Last week, I argued that there weren’t yet clear signs of a Republican comeback in polls of the midterm elections.
This week? The case for a GOP rebound is a little better.
For one, Republicans can cite some individual polls that have been favorable for them. There was a poll this weekend from ABC News and the Washington Post, which found Republicans 5 points ahead on the generic congressional ballot among likely voters — a result which would all but assure the GOP of winning the House and make the Senate an uphill battle for Democrats. Republican candidates have also pretty clearly gained ground in some individual Senate races, such as Pennsylvania and Wisconsin.
There are counterexamples, though. Another highly rated pollster, Selzer & Co., published a poll this week showing Democrats 4 points ahead on the generic ballot among likely voters. Meanwhile, Democrats can cite a few strong recent Senate polls of their own in states such as Ohio and New Hampshire.
Overall, Democrats’ chances of winning the Senate are 68-in-100, according to the Deluxe version of our forecast. That’s down from a high of 71-in-100, which persisted from Sept. 13 to Sept. 20, though that change isn’t large in the big scheme of things. In the House, Democrats’ chances are 31-in-100, which reflects an even smaller change; their chances were 32-in-100 on Sept. 23 and 24.
Still, we’re overdue for a reminder of why Republicans still could do quite well for themselves in November. So here are five major reasons for hope for the GOP — and concern for Democrats.
As an added twist, we’ll engage in a bit of model talk as we review these reasons. That is, I’ll tell you whether I think these reasons for hope/concern (depending on who you’re rooting for) are adequately accounted for by our model — or whether they reflect information outside of its scope and assumptions.
The fundamentals still ought to be pretty good for Republicans
Sure, this election could be an exception to the usual midterm pushback against the president’s party. But, it’s not always smart to bet on exceptions. So it’s worth going back over the reasons why I and other analysts initially expected this election to be a strong one for Republicans.
- The opposition party — in this case, Republicans — has a long history of doing well in the midterms. It’s one of the more robust trends in American politics.
- Though his numbers have improved recently, President Biden remains fairly unpopular, with an approval rating of 42 percent.
- Americans remain quite unhappy about the direction of the country, which tends to hurt the incumbent party.
- Republicans have a big structural advantage in the Senate given that they do well with white, rural voters and that white rural voters are considerably overrepresented in the Senate relative to their overall share of the U.S. population.
Does the FiveThirtyEight model account for this? Yes. The Classic and Deluxe versions of our model use a blend of polls and fundamentals when evaluating the overall political environment and individual races.
This makes a pretty big difference in the House, where our Lite forecast, which tries to predict the race based on polls alone, gives Republicans a 61-in-100 chance of a House takeover. Those odds jump to 68-in-100 percent in the Deluxe forecast, which takes fundamentals into account. This also boosts the odds of Republicans, who have a 23-in-100 chance of winning the Senate in Lite but a 32-in-100 chance in Deluxe.
The differences between the forecasts were larger at earlier points in the cycle; there isn’t as much time left now for the polls to converge toward the fundamentals. Still, the model assumes that Republicans are more likely to gain ground than to lose ground in the polls between now and November.
The “issue environment” could get better for Republicans
This is the argument recently advanced by Nate Cohn of The New York Times. During the summer, Democrats benefited from media and voter attention to the Supreme Court’s decision to overturn Roe v. Wade, the Jan. 6 hearings, and the FBI investigation of former President Donald Trump. These issues tend to remind voters of Republican power and influence and potentially make the election more of a 2020-like choice between Democrats and Republicans and less of a referendum on Biden and Democrats’ performance.
But now, issues like immigration and the economy are again getting more headlines. That’s not to say abortion and those other issues won’t matter, but the summer could have reflected a best-case scenario for Democrats.
Does the FiveThirtyEight model account for this? Not exactly. The model accounts for uncertainty due to future news events. But it doesn’t make any assumptions about whether these surprises tend to benefit one party over another.
But … I’d be careful overthinking this. People who cover politics for a living, including me, live in a bubble full of people who pay far more attention to day-to-day political developments than regular voters do. Pundits’ guesses about which stories voters will care about, or how they will affect their voting preferences, are often wrong. Moreover, the life cycle of stories tends to be pretty short. Florida Gov. Ron DeSantis’s decision to send immigrants to Martha’s Vineyard was a huge story until about a week ago, but now we have a devastating hurricane and a new wave of economic concerns that have usurped it. So in practice, assuming that future news events create uncertainty in both directions will usually be the best course.
Republicans could overtake Democrats in advertising
If I were actually betting on election outcomes at prediction markets — and no, I don’t do that for a variety of reasons! — I’d pay attention to the flow of campaign advertising. Advertising may not matter that much in terms of moving votes in an era with relatively few swing voters. But it certainly matters some, and the effects can be short-lived. So if one candidate had a temporary edge on their airwaves, it might lead to a temporary surge in the polls, which would then recede.
So one potential source of concern for Democrats would be if they’d dominated the advertising race during the summer, but only temporarily.
Is this happening? I honestly don’t know. There are some accounts of Republican groups having boosted spending recently. But while public fundraising and campaign spending is relatively easy to track, outside spending also plays a large role and is more opaque. Moreover, outside spending has tended to favor conservatives and Republicans this cycle. With that said, not every dollar is spent equally well. Outside groups don’t get the same favorable advertising rates that campaigns themselves do, and it’s cheaper to buy ads in advance rather than at the last minute.
Does the FiveThirtyEight model account for this? Not really. The model doesn’t make any assumptions at all about advertising. It doesn’t discount or adjust a poll because it came during an advertising blitz, for example. Nor does it make any assumption about future advertising spend. The model does account for fundraising; money raised from individual donors is one of the factors that the model evaluates in addition to the polls. But this is more because fundraising tends to be a proxy for grassroots support rather than because of any additional advertising leverage it provides.
