Pollsters Have Fixed Mistakes From Past Elections, Right? Actually, No.

We all remember the run-up to the 2016 presidential election. Hillary Clinton was ahead in the polls and ahead in the betting odds. She was also ahead in the media forecasts.

Depending on which source you consulted, Hillary Clinton’s odds of winning the 2016 presidential election were between 85% and 93%. Donald Trump was given no chance at all of winning, or practically none. Then Trump won.

The political elites went crazy and the attacks on Trump have never stopped since. But, professional pollsters had another task to attend to.

What went wrong? How could their supposedly scientific statistical methods miss the result by so much? There are a lot of ways for polls to get things wrong.

The skew between Democrats and Republicans can be off relative to the actual voting population. Whether the polling population is based on “all voters,” “registered voters,” or “likely voters” makes a huge difference; (the only really accurate polls involve likely voters, but even that can be tricky depending on your definition of “likely”).

This article touches on all of those issues, but it focuses specifically on the difference between college-educated and non-college educated white voters. Pollsters had assumed that the college and non-college white Republicans voted the same way. This meant that distinguishing between the two was unimportant. That turned out to be a huge mistake.

Non-college educated white Republicans voted overwhelming for Trump whereas college-educated white Republicans still favored Trump but by a smaller margin. The undercounting of non-college educated whites, especially in the upper Midwest, resulted in surprise victories for Trump. The same problem persisted in 2018 (Democrats took back the House but fell short in many races for governor and senator that polls said they would win).

Has the problem been fixed in the 2020 polling? Actually, no. Pollsters recognize the problem, but can’t decide if the college/non-college skew has predictive power.

It could just be a random variation without much consequence. That’s fine; it’s their problem.

For our purposes, it’s fair to infer that past is prelude and Trump may again have hidden sources of strength in key states like Michigan and Florida that are not showing up in the polls. If so, the pollsters may miss the mark again as Trump enjoys four more years in office.

Corporate leaders and institutional fiduciaries looking to incorporate state of the art predictive analytics to their risk mitigation and strategic analysis should click the link to learn more about Raven Predictive Analytics®.