It’s hard to reconcile pre-election polling of independents with what the exit polls told us.
It’s hard to reconcile pre-election polling of independents with what the exit polls told us.
O ne of the most striking findings in the exit polls from the 2022 midterm elections was that Democrats won independent voters. According to the exit polls, they won independents in the national vote for the House by a narrow margin of 49 percent to 47 percent, while winning those voters by a decisive margin in many of the headline races in battleground states.
It’s a finding that is surprising as a historical matter, and one that should give Republican campaigns serious pause. But is it accurate?
My alarms went off on the exit polling on independent voters for three reasons:
- Historically, there is a powerful tendency for independents to support the party out of power. Things that are typically true in the past don’t always continue forever, but it’s a strong enough pattern that we should look carefully at data saying things really were different this time.
- Whenever I looked at the crosstabs of polls during the past year and a half, Democrats were getting absolutely murdered with independent voters, who had soured on Joe Biden; it was surprising that they would take the opposite turn at the end.
- Looking at the House popular vote, Republicans got more votes than Democrats did. Thus, if Republicans lost independents, the electorate must have been more Republican-leaning than any electorate in a very long time.
None of these points indicates with certainty that the poll results must be wrong. Sometimes, the data surprise us! But they do suggest a closer look is warranted.
Begin with what was shown in the exit polls:
According to the exit polls, Republican candidates got beaten soundly among independent voters in both races in Pennsylvania and Arizona, in the Michigan and Wisconsin governor’s races, and in the Georgia, New Hampshire, and North Carolina Senate races. Those voters were narrowly in favor of one side or the other in most of the other races where we have exit polls, with only three incumbent Republican governors bucking the trend, in races that ended up as blowouts: Mike DeWine in Ohio, Chris Sununu in New Hampshire, and Ron DeSantis in Florida.
That is not how elections, especially midterms, typically go. Since 2004, looking at either the presidential vote or the national House vote in midterms shows that the party in power has lost independent voters in every election. Barack Obama and George W. Bush both lost independents while winning reelection, Bush by two points and Obama by five. In midterms, independents have gone for the out-of-power party by double-digit margins in four straight elections, producing the familiar wave dynamic that many of us expected to see repeated this year:
Interestingly — we’ll come back to this — independents made up 31 percent of the exit poll sample this year, up five points from 2020 and short of the 2016 high point of 34 percent. Not coincidentally, the 2016 election featured two candidates who were both widely loathed by the voters in general.
Should we have been warned by the pre-election polls that independents were in a different mood this year? Not if we were looking at the national polling. I drilled down into the polls that were in the final RealClearPolitics poll average. Breakdowns of independent voters were not available for every poll, but looking at the Biden job approval polls, if we average the five polls that had party-identification crosstabs, they showed Biden to be underwater with independents by an average of 26 points.
Those five polls were somewhat more hostile to Biden than the ones that had no publicly available crosstabs, and the final exit polls showed Biden doing slightly better in his approval rating overall, with 44 percent rather than 42 percent approval, but still 55 percent disapproval. In short, there is really nothing here that would lead a careful reader of the polls to suspect that Biden was anything other than deeply unpopular with independent voters.
We do, however, begin to see a problem with the polls: a wide variance in the estimated share of the electorate composed of independents. The estimates ranged from the Reuters/Ipsos poll (conducted among adults, not registered or likely voters), which had independents as 10.6 percent of respondents, to the ABC News/Washington Post poll of registered voters, which had independents as 39 percent. On average, these polls undercounted independents relative to the final exit polls by 2.4 points, projecting an electorate that was 28.6 percent independent rather than 31 percent.
That brings us to the generic-ballot polling. The generic-ballot polls were not bad as far as projecting the final outcome: a national popular vote advantage of around three points for Republicans. A few pollsters were way off, such as the Politico/Morning Consult poll that had Democrats up five, but they averaged out well. Yet, among the eight polls that provided crosstabs, Republicans led with independent voters in seven of eight polls, by an average of ten points. That’s a pretty big miss. And once again, we see an enormous variance in the independent share of the respondents, ranging from 9 percent to 39 percent.
