Clinton support in the final week of the election, and three (LA Times/USC Tracking, Gravis, and IBD/TIPP Tracking) overestimated Trump support. McClatchy/Marist was the closest to the popular vote, with a final two-person matchup that had Mrs. Clinton winning by two percentage points. As of this week, Mrs. Clinton had 48.04% of the popular vote and Mr. Trump 45.95% – a Clinton advantage of 2.09 percentage points.
The debates on the most accurate polling methodologies will continue beyond. The differences may account for why some polls were better than others. For example, McClatchy/Marist used only live surveys (live landline and cell phone calls), while one of the least accurate polls in the final week, NBC News/SurveyMonkey, incorporated non-live surveys in their sample. Polling outlets – especially those that focus on state surveys – must discover why their methods failed in 2016 and adjust their methodology and samples in elections to come.