One of the weird things about this cycle is that in many ways, it truly is not like any other. President Joe Biden is at 44% approval with registered and likely voters, and yet the generic ballot and special election results indicate a tight national environment. Former president Donald Trump continues to command an unusually high percent of the news cycle. And while the conventional wisdom states that the midterm shift against the president’s party comes partially from thermostatic effect against unpopular policies enacted by the party controlling Congress, the biggest policy shift in recent memory actually came from the out-party through the conservative Supreme Court’s decision to overturn the right to an abortion in Dobbs v. Jackson.
All of these things mean that the race for Congress is much closer than you would expect in a midterm election with an unpopular president. Democrats have seen a string of special election victories and overperformances since Dobbs, and their generic ballot position appears to be significantly stronger than it was before the ruling. However, Republicans have recently taken the lead in the FiveThirtyEight polling averages, and while the overall picture is still quite complicated, there is some serious reason to believe that they have seen electoral gains of late.
To begin with, the polling average on FiveThirtyEight has gone from D+1.9 two weeks ago to R+0.1 today, a shift of roughly two points. Another, more detailed lens looks at poll-by-poll movement from individual pollsters, and it shows the same picture. These were the same metrics that first indicated to us that Democrats had made some clear gains in the wake of Dobbs v. Jackson, and now, they tell us that Republicans may be seeing the expected shift in their direction as the election nears. If we take the average shift between the last September poll and the most recent October poll from 10 nonpartisan pollsters, the picture is reinforced: Republicans have gained around two percentage points on the generic ballot of late.
It’s important, of course, to note that the Democrats haven’t been losing votes so much as Republicans have simply been consolidating undecideds. The Democratic vote share in the FiveThirtyEight averages is 45.2%, which is actually 0.1% higher than their vote share they had on September 22, when their lead was at its widest (D+1.9). To some degree, this might be expected — polling suggested most of the undecideds were quite hostile to the president, and it’s almost unheard of for an unpopular president’s party to gain among them as the election nears. But other flags suggest that part of this also has to do with recent economic developments.
A Washington Post analysis from Philip Bump suggests that presidential approval and vote share track gas prices fairly closely, and data from fuel price tracker GasBuddy shows that the price of gas had gone up by roughly 40 cents in the two weeks since the Democratic peak on the generic ballot. Unsurprisingly, economic anxiety has also now begun dominating the concerns of the American electorate, and this is an issue on which the GOP benefits by being both out of power congressionally and being the more trusted party overall.
It’s too soon to confidently say if this Republican edge is locked in. Traditional midterm fundamentals suggest that it might be, and our prior is indeed to suggest that November will result in Republican gains. If the election was held today, we would predict Republicans winning the generic ballot by about a point or so, and it is extremely unusual for the president’s party to retake the lead as the day of the election approaches. But there are always possible exceptions to consider. GasBuddy charts suggest that the price of gas is falling, and recent moves by the Biden administration suggest that they are acutely aware of it. CNN analysis also suggests that gas will continue to fall, which may help Democrats overall. But there is also a lag between the price dip and a rise in approval, and so the counterargument would be that it may be too little, too late for Democrats, especially with so much of their vote being banked in the early vote.
The other point to make in favor of Democrats is the bounce they’ve enjoyed in turnout since Dobbs v. Jackson, which gave them a shot in the arm with base turnout and a significant bulwark of secular pro-choice voters. This is the type of boost that carried them to victory in special elections for New York’s 19th congressional district and Alaska’s At-Large congressional seat while coming close in the New York’s 23rd, Minnesota’s 1st, and Nebraska’s 1st congressional districts, all of which were handily won by Trump in 2020 and yet saw leftward swings. If this holds, Democrats could enjoy a better-than-expected night, and if the gains are especially concentrated in certain swing states and suburban seats, it could see them possibly gain a seat in the Senate while having a shot at keeping the House.
This is an unlikely scenario, however, and shouldn’t be the default. The electoral boost Dobbs has provided Democrats in turnout and among independents likely insulates them from a 2010-level catastrophe that they very well may have faced otherwise. But it is probably not enough to make them win the popular vote; instead, it makes them lose it by less and avoids the electoral wipeout that loomed before the ruling.
One argument that we have repeatedly heard Republicans make is that a GOP victory is virtually guaranteed this year because of prior polling errors. This, however, is not one that would be backed by empirical evidence. Polling error is historically not correlated cycle-to-cycle, and polls missing in one direction for one cycle does not mean they’ll continue to miss in that direction in future cycles. Long-time political watchers may remember 2014, when Democrats had staked their arguments for keeping a majority on polls undercounting Democrats as they did in 2010 and 2012, only for polls to miss low on Republicans instead. If it was easy enough to simply add points to Republicans, pollsters would just be doing that.
Moreover, there are reasons to think that polls are better this year. In 2020, one of the main reasons polls missed was that they simply failed to reach a representative sample and missed counting many Republicans. Enough ink has been spilled on the 2020 misses, and we will save our thoughts on that for another time. This cycle, however, there is evidence that things may not be the same.
Take, for example, the difference in response environments (the people who answer a poll) between 2020 and 2022. Nate Cohn noted that while in October 2020, white Democrats were 23 percent more likely to respond to polls than white Republicans were, this gap was only 5 percent in September 2022. Moreover, we’ve already seen some good examples of overly-Republican polls this year missing badly, thanks to the post-Dobbs special elections. Several polls in the specials handily underestimated Democrats quite substantially. In New York’s 19th district, for example, not a single poll had the race closer than a 3-point Molinaro win, but Ryan ended up winning by 2.
This is not to say that Democrats will beat their polls in November; we’re not banking on that. But it is simply making the point that assuming a GOP polling overperformance based on past polls is not locked in and may be misguided. Modeling electorates is hard, and there isn’t yet any real correlation across many cycles to suggest one party will continue to be disproportionately overestimated in them. Polls of registered voters show a very close race that is Democratic on the aggregate, while polls among likely voters generally show a more Republican edge. If polls struggle to model Democratic turnout properly, as they did in the specials, Democrats could conceivably beat their polls too. Alternatively, if polls cannot reach low-trust, low-propensity Republicans, especially in states like Ohio, we might see a significant Republican overperformance. There is simply no real way to tell just yet.
We would probably favor Democrats to keep the Senate narrowly at the moment while comfortably favoring Republicans to take the House. Elections are extremely noisy, and every cycle provides us with a trove of data that will be overanalyzed and overfitted on for decades to come. If predicting the future was that easy, everyone would be really good at it. All we can say is that for now, Republicans have made some serious gains, and while there are small rays of hopes for Democrats trying to swim against the tide in keeping a trifecta, they’ve got large amounts of ground to make up.
Ed Note, 10/21 9:53 AM: The table in this post was updated to have the correct generic ballot numbers for the NY Times/Siena poll
I’m a software engineer and a computer scientist (UC Berkeley class of 2019 BA, class of 2020 MS) who has an interest in machine learning, politics, and electoral data. I’m a partner at Split Ticket, handle our Senate races, and make many kinds of electoral models.
Leave a Reply