By David Thornton
The resulting study, released by the American Association for Public Opinion Research, “found that the biggest culprit was state-level polling underestimating the level of Trump’s support, most importantly in Pennsylvania, Michigan, and Wisconsin,” observes Business Insider in a classic understatement.
National polls showed Hillary Clinton with an average lead of about three-points, which was very close to the actual popular vote result which Clinton won by two points. The polls were less accurate at the state level, where they showed a tight race, but still pointed to a Clinton victory.
The three perennially blue states in the upper Midwest flipped to Trump and enabled his path to 270 electoral votes. The fundamental question is why pollsters underestimated Trump’s support in these states. On that issue, the analysis found three main factors that likely caused polls to be off in those states.
A major factor was that a large segment of voters waited until the last week before the election to make their decision. In Florida, Pennsylvania and Wisconsin, those late-deciding voters broke decisively for Donald Trump. In Wisconsin, they chose Trump by a 30-point margin. The margin for Trump among these voters was 17 points in Pennsylvania and Florida.
A second major factor was that many polls included too many college graduates in their samples and their assumptions about the electorate. “Voters with higher education levels were more likely to support Clinton,” the report said. “Furthermore, recent studies are clear that people with more formal education are significantly more likely to participate in surveys than those with less education.”
The third factor was that many Trump voters did not admit to preferring Trump in pre-election polling. There had been speculation about reluctance of Trump supporters to admit their preference to pollsters, the so-called “Shy Trump effect,” but the report notes that this effect could also be attributed to late-deciding voters.
Two additional factors were also considered to be less compelling reasons for the polling problems. The report notes that in 2016, turnout grew more in Republican counties than in Democrat counties when compared to 2012. This could have caused an overcounting of Democrat demographics while underestimating Republican support. The report also notes that Donald Trump’s name appeared above Hillary Clinton’s on the actual ballot while polls tended to randomize the order of the candidates. The report considers these effects to be insignificant.
Regarding the pre-election forecasts that Clinton was a shoo-in, the report notes that polling and forecasting are two different things. “Pollsters and astute poll reporters are often careful to describe their findings as a snapshot in time, measuring public opinion when they are fielded,” the report notes. “Forecasting models do something different — they attempt to predict a future event. As the 2016 election proved, that can be a fraught exercise, and the net benefit to the country is unclear.”
Polls are already history when they are published. The measure public opinion on the dates that they are conducted, they do not predict future events. Polling results and trends can be used by forecasters to make predictions, but, if public opinion is changing rapidly, as it did in 2016 with FBI Director Comey’s eleventh hour letter to Congress, then polling is less effective as a forecasting tool.
The analysis also found that there is no consistent partisan bias in recent US polling. While polling underestimated support for Trump last year, the reverse has been true in other recent elections. “Whether the polls tend to miss in the Republican direction or the Democratic direction is tantamount to a coin flip,” said the report.
In 2012, late-breaking support for Barack Obama in the wake of Hurricane Sandy and the Obama bromance with Chris Christie derailed what many thought would be a win for Mitt Romney. In 2004, early exit polling showed that John Kerry would defeat incumbent George W. Bush. When all the votes were counted, Bush won a convincing victory.
Looking back at the 2016 polls, which are still available on Real Clear Politics, the signs were there. Pennsylvania showed a 1.9-point lead for Clinton, which was well within the margin of error of most polls. The latest poll before the election gave Trump a one-point lead. In Michigan, the average showed Clinton up by 3.4 points, still within the margin of error, while the last poll showed Trump with a two-point lead. Wisconsin polling was farther off, showing a 6.5-point lead for Clinton. Trump did not lead in any Wisconsin polls, but won all three states by less than one point on Election Day.
The bottom line for political observers is that polling is not an exact science. This is particularly true in a country that is as closely divided as the United States is today. While polls are useful to give a snapshot of public opinion, they can’t be expected to accurately predict an election winner in a tight race where poll results are within the margin of error.
In retrospect, the fact that polls were as tight as they were going into the final week before the election should have been a red flag for Clinton supporters. For an opponent who was as unpopular as Donald Trump to be within the margin of error of the Democrat candidate in what was almost universally assumed to be a coronation more than an election indicated serious problems with the Clinton campaign.
The surge of Trump voters in the final week may be partially due to the continuing dribbles of scandal from the hacked emails and Director Comey’s decision to reopen the investigation a week before the election, but it also indicates a more fundamental problem for the Democrats. Many of these late Trump voters undoubtedly went for Trump because Democrat messaging failed to convince them that Hillary would improve their own lives and finances.