This week, a Siena College/New York Times poll showed President Biden with just a 33 percent approval rating, a result so poor that it touched off speculation — including from yours truly — about whether he would even run again in 2024. The Siena/New York Times number is on the low end of the polling consensus, but Biden’s approval rating in our polling average — about 39 percent24 — is nonetheless a historically low number.

And yet, the same poll showed a neck-and-neck race for Congress. Democrats led by 1 percentage point among registered voters25 on the question of which party voters preferred controlled Congress and trailed by 1 point among likely voters.

What to make of this seeming divergence? How much does the president’s approval rating actually matter for predicting congressional outcomes?

From a zoomed-out perspective, the answer is that there’s a reasonably strong relationship. If you knew nothing else about the race for Congress, you’d expect an unpopular president’s party to lose seats. And indeed, that’s likely what will happen this year, too. Republicans are an 87 percent favorite to take over control of the House, according to the Deluxe version of our forecast. The Senate remains much closer to a toss-up, but that has more to do with poor Republican candidates than anything the Democrats are doing well.

Watch: https://abcnews.go.com/fivethirtyeight/video/democrats-continue-win-georgia-2022-fivethirtyeight-politics-podcast-86847505

But predicting the number of seats lost in Congress based on the president’s approval rating is not quite the question we’re interested in from a forecasting standpoint. Rather, we want to know how much the president’s approval rating matters given all the other information we have about the race. In other words, is Biden’s poor approval already “baked in” to the congressional generic ballot polls and polls of individual House and Senate races? Or is there reason to think that the Democrats’ standing will get worse between now and November?

The statistical answer is that it’s mostly baked in. Warning: The following paragraphs are going to be a bit technical. If you want more of an intuitive answer, skip ahead to the bolded bullet points below.

The way our model addresses this is by looking at every congressional race since 1990 and evaluating how predictable the movement in the generic ballot has been based on baseline conditions, which we sometimes refer to as the “fundamentals.” Specifically, the factors it looks at include the president’s approval rating, the result of the previous election for Congress, whether or not the election is a midterm and the degree of political polarization. (Times of high political polarization — like now — will tend to produce less dramatic swings in races for Congress because there are fewer swing voters.)

Watch: https://abcnews.go.com/ThisWeek/video/buy-dems-energizing-abortion-evens-midterms-equation-silver-86558599

Right now, those “fundamentals” expect Democrats to eventually lose the House popular vote by about 8 points, which would be an awful result for the party and would very likely result in its loss of both chambers of Congress. By comparison, if Biden had a breakeven approval rating instead of being about 17 points underwater, the “fundamentals” would predict Democrats to lose the popular vote by around 4.5 points, which would still mean almost certain doom in the House but might be enough for them to save the Senate.

However, the model also weighs those “fundamentals” against the current state of affairs. Right now, Democrats trail in our generic congressional ballot polling average — a proxy for the House popular vote — by about 2 points. But that’s actually more like a 4-point deficit among likely voters, since Republicans are likely to have a turnout advantage in November. Our model accounts for this, but the model also accounts for factors aside from the generic ballot in forecasting the House popular vote, and when we take into consideration those factors, our model predicts Democrats to lose the popular vote by almost 6 points, not that far from what the “fundamentals” show.

Watch: https://abcnews.go.com/fivethirtyeight/video/democrats-supreme-court-fivethirtyeight-politics-podcast-86637136

Even if there were a bigger gap, though, the “fundamentals” ultimately do not get all that much weight in the model. The reason is simply that, even at this fairly early point in the cycle, the generic ballot (at least if you properly adjust it to account for likely voters) and other indicators directly related to the current election have historically been more reliable predictors than the “fundamentals.” The model does expect conditions to get a bit worse for Democrats, but really just a bit.26

So what is the intuition behind this? Here are a few factors to keep in mind:

1. Voters have good reasons to disapprove of Biden without wanting Republicans in Congress

When your approval rating has fallen into the 30s, you’ve not only lost the confidence of most swing voters but also some members of your own party. The Siena/New York Times poll, for instance, showed Biden with only a 70 percent approval rating even among Democrats. However, 90 percent of Democrats in that same poll prefer Democratic control of Congress, compared to just 4 percent who want the GOP in charge.

One concern for Democrats is that those disaffected voters won’t turn out. Still, there’s no particular reason to expect them to vote Republican if they do. A lot of them think Biden is too old — a concern also shared by many independent voters — but that’s more a factor for 2024 than in congressional preferences for 2022.

And on many issues — from abortion to LGBTQ rights to the integrity of the 2020 vote — Republicans are adopting highly right-wing, partisan positions that have little appeal to swing voters and might even motivate otherwise disaffected Democrats to turn out. Parties generally pay a penalty for ideological extremism. In other words, although Democrats have also adopted unpopular left-wing positions on many issues, Republicans aren’t as poised to capitalize on a high inflation and poor electoral environment for Democrats as a more moderate, less Trumpian version of the party would be.

2. It’s usually best to trust a direct measure over an indirect one

This is just a good principle of statistical analysis. If you have a direct measure of the quantity that you’re interested in, there’s not much need for a proxy or an indirect one.

Suppose, for instance, that you’re trying to estimate the volume of home sales in — I don’t know — Indianapolis. You could imagine some clever ways to get at this. You could drive around town and count the number of “FOR SALE” signs. Or you could track the number of clicks on Zillow and other websites that list homes for sale. But all of that is beside the point because home sales can be directly measured, albeit with something of a lag until reports are compiled.

Likewise, if you’re interested in races for Congress, and you ask voters how they’re going to vote for Congress and also how they feel about the president, voters’ preference for Congress is the direct measure and the one that should be more reliable. It’s presumptuous, frankly, to suggest otherwise and to disbelieve a voter who says she disapproves of Biden but also wants Democrats to stay in charge of Congress.

Watch: https://abcnews.go.com/fivethirtyeight/video/political-consequences-overturning-roe-wade-85819633

3. Biden and Democrats weren’t that popular to begin with

In the national exit poll in November 2020, 52 percent had a favorable opinion of Biden and 46 percent had an unfavorable opinion. That’s considerably better than his numbers now, and Biden won a fairly comfortable victory in the popular vote. But, it also wasn’t the sort of sweeping mandate that, say, former President Barack Obama had in 2008, which was accompanied by approval and favorability numbers that initially soared into the 60s and 70s. Moreover, Democrats rode into Obama’s first term with 257 House seats, far more than the 222 they held after the 2020 election.

Part of the reason that the 2010 midterms were so awful for Democrats was because they had a long way to fall from being about as popular as a party probably could be in modern American politics. In 2022, Democrats don’t have that problem because they weren’t very popular to begin with. They barely held onto the House.

So while goodwill toward Biden may have been just enough to get him over the hump in 2020 — and a lot of that goodwill has now evaporated — conditions aren’t necessarily that different than they were two years ago. The major parties are both unpopular, there are few if any nationally beloved political figures and the country is highly polarized. What’s more, with unpopular former President Donald Trump potentially set to declare a 2024 bid soon, he could also be a factor in the race — maybe one that helps Democrats.

4. So far, presidential approval and the race for Congress have diverged, not converged

Finally, I’d note that if you had predicted some months ago that polls for Congress and Biden’s numbers would have converged toward one another, you would have been wrong. Since May 1, Biden’s approval rating has declined by about 9 points:

And yet, the generic ballot has been essentially unchanged:

Instead, as voters have gathered more information about the race, they have drawn more of a distinction between how they feel about Biden and what they’d like to see happen in Congress. Maybe this trend will reverse itself. But the “fundamentalists” — the analysts who think the races for Congress are predictable based on presidential approval and other baseline conditions — have been wrong so far.

CORRECTION (July 15, 2022, 11:06 a.m.): A previous version of this article calculated the change in Biden’s approval rating from May 1, 2021 — not May 1, 2022. That calculation has been updated to reflect the change in Biden’s approval since May 1, 2022.