Poll Fretting

It’s around that time in the election sweepstakes where I fret about the polls. One question keeps running through my mind:

How accurate are they?

In 2016 and 2020, the polls underestimated Trump’s support by a significant margin. Perhaps to compensate, some predicted a Republican “red wave” that did not come to pass in the 2022 midterms.

For myself and others feeling the poll angst, it’s helpful to remind ourselves of the mammoth task pollsters face regarding presidential elections.

Usually, polls consist of 500–1000 individuals who are supposed to predict an election in which as many as 150 million Americans vote. Think about that. One thousand people or fewer are trying to predict the behavior of millions.

Sometimes polls consist of 2000–3000 people but that’s for longer studies and the length of time needed to carry them out is not always practical for the time crunch of a presidential election.

Pollsters have to carefully select individuals in proportions that correlate to the ratios of the nation as a whole. They need to avoid over representing a particular group, such as blue collar manufacturing workers, while not underrepresenting another, such as Catholics and religious leaning voters. At the same time, just because someone is a blue collar worker or a religious person does not guarantee they will vote a certain way. There are so many tendencies and probabilities to factor in.

Plus, consider the fabled “October Surprise”- an event in October that can affect the outcome of a November election. The FBI’s reopening of the Hillary Clinton email investigation 11 days before the 2016 election is an example. The leaking of Mitt Romney’s 47% of the U.S. population are “takers” comment in 2012 is another.

Despite this intense juggling, pollsters produce polls that have margin of error of 3–4 points, typically.

Rather than expecting a perfect prediction, think of polls are like a weather forecast. We definitely would rather have polls and forecasts, but both are not 100% accurate. They are educated guesses as people and the weather operate in ways that defy the experts.

Let’s return to 2016 to emphasize the point. The polls were pretty accurate regarding the population vote. Trump lost it by almost 3 million or 2%. The overall average was about a 4 point advantage for Clinton, which was within the margin of error. Through the Electoral College and 77, 744 votes in three swing states, Trump won. It’s a tall task to for pollsters to predict that someone can lose the popular vote yet win the election. In 40 some presidential elections, that has happened only 5 times. Of those five, the 2000 presidential election is a case unto itself with its complicated legal maneuverings.

For 2020, Pew Research pointed out that a poll of 1000 that gave Biden a 12 point average could shift to a 4 point Biden advantage, what actually happened, by switching only 38 votes. The fact that so few can change the outcome so much is staggering.

Pollsters already have to transpose the views of hundreds or a few thousand to millions. Then add in an October Surprise and other election craziness and it’s a miracle they get as close as they do.

My gut tells me that pollsters are doing a better job of accounting for Trump’s true level of support. Their modifications, the multiple assassination attempts against Trump and Trump’s legal battles lead me to believe that the “shy Trump voter” of 2016 is being tracked accurately and people are more likely to publicly claim their support for him.

At any rate, the moral of the story: the polls are a guideline, not an absolute. Get out and vote because voters determine elections.

Not polls.