The 3 Big Questions I Still Have About Election Day

The 3 Big Questions I Still Have About Election Day

With less than 24 hours to go until we freeze our model,1 we’re probably not going to see a lot more changes in the topline forecast. Republicans have a 54 percent chance of winning the Senate and an 83 percent chance in the House, according to our Deluxe forecast. Those numbers have been relatively steady for the past few days.

So let’s instead ask three big forecasting questions about the race. These aren’t questions like “how much will abortion matter,” which I hope we’re doing a good job of addressing elsewhere at FiveThirtyEight. Instead, they’re questions that will inform our understanding of future elections from an analytical, polling and forecasting standpoint.

Question 1: Will the polls be systematically wrong?

We’ve covered this topic so often that you might think I have nothing more to say about it — but you’d be wrong!

That’s partly because it really is the whole ballgame, at least when it comes to which party keeps control of the Senate. If Republicans beat their polling averages by, say, 3 percentage points across the board, it would be very unlikely for Democrats to salvage the Senate even if there’s some state-to-state variation. (Perhaps they could hold on in Arizona and New Hampshire, but they’d be considerable underdogs in Nevada, Georgia and Pennsylvania, where they need to win two out of the three.)

Likewise, if Democrats beat their polls by 3 points across the board, the picture is very rosy for them. In that event, they’d be favorites in Nevada, Georgia and Pennsylvania. Arizona and New Hampshire would probably be out of reach for Republicans. But conversely, Ohio, North Carolina and Wisconsin would be in reach for Democrats. The House would also be highly competitive in this circumstance.

Which type of polling error is more likely? If you’ve been following these election updates, you’ve probably noticed me wrestling back and forth between a “trust the process” mentality, where I take the FiveThirtyEight’s model’s output as gospel, and a concern that the model may still be underestimating the chance of a pro-Democratic bias in the polls as we saw in 2016 and 2020.

Watch: https://abcnews.go.com/fivethirtyeight/video/parties-making-final-pitches-voters-fivethirtyeight-politics-92670196

For what it’s worth, I’ve mostly landed on “trust the process.” My personal view of the race is pretty well aligned with the FiveThirtyEight Deluxe model. The polls could very well be biased against Republicans again. The best reason to think so is probably the “Nathan Redd” argument that as polling gets more difficult, you should put more faith in the fundamentals. Usually, the president’s party has a rough midterm, especially when the president has a 42 percent approval rating and inflation is at 8.2 percent.

But it’s not hard to imagine how the polls could be biased against Democrats instead. After 2016 and 2020, pollsters face more reputational risk from again missing high on Democrats than the other way around, and that could consciously or unconsciously affect decisions they make at the margin, or even which polls they release to the public. Moreover, the composition of polling averages has considerably changed, with fewer “gold standard” polls and more quick-and-dirty ones that tend to show more favorable results for Republicans.

I’m not sympathetic to Democratic complaints about Republican pollsters “flooding the zone.” Not all polls with a Republican-leaning house effect actually have any formal affiliation with the GOP. Also, to the extent that polls do have a house effect — that is, consistently leaning toward Democrats or Republicans — our model adjusts for it.

Besides, being the University of Chicago economics major that I am, I mostly trust the market to sort everything out. Firms with a Republican house effect will lose business and credibility in future election cycles if Democrats have a good night. Conversely, some traditional pollsters like Monmouth University aren’t even publishing final horse race numbers in the races that they’re polling. If GOP-leaning firms like Trafalgar or InsiderAdvantage are willing to put their credibility on the line and Monmouth isn’t, that tells you something.

Still, we’re a long way removed from the Golden Age of Polling circa 2006-2012 when “gold standard” pollsters (live-caller telephone polls with transparent methodologies) could be counted upon to set a reliable benchmark. Nobody in the polling or election forecasting community has any right to be all that confident about what will happen on Tuesday. That might make you want to give up and trust the vibes or insider sentiment about the race, which is bearish for Democrats. But the track record of vibes is that they’re somewhere between useless and worse-than-useless, like in 2016 when insiders thought Hillary Clinton would have a cakewalk.