One potential problem emerges: Even aside from varying methods of determining who is likely to vote, different pollsters use different approaches to identifying who is an independent voter. Some pollsters are more willing to rest on how a voter is registered, others on how a voter self-identifies; some press harder than others to find out which way a voter leans. When I interviewed Robert Cahaly of Trafalgar in September, he expressed his deep skepticism of voters’ partisan self-identification, and you will notice that Trafalgar’s polls consistently show fewer independents in the electorate than do other pollsters. That’s not because Cahaly isn’t talking to those voters, it’s because he’s classifying more of them as partisans.
In order to make sense of how polling of independent voters conflicted with the exit polls, I decided to look in detail at the polls in seven high-profile statewide races. I selected big, disappointing Senate races but also two Republican Senate wins and a race for governor. The results allow us to look more closely at how the two sets of polls collide.
Pennsylvania Senate
The Pennsylvania Senate campaign was the headline race of 2022, deciding control of the Senate. At this writing, it is the only Senate seat to change parties this year. There were five polls in the final RCP average, to which I added the most recent nonpartisan poll (from Emerson) in order to get a deeper sample of polling of independents. Even adding Emerson, Mehmet Oz had a by-decimal-points lead in the polls, but he lost by four points. According to exit polls, he lost by 20 among independent voters.
Pennsylvania’s electorate had very few independents. According to exit polls, 40 percent of the voters were Republicans and 37 percent Democrats, with the remaining 24 percent independents (yes, I know that adds up to 101 percent, but it’s Pennsylvania). Oz’s loss, according to the exit polls, is explained by two simple calculations: He lost 10 percent of self-identified Republicans, while John Fetterman lost just 5 percent of Democrats, and he was routed among independents.
The polls never saw it coming; they had Oz winning independents by nine points. Marist was the only pollster even remotely in the ballpark of a 20-point rout for Fetterman among independents. Insider Advantage and Remington, both Republican pollsters, had Oz winning by 32 and 23 points, respectively, among independents.
Nevada Senate
Nevada was the Republicans’ biggest Senate disappointment, with Adam Laxalt leading for months in the polls all the way to his defeat. It is therefore a ripe field for investigating polling failures. All of the final polls attempted to limit their samples to likely voters, and the larger the sample size, the surer the poll was of Laxalt’s lead.
On average, the polls correctly projected the share of independents in the electorate; but they expected Laxalt to win them by 15 points, where he instead lost them by 3. That alone was more than enough to explain the outcome. Explaining why the exit polls of independents diverged so far from pre-election polls is another story.
New Hampshire Senate
New Hampshire was another state where the polling showed a close race but the voters said otherwise. Pre-election polls were all over the lot on independent voters, ranging from a ten-point lead for Maggie Hassan in Emerson to 19-point leads for Don Bolduc in the Daily Wire/Trafalgar poll and in one version of the St. Anselm poll. The outliers, however, tended to be polls that showed a much smaller independent vote than the 43 percent that showed up — a figure in line with New Hampshire exit polls in past elections. Presumably, the pollsters showing a big Bolduc lead among independents were, in part, classifying a lot of registered independents as Democrats. The St. Anselm poll makes that apparent: Bolduc led by 19 points among self-identified independents, who made up just 8 percent of the electorate, but he and Hassan were tied among registered independents, who made up 36 percent. Of course, given how the final results diverged from several of these polls, it stands to reason that the pollsters simply undersampled independents and oversampled partisans. And even Emerson underestimated Hassan’s margin of victory.
