The Washington PostDemocracy Dies in Darkness

Opinion Polls are useful. They just can’t predict elections in swing states.

President Biden's popularity has improved substantially from his lowest point this summer, according to polls. (Andrew Harnik/AP)

Polls now suggest Democrats have a good chance of keeping control of the U.S. Senate and a smaller-but-real chance in the House, after it seemed for much of President Biden’s tenure that the party was destined to have a bad midterm cycle. But polls also wrongly suggested Hillary Clinton would win the 2016 presidential election, Biden would carry some swing states by substantial margins and Democrats would win a 52- or 53-seat Senate majority in 2020.

So there is a lot of discussion about whether the polls are again missing Republican voters and overstating the Democrats’ chances this year.

I am not sure whether the polls will correctly project the winner of either chamber. But this is the wrong conversation about polling in the first place — or at least it’s wrong for this to be essentially the only conversation about polling.

The real problem isn’t the polls. It’s that journalists, people in politics and even regular voters don’t use polls correctly. Polls are valuable, but not if they are used as the political version of Vegas odds for the Super Bowl.

Follow Perry Bacon Jr.'s opinionsFollow

So here’s a guide to three ways we shouldn’t use polls and four ways we should. Let’s start with the misuses of polling:

Polls should not be used to confidently predict the winner in statewide races in swing states. RealClearPolitics, which averages polls from various outlets, correctly projected the winner of 48 of the 50 states in the 2020 presidential election and that Biden would win overall. (The average wrongly suggested that Trump would win Georgia and Biden would win Florida.) The Economist and FiveThirtyEight, which do so-called forecasts that rely on polling, also missed only two states and projected a Biden victory. (They each had Florida and North Carolina for Biden.)

Despite that fairly good showing, we should be cautious about polling, polling aggregation and modeling in swing states. We should not think of these sites as being right 96 percent of the time. Anyone who follows politics closely could have predicted the winner of at least 40 states. Missing 20 percent of the swing states isn’t bad, but it’s not impressive, either. And in some key states (Ohio and Wisconsin in particular) where the 2020 winner was projected correctly, the polls were still way off.

Virtually every poll, forecast and election expert wrongly projected Clinton to win four years earlier, all because they missed a few swing states. Much of the 2020 polling suggested Democrats would win the U.S. Senate elections in Maine and North Carolina, thereby whiffing on more than a quarter of the hardest-to-predict Senate races.

I simply do not think polls can tell us with much certainty whether Sen. Raphael G. Warnock (D-Ga.) will survive his challenge from Republican Herschel Walker; nor can they predict other races in states where the parties are closely divided. And because Senate control in particular is likely to come down to these few races, polls can’t offer much guidance about which party will win the chamber, one of the core questions of the November elections.

In short, the most interesting, talked-about data we get from polls — their findings in close races — simply isn’t that useful or reliable.

This is a huge problem because most people, myself included, find it hard to accept “we just don’t know who’s ahead in this race” when a stream of polls are constantly released that give us the impression that we know who’s really ahead.

Polls should not be used to determine what policies elected officials and activists push for. “Popular” is not a synonym for right, smart or morally correct. The 1963 March on Washington was opposed by a plurality of Americans at the time. The Rev. Martin Luther King Jr. was an unpopular figure at his death. I worry that if King lived today, he would have to read news stories by political analysts about the unpopularity of protests he was leading, with the implication that those protests either shouldn’t happen or should use only poll-tested, popular language.

The Republican lawmakers who are now passing strict bans on abortion are likely very aware of how unpopular their approach is. They believe these bans are justified anyway. Biden’s decision to consider only Black women to replace retiring U.S. Supreme Court Justice Stephen G. Breyer didn’t initially poll well but had other important rationales.

It is perhaps electorally unwise for politicians in swing states to push unpopular ideas, but most elected officials and activists aren’t politicians in swing states.

