What do election polls tell us?Contributor Meera Pandit
Over the last four years, the frequent refrain about election polling is the polls were so wrong in 2016. But just how far off were the polls?
In 2016, the final RealClearPolitics poll showed Clinton with a 3.2% lead. She won the popular vote by 2.1%, only 1.1%-points less than polls suggested, as shown in the chart below. This difference was more than we saw in 2004 and 2008, but less than both 2000 and 2012. Part of the dissonance is likely not caused by the inaccuracy of polls, but rather the disconnect between the popular vote margin and the winner-take-all Electoral College, which delivered President Trump a solid 77-vote margin of victory, despite losing the popular vote. Further undermining the reputation of the polls is the conflating of polls and probabilities. The vast majority of political commentators and news outlets predicted a high probability of a Clinton victory.
Of course, polls are by no means perfect. They are a statistical tool, which means they inherently have errors. Even averaging over the three weeks before the election, the weighted-average absolute error of national presidential polls since 1972 has been a substantial 4.1% points1. Polls also are subject to selection bias, in which some individuals are more likely to be polled than others. One notable example of both selection bias and incorrect statistical weighting of the samples was the underrepresentation of voters without a college education, who predominantly favored President Trump in 2016. The polls also cannot account for every factor, and in 2016, the president was aided by late-deciding voters. That being said, in 2020, surveys suggest that as few as 6% of voters are still undecided2, 14.5 million people have already voted3, and there are likely to be fewer “silent” Trump voters.
Not only do we not yet know the outcome of the election, but we also don’t know how markets might react to that outcome. Therefore, investors should focus on what is known—low interest rates, improving earnings and a new economic cycle—rather than the polls, to guide their investment decisions.
Popular vote spread minus final poll spread