Three Reasons Why Midterm Polls Could Be Wrong

On Nov 22, 2022

Comment

With a week remaining before the midterms, the polls say that Republicans are very likely to win a solid but not huge majority in the House. The Senate remains a toss-up.

For all the attention their errors receive, pollsters really do excellent work and make valuable contributions to democracy. We know a lot more about what our fellow citizens think about policy and politics because of systematic measurement of public opinion; it’s far better than relying on focus groups or (just) shoeleather reporting, even though the latter can add useful insights. And, yes, horse-race polls give us good estimates for the question many of us have: Who will win?

Polling is always subject to some error because of random variation. Indeed, we should take the findings as estimates giving us a range where the truth is probably found, rather than indications about exactly what people are thinking. And as good pollsters never hesitate to remind us, polls are snapshots of a moment in time, not predictions about the future.

But it’s also true that more systematic statistical bias can seep in to election polling, and so the polls sometimes wind up off by larger margins. By statistical bias, I mean effects caused by neutral procedures — such as screening for likely voters — that might inadvertently produce results favorable to one or the other party. Reputable polling outfits, including partisan ones, try hard to get it right because their reputations and therefore their income depend on it, and polling averages will typically exclude firms that appear to be deliberately slanted for one party.

This midterm seems to have more than the usual amount of uncertainty. And it appears that the uncertainty could work in favor of either party. While unlikely, it’s still possible that the best predictions are wrong, and we’re really heading for a major Republican landslide — or a break-even election that could even leave Democratic majorities in both chambers of Congress.(1)

Here are three reasons the polls could be off.

First, normal polling error could be more of a problem in this election because there are simply fewer polls in 2022 than there used to be. The less polling you have, the more unreliable the estimate will be.

Some high-profile races have had quite a few surveys; For the Georgia Senate contest between Senator Raphael Warnock and former football star Herschel Walker, RealClearPolitics has collected 11 polls taken entirely in October. But in the North Carolina Senate election, also hotly contested but with a lot less star power and national attention, there have been only four October polls. Fewer polls heightens the chance that the estimates will be off.

Second, it is harder to do polling these days. Between the demise of landline phones and the ever-shrinking share of voters willing to talk to pollsters, the old way of doing things that persisted for some 50 years is pretty much gone.

Public opinion professionals have done a great job finding innovative ways to reach voters on cellphones and online, but with methodology forced to change rapidly from cycle to cycle, it’s impossible to have results as reliable as they were in the heyday of stable polling. What’s more, it’s possible that this year’s mix of new and continuing methods could wind up systematically missing or overrepresenting some groups of voters, which could add up to an across-the-board statistical bias for one party or the other.(2)

A final challenge is the change in who votes and how. In the last several elections, we’ve had constant change in voting rules. Some states have made voting easier with automatic voter registration, relaxed rules for absentee voting and extended early voting. Other states have made it more difficult by reversing those policies. Republican voters, who until 2020 were more likely to use absentee ballots, are now much less likely to vote that way. All of this changes how likely different groups are to vote, but not necessarily in predictable ways.

Similarly, turnout has been unusually high in recent elections, but there is no way to know whether that will continue in 2022 — and that, too, may differ for different groups and different locations. If pollsters only had to figure out which candidates the public preferred, their job would be much easier. But horse-race polls close to the election also incorporate “likely voter” screens that are really just educated guesses.

Unexpected turnout, whether high or low, could produce errors in either direction, leading to estimates that are too friendly to Democrats or to Republicans.

None of these possible sources of polling error reflect deliberate attempts to falsely portray one party as winning. The point is that polling itself adds a lot of uncertainty to what we can know until the election results come in. Right now, the best advice is just to be patient.

More From Other Writers at Bloomberg Opinion:

The Email That Every Campaign Is Sending Right Now: Ramesh Ponnuru

Democrats Were Smart to Meddle in GOP Primaries: Jonathan Bernstein

Seen the Latest Polls? Now Forget Them. Here’s Why: Jonathan Bernstein

(1) I’m focusing on unknown errors that could be in either direction; for why Nate Silver’s projection models find some specific likely errors in various states, see Silver here.

(2) It’s actually more complicated than that: There could be multiple biases, yielding different biases in different states.

This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.

Jonathan Bernstein is a Bloomberg Opinion columnist covering politics and policy. A former professor of political science at the University of Texas at San Antonio and DePauw University, he wrote A Plain Blog About Politics.

More stories like this are available on bloomberg.com/opinion