Also, public opinion isn’t that set anyway …

Polls should not be used to make specific claims about public opinion on issues Americans aren’t that interested or engaged in. The broader American public is just not very informed about politics and government. Polls asking Americans to, say, rank issues they care about most, choose their preferred policy among several options or pick their favorite candidate among people from the same party, often aren’t capturing any real, deeply rooted public opinion.

Most Americans are aligned with a party (Republican or Democratic) and a general ideology (liberal, moderate, conservative) and have some general inclinations on policy (government spending is too high; Social Security is good). But they can change their minds quickly on a wide range of issues, particularly if leaders in whatever party they are aligned with change theirs.

So what is polling good for?

Polling should used to assess if there is a gap between politicians’ behavior and public opinion on issues that are simple to understand and in the news often enough that it is likely most voters have some real opinions. The U.S. government system was intentionally created to have some checks on popular sentiment, such as the Senate and guarantees of certain rights in the Constitution. But even if the public doesn’t always get its way, it’s useful to understand where the public is and whether government policy is in sync with public opinion.

The volume of polling showing most Americans oppose near-total bans on abortion and the defeat of an antiabortion ballot initiative in Republican-dominated Kansas suggests there is a clear public sentiment on the issue. The abortion restrictions being passed by Republican-dominated states aren’t improper or illegitimate simply because they are unpopular. But without polls (and the defeat of an antiabortion ballot initiative in Kansas), we might assume these bans are passing because of mass support for them — as opposed to them passing despite mass resistance.

Polling should be used to assess changes in public opinion. I am not sure whether 43 percent of American adults approve of Biden, as FiveThirtyEight’s average of polls suggests. Perhaps the real number is 39 percent or 47 percent. Polls do not capture the exact percentage of Americans who support a given idea or candidate.

But many organizations regularly conduct polls using the exact same methods over several months or even years, which means we can trust the changes in public opinion that they show. It is almost certainly the case that Biden is more popular than he was in late July, when the average of those same polls put him at about 38 percent approval.

Why is analyzing change in public opinion important, if we aren’t sure about the overall number? Because much of political coverage and politics itself involves trying to figure out what actions or tactics worked electorally — and therefore if they should be repeated or discarded. For example, if Biden’s poll numbers had plunged immediately after he announced his student debt cancellation proposal, future politicians might have been nervous about debt cancellations. Instead, Biden’s numbers went up.

The polls haven’t proved that debt cancellation is an electoral boon. But the results we have so far suggest debt cancellation also isn’t election kryptonite.

Polling should be used to assess which elections don’t need as much coverage as others. Despite the mixed record they have for projecting the winner of races in swing states such as Wisconsin, when polls and forecasts show a candidate in a congressional race is an overwhelming favorite or underdog, they are usually correct.

People choosing which candidates to volunteer for or donate money to can rely on these polls to avoid investing in those with no chance to win. News organizations can downplay coverage of these races.

It’s not that likely blowouts shouldn’t be covered at all. I’m glad reporters in Kentucky are tracking the Senate campaign of Democrat Charles Booker, who is running a long-shot race against incumbent Sen. Rand Paul (R). Perhaps the forecasts are wrong and Booker will win. More important, I want to know what issues Booker raises in his campaign, even if he doesn’t have much of a chance.

Polling should be used to identify smaller groups within the electorate and their views. We only have a big, mass-turnout national election every four years. From those elections, we get fairly limited data — essentially how many votes various candidates received, broken down by neighborhood, city and state. With polling, we can see the disparate voting patterns of various demographic groups, such as Black Americans or people who describe themselves as White evangelicals. So-called exit polls conducted around elections capture the broader dynamics of the electorate (most White voters in the South back Republicans, for example), even if they, like other polls, are often wrong in their precise details.

Non-election polling gives important insights into those groups’ political views. For example, the Pew Research Center occasionally does surveys asking thousands of Black Americans questions. The result is a much richer, more complicated portrait of Black views than simply that Black people overwhelmingly back Democratic candidates.

Polls are very useful. They’re just not useful at predicting winners in close races in swing states — the one thing everyone wants them to be useful in doing.