Watch: https://abcnews.go.com/fivethirtyeight/video/governors-races-watch-fivethirtyeight-92674801

It’s also worth mentioning that the FiveThirtyEight Deluxe model is slightly more favorable for Republicans in the Senate than the polls-only version of our forecast, which essentially means that it does predict a little bit of pro-Democratic bias in the polls, especially in some of the redder states such as Ohio. So by endorsing Deluxe’s view, I’m putting a pinky finger on the scale for the view that polls will again have a Democratic bias, while also being open to a mere severe bias in either direction.

I do think this is a big year for the pollsters. In 2016, the polling error wasn’t really all that bad, and pollsters had a semi-decent array of excuses from shifts in voter coalitions that made it important to weight polls by educational attainment to a crazy October and November news cycle, including the Comey letter. In 2020, COVID-19 presented real challenges for polling. This year, there aren’t as many contingencies. Most Americans have given up on COVID-19 precautions, there have been no late-breaking news events with obvious electoral implications, and Donald Trump isn’t on the ballot. 

Question 2: How big will the turnout gap be?

 I’m sorry to repeat the biggest cliche in election analysis, but if the polls are roughly in the right vicinity, control of the Senate will come down to turnout. If you care about the outcome and haven’t voted, you should do so.

On that front, Democrats got good news on Sunday with two major network polls showing a relatively small turnout gap. (I’ll use the terms “turnout gap” and “enthusiasm gap” interchangeably here; in both cases, I refer to the difference in margin between the likely voter and registered voter version of a poll.) The first, from our colleagues at ABC News and The Washington Post, showed Republicans 1 point ahead on the generic ballot among registered voters but 2 points ahead among likely voters, making for only a 1-point enthusiasm gap in the GOP’s favor. Meanwhile, an NBC News poll had the generic ballot tied among registered voters but Democrats 1 point ahead among likely voters, meaning that there was actually a tiny enthusiasm gap in Democrats’ favor.

In both cases, that reflects improvements in Democratic enthusiasm from earlier this cycle. The previous ABC News/Washington Post poll had shown a 4-point turnout gap favoring Republicans, and the prior NBC News poll had Republicans gaining 2 points from their likely voter screen instead of it helping Democrats.

Most other polls show a modest enthusiasm gap in favor of Republicans. Republicans gain 3 points from the likely voter screen in CNN’s recent generic ballot poll, for instance, 2 points from Siena College and The New York Times, and 1 point from Echelon Insights. But these numbers can be noisy. In Marquette Law School’s early October poll of Wisconsin’s U.S. Senate race, the race was tied among registered voters but Republican Sen. Ron Johnson led by 6 points among likely voters. In Marquette’s more recent poll, however, Johnson led by 3 among registered voters but actually lost a point and led by 2 among likely voters.

If there could be anything from a 6-point turnout gap favoring Republicans (as in Marquette’s October poll) to a 1-point turnout gap favoring Democrats (as in its latest one), that creates a wide range of plausible outcomes. And neither of those scenarios are necessarily crazy. There was roughly a 6-point enthusiasm gap favoring the GOP in the 2010 midterm, for example. 

A turnout gap of any magnitude favoring Democrats would be unusual — typically voters from the opposition party have more enthusiasm at the midterms than the president’s party, and typically Republican voters are more likely to turn out than Democratic ones. But political coalitions are shifting, with Democrats increasingly relying on college graduates, who are considerably more likely to turn out to vote. Moreover, there are some tangible signs of high Democratic enthusiasm. Their House and Senate candidates have raised more money than Republican ones, and Democrats performed very well in a series of ballot initiatives and special elections over the summer, often on the basis of superior turnout.

A Wonky Aside About The Generic Ballot, the House Popular Vote, and the “National Environment”

Before we get to the third question, a note on terminology. You’ll often hear us use the terms “generic ballot,” “House popular vote” and “national environment,” but they mean somewhat different things.

The FiveThirtyEight generic ballot tracker combines what are really two types of polls. One set of polls asks voters which party they’d rather see in control of Congress. Another type asks them whether they plan to vote for the Democratic or Republican candidate in their U.S. House race.2 Since these questions tend to produce similar results, we combine them to increase the sample size and consider both to be generic ballot polls.

After the election, you can evaluate these polls based on how well they predicted the House popular vote. That’s the number you get when you add up votes for Democratic and Republican U.S. House candidates in all 435 Congressional districts. In 2020, for instance, Democrats won this measure by 3.1 percentage points — similar to, but less than, Joe Biden’s 4.5-point win in the presidential popular vote.