Republicans are more likely than not to have an enthusiasm advantage
One reason that the aforementioned ABC News poll was so worrying for Democrats is because of the relatively large enthusiasm differential it showed between registered voters — among whom Democrats trailed on the generic ballot by only 1 percentage point — and likely voters, among whom they trailed by 5. That 4-point “enthusiasm gap” was somewhat reminiscent of the 2010 midterms, when there was roughly a 6-point gap between polls of registered and likely voters, leading to a catastrophic election for Democrats.
But other polls do not show the same enthusiasm disadvantage for Democrats. As Cohn mentions, for instance, there was only a 1-point enthusiasm gap in The New York Times’s recent generic ballot polling. Other polls in the wake of the Dobbs decision found that while Republicans were more excited than usual to vote at the midterms, so were Democrats. Democrats have also tended to have strong turnout in recent special elections and ballot referendums. The consensus of the evidence points toward a modest enthusiasm gap favoring Republicans — enough to make matters harder for Democrats, especially in the House — but nothing like the magnitude of 2010.
Does the FiveThirtyEight model account for this? Yes, the model explicitly accounts for the likely GOP enthusiasm advantage. Specifically, it adjusts polls of registered voters to make them equivalent to polls of likely voters.
Initially, this adjustment is based on historical patterns: Republican voters have historically been more likely to turn out than Democratic ones, and voters of the opposition party are more likely to turn out than voters from the president’s party. In 2018, those factors offset one another (Republicans had the traditional enthusiasm advantage, but Democrats were the opposition party). But this year, they’re both working in the GOP’s favor.
I can also see an argument that this assumption is too generous to Republicans, though. Voters at the midterms tend to be older, whiter and more well-educated than voters in presidential years. Historically, these factors all tended to favor Republicans. Now, however — while Republicans are still older and whiter — Democrats tend to dominate among college-educated voters, who are considerably more likely to turn out to vote.
There’s still some reason for Democrats to be concerned about the reliability of the polls
Let me put this carefully. Two weeks ago, I wrote about why I don’t agree with the presumption that polls — despite their shortcomings in 2016 and 2020 — are biased toward Democrats:
Here, I’m going to present something of a rebuttal […] to the presumption I often see in discussion about polling that polling bias is predictable and necessarily favors Democrats. My contention is that while the polls could have another bad year, it’s hard to know right now whether that bias will benefit Democrats or Republicans. People’s guesses about this are often wrong. In 2014, for example, there was a lot of discussion about whether the polls would have a pro-Republican bias, as they did in 2012. But they turned out to have a pro-Democratic bias instead.
There are a lot of qualifications here. It doesn’t mean that the polls can’t have a bad year. But the idea behind the FiveThirtyEight model is that they’ll be biased toward Democrats in some years, they’ll be biased toward Republicans in some years, and they’ll be unbiased in some years — and it’s hard to know what you’re getting ahead of time.
I’ve read various rebuttals to this article, some of which I find a little lacking. For instance, I simply don’t think it’s correct to claim, as Andrew Prokop recently did in Vox, that polls overestimated Democratic performance in 2018. To treat anti-Republican bias in the polls as a sort of inescapable empirical regularity based on a sample of two election cycles, neither of which was the most recent midterm, is making too much of limited evidence.
But does that mean I give no credence at all to what we might call the Predictable Democratic Polling Bias (PDPB) hypothesis? No, of course not. Good fox-like forecasters can hold multiple ideas or hypotheses in their head at once. PDPB has some things going for it. My mental model of the race might be something like: mostly the FiveThirtyEight forecast, but with a little smidge of weight that PDPB is real.
What’s the case for PDPB? Again, I don’t think it’s an empirical one. Empirically, the evidence strongly shows that polling bias is unpredictable.
But the theoretical case that analyst Richard Hanania makes for PDPB in his recent Substack post is plausible. If, fundamentally, Republicans are less likely to answer surveys because (for instance) they have lower social trust, that’s not necessarily an intractable problem. But it could be one that hasn’t been addressed yet. It could take pollsters a couple of election cycles to work out the right approach, especially if it requires abandoning traditional polling best practices.
Interestingly enough, PDPB wouldn’t necessarily doom Democrats in the Senate. Even if polls had a 3-point pro-Democratic bias, Democrat John Fetterman would still be favored in Pennsylvania, where he leads by 6 points in our polling average. Democratic incumbents Raphael Warnock in Georgia and Catherine Cortez Masto in Nevada would technically be underdogs (they currently lead by 2.1 and 0.7 percentage points, respectively). But Georgia and Nevada have had fairly accurate polling in recent years. Any amount of PDPB would doom Democrats’ chances in the House, however.
Does the FiveThirtyEight model account for this? Sort of.
Although the model doesn’t assume any sort of permanent polling bias, it does give some weight to the fundamentals, and as I mentioned above, the fundamentals are more favorable to Republicans than the polls alone. The Deluxe forecast also considers expert ratings, and although how it uses these ratings is slightly complicated (read the methodology section for more), those do help Republican odds in the Senate.
Perhaps if you give a little bit of extra mental weight to PDPB, it would be enough to account for the relatively small differences between the FiveThirtyEight forecast and betting markets. For instance, betting markets give Republicans a 39 percent chance of winning Senate control, as compared to 31 percent in our Deluxe forecast, and a 75 percent chance of House control, instead of 68 percent. Those are small enough differences that I’m not sure I’d make huge bets on either side.