Arizona Senate
Even including the Marist poll, which was left out of the RCP average in favor of polls taken closer to the election, the polls had Blake Masters in a dead heat with Mark Kelly. Instead, he lost by four. These polls did not assume that Masters would rout Kelly among independents, as some of the others did, but they had him hanging four points back, while the exit polls say that he lost independents by sixteen. Again, the polls showing Masters to be competitive among independent voters tended mainly to greatly understate the independent share of the electorate — 29.1 percent compared with 40 percent in the final exit polls. Remington, the only pollster who projected Masters to win independents, still had him losing and coded only 24 percent of respondents as independents, suggesting that it was very aggressive in categorizing Democrat-leaning independents as Democrats.
North Carolina Senate
Unlike the other Senate races discussed above, Republicans won the open (Republican-held) Senate seat in North Carolina behind Ted Budd. North Carolina is quite different from New Hampshire: it has a highly racially polarized electorate with comparatively few true independents. Pre-election polls nailed the outcome, as Budd won 51 percent of the vote, just as in the poll average; his opponent, Cynthia Beasley, got 47 percent, up from 45 in the final polls.
And yet, once again, the exit polls diverged dramatically from the pre-election polls in assessing the turnout and sentiments of independent voters. Those voters made up 31 percent of the exit poll sample, compared with 26.5 percent of the pre-election poll average. Moreover, the pollsters that were closer to estimating the exit-poll share of independents were further away from capturing their sentiments, projecting double-digit Budd margins compared with the six-point Beasley win reflected in the exit polls. But the exit polls saw the electorate as 37 percent Republicans and 32 percent Democrats, which was the difference.
Florida Senate
While Budd won a fairly close race in his state, in Florida Marco Rubio won in a blowout by nearly 17 points. Yet even Rubio is said by the exit polls to have won independent voters by one point, where pre-election polls had him winning by almost ten. Again, Rubio’s margins were larger in polls that greatly lowballed the number of independents compared with the exit polls. But Florida’s electorate was, according to those exit polls, a whopping 42 percent Republicans and only 28 percent Democrats. That suggests, at minimum, that the strong brand of the DeSantis/Rubio party in Florida attracts a lot of the state’s previously independent voters to identify as Republicans now — a conclusion consistent with the sharp increase in Republican voter registration.
Wisconsin Governor
One of the most disappointing races for governor, from Republicans’ perspective, was in Wisconsin, where Tim Michels led Tony Evers narrowly in the polls but lost by three points (three points being a big margin by Wisconsin standards). Only Trafalgar, which saw the Wisconsin electorate as just 11.4 percent independent compared with the 30 percent in the exit polls, had Michels winning those voters; but several polls had him competitive. The Marquette poll, traditionally the gold standard of Wisconsin polls, had a dead heat, with Evers leading by one among independents. Instead, the exit polls showed him winning those voters by six.
Some Conclusions
It is easier to spot the problems with this picture than to come up with answers. We know that exit polls are likelier than pre-election polls to be accurate, but both have their flaws. Clearly, part of the difficulty is the variance among pollsters in defining “independent” voters, a group not easily defined by the voters themselves. Voter identification with a party is notoriously fluid and affected by the voters’ feelings toward particular candidates and issue environments, a fact that has convinced me since 2012 to stop trying to use party-ID surveys to predict elections or to use the projected partisan composition of the electorate to critique pre-election polls.
We can, however, put forward some hypotheses. One is simply that a big chunk of this year’s undecided voters broke hard against Republican candidates because several of those candidates were so bad. But that doesn’t hold water for, say, Marco Rubio, an experienced politician who won a lopsided victory. Another is that Democrats managed to drive to the polls a fair number of voters who were not strong partisans but cared a lot about abortion. That, too, seems an inconsistent thesis: It doesn’t explain why DeSantis and DeWine nonetheless did much better than their own state’s Senate candidates with independents despite signing laws restricting abortion, and it would be more plausible if we saw a large surge in the youth vote, which we did not (young voters being the sort of people who might vote on one issue without feeling allied with either party). If turnout was the main explanation, it would also not suffice to explain how the polls as a whole were on the nose in North Carolina and if anything understated the Rubio and DeSantis wins in Florida.
We do know this much: Republicans will need a more rigorously considered answer if they intend to do better next time.