November 8, 2023

This week’s election should be a wake-up call for everyone who thinks polls are valid predictors. The best we can hope for from polls, especially a full year before a presidential election, is a sense of the current trends, whether momentum seems to be shifting. But none of the major polls taken over the last few months accomplished even that.

By any objective measure, the 2023 election was both a big win for Democrats and a big loss for politicians who tied themselves to Donald Trump’s rhetoric. If you want evidence, consider two governors who are being watched as rising stars and future presidential candidates.

Republican Glenn Youngkin of Virginia had confidently staked his future and reputation as a leader on his promise to sign a law outlawing most abortions after fifteen weeks if voters delivered a Republican trifecta (control of the Assembly, Senate, and Governorship.) Before the election, Democrats controlled the state Senate by a slim margin, and Republicans controlled the House, but voters gave Democrats victories in both chambers. Polls had suggested that Youngkin’s momentum as a generally pro-Trump leader was on the rise, but support for abortion rights killed it. Ever since the Dobbs decision that reversed Roe v Wade, voters in states that put abortion rights on their ballot unanimously protected them in state laws and Constitutions. Still, pollsters didn’t seem to get the correlation between the positive momentum on that issue and its effect on statewide elections.

The vote to protect abortion rights in Ohio wasn’t shocking but the margin of victory in a state that has become continuously redder over recent decades was larger than anyone expected. Another serious momentum shift that polls should have detected but didn’t.

Then there’s Andy Beshear, the Democratic Governor of Kentucky. In 2020, Trump won Kentucky by a two-to-one margin. In 2023, Beshear ran against a powerful Republican coalition that included Mitch McConnell and Rand Paul. Yet, his margin of victory this week was twelve times larger than his first term win in 2019. The increased margin of victory is attributed to support for abortion rights and the money pouring into Kentucky from the infrastructure bill promoted and signed by President Biden last year. The polls failed to predict either the degree to which abortion lifted Beshear in one of the reddest states in the country or the general rise in his popularity. Finally, there’s the Mississippi Governor’s race. Republican incumbent Tate Reeves beat Elvis Presley’s cousin Brandon, but the margin was smaller than both Reeves’ victory in 2019 and Trump’s win over Biden in 2020.

By any measure, these results imply a rising Democratic tide, which seems to contradict the national polls that suggest Trump and Biden are running about even in the 2024 election. That’s a very serious matter, because voters are victims of our polarized media. When poll after poll suggests that the country is tending to Trump, most people believe them. But the great majority of people who unquestioningly swallow poll results really have no idea of how polls work or whether they’re likely to be accurate, and the major media outlets do little or nothing to educate their viewers. Instead, poll reports are presented like sporting events. The jargon used sounds like an account of a horse race because that builds ratings.

My years of experience with statistical sampling and prediction models leave me skeptical about the way polls are conducted today. We’re told almost daily that Trump is a shoe-in for the Republican nomination and that his popularity continues to grow in spite being on trial for ninety-one felonies, including those stemming from the January 6th insurrection at the Capitol and the obvious phenomenon that American women are furious over the Dobbs decision, which they properly, closely associate with Trump. Are the polls correct?

I ask whether the thousand or so people sampled in a typical poll are representative of America’s electorate. Who are they? Think about how we use our phones and email accounts. They are the sources of polling data. Do you answer calls from unknown callers? Do you respond to polling emails knowing that nearly all of them are just bait for fund-raising? Most of us answer “No” to both questions these days. If that’s true, how can polling results accurately reflect the view of the electorate as a whole. The notion violates every rule of statistical sampling.

This week’s election should give every voter and pollster pause. It’s a serious problem, but one that’s easily fixable, at least in part. Polling organizations are generally professional and nonpartisan, but the media they work for make no effort to provide meaningful background for voters. They all need to put political biases aside and do their jobs, informing the public accurately, and making sure we all understand that relying on polls can be dangerous.

Dewey did not beat Truman in 1948, no matter what the polls said.