One potential source of divergence between these measures is that in some districts there’s no Democrat or Republican on the ballot, either because the party didn’t bother to nominate a candidate or because it’s a state like California and the top two finishers in the primary advanced to the general election regardless of party.

This year, there are considerably more districts with no Democratic nominee than with no Republican. Specifically, there are 23 House districts with no Democrat on the ballot3 but 12 with no Republican. Moreover, the districts with no Democratic nominee tend to be more competitive than those with no Republican one, meaning that Democrats are sacrificing more votes.

It’s slightly tricky to calculate exactly how big this effect is, but it will likely shift the final House popular vote margin by at least 1 percentage point toward Republicans, and probably more like 1.5 percentage points. In other words, if the final generic ballot margin was Republicans by 3 percentage points, we’d expect them to win the House popular vote by more like 4.5 percentage points because of all the districts with missing Democratic candidates.

Finally, you’ll sometimes hear an analyst like me describe the “national environment,” usually in a context like the following: “Democrats will need a D+1 or D+2 national environment to keep the House.” This term is more ambiguous, although I think of it as referring to what you’d expect to happen in a perfectly neutral setting: a district with no incumbent, two “average” candidates, and no partisan lean.

Let’s sort through how all of these numbers work this year:

  • Our generic ballot tracker has Republicans ahead by 1.1 percentage point.
  • However, our model uses a slightly different version of the generic ballot that includes a likely voter adjustment.4 This helps Republicans, and they’re ahead by 1.7 percentage points in this version.
  • Here’s where it gets more complicated. Our model also makes a forecast of the House popular vote. The estimate is determined by literally forecasting the popular vote in each district one race at a time. It is not just based on the generic ballot: it also accounts for factors such as district-by-district polling and incumbency. We also forecast how many people will vote in each district, accounting for past turnout and even factors such as population growth. Currently, this forecast has Republicans winning the House popular vote by 4.0 percentage points
  • However, as mentioned earlier, when you tally up the race-by-race forecasts, Democrats get a lot of zeros in their column because of all the districts with no Democrat running. If you back out this effect — if both parties had a candidate on the ballot in every district — Republicans would be predicted to win the House popular vote by more like 2.5 percentage points instead. This is roughly what I think of as FiveThirtyEight’s forecast of the national environment. It’s about how you’d expect a Congressional race to go in an average district with no incumbent or partisan lean.

So in case you’re wondering why Democrats have only a 17 percent chance of winning the House despite our generic ballot average being within a percentage point: our more detailed forecast suggests that Republicans have more like a 2- or 3-point advantage. On top of that, they have a slight advantage because of how voters are distributed between districts, although this is less than it was before redistricting. Thus, Democrats would have to beat our forecast of the national environment by 3 or perhaps 3.5 percentage points to be favored to win the House. That’s hardly impossible — we’re talking about a normal-sized polling error — but it’s a little tougher than our generic ballot average implies.

Question 3: How much does candidate quality matter?

Here’s an important fact about this election that’s become somewhat obscured. If it weren’t for the candidates — some relatively strong Democrats and some relatively weak Republicans — Democrats would be completely screwed in the Senate, barring a major polling error.

As I said, FiveThirtyEight estimates that the national environment favors Republicans by about 2.5 percentage points. Now take a look at the states where pivotal Senate races are being held: Nevada, Pennsylvania, Wisconsin, North Carolina, Georgia, Arizona and Ohio are all Republican-leaning relative to the country overall. (Consider that Biden underperformed his 4.5- point national popular vote victory in each of these states, even though he won several of them.) New Hampshire is Democratic-leaning according to our index, but just barely, enough that you’d expect it to also go Republican in a year where the national environment is GOP +2.5.

So if all races went according to the national environment plus the state’s partisan lean, Democrats would lose the seats they currently hold in Nevada, Georgia, Arizona and New Hampshire while failing to make any gains from Republicans, resulting in a 54-46 GOP Senate. That sort of outcome is not out of the realm of possibility by any means at all, but it’s relatively unlikely. The GOP may well pay a price for its inexperienced, unpopular and in some cases scandal-plagued candidates. Just how much of one could determine which party winds up with Senate control.

Watch: https://abcnews.go.com/fivethirtyeight/video/election-deniers-favored-win-midterm-races-fivethirtyeight-92803581

Powered by WPeMatico

Comments are closed.
%d bloggers like this: