The good, bad and ugly outcomes in the Senate

Politics

The good, bad and ugly outcomes in the Senate

Can Democrats still win the Senate if the polls overestimate them? And how ugly could the map get for Republicans if the polls are wrong in the other direction?

Nate Silver

Oct 17, 2026

∙

Paid

77

2

Share

Troy Jackson shakes hands with Susan Collins before their Oct. 15 debate. Based on current polling averages, Maine would be the tipping-point state. Charly Triballeau/AFP via Getty Images.

Our standing rule at Silver Bulletin: Always account for the possibility of

systematic polling errors

. What if the polls are off by 3-4 points in the same direction in every state? I’m choosing 3-4 points since that’s basically what we refer to as a “

normal-sized polling error

” around these parts. Systematic polling errors in this range are fairly common; that’s roughly the margin by which

polls underestimated Trump

in 2016, 2020 and 2024. There can be larger errors in individual states, and larger systematic errors are possible too, but those fall more in the tails than in the thicker part of the probability distribution.

Since polling error is

basically the critical question

at this point, we’ll probably attempt at least one more big-picture analysis of it before Nov. 3. One point that sometimes gets lost, that my previous newsletter

didn’t really address

, is that pollsters are very aware of the problem. So in predicting polling error, in some sense you’re predicting the behavior of human beings and how they adjust to changing conditions in the polling landscape.

But leave that philosophical question aside for now — because after a long week of travel,

1

I could use a simpler Model Talk. I just want to take that “what if polls are off by 3-4 points in every state” question more literally.

We’ll take two approaches into this, the first of which is super simple: just looking at our table of current polling averages. Then something slightly more complicated. We’ll take today’s batch of 40,000 FLIPR Deluxe simulations and sort them into three buckets: “no systematic national bias,” “polls overrate Democrats,” and “polls overrate Republicans.” We can answer a few important questions alongside this:

Are we at the point where Democrats are guaranteed to pick up the Senate

unless

there’s a systematic polling error?

How much capacity do Democrats have to absorb polls overrating them again?

And how bad could the map get for Republicans if polls

underrate

Democrats instead? Yes, that can actually happen, as they

did in New Jersey and Virginia last year

.

Democrats might survive a modest polling error

Here are FLIPR’s

current polling averages

in all Senate races that are polling within the single digits. They include both the states that Democrats are targeting

and the ones they’re defending

. For purposes of flipping the Senate, this distinction basically doesn’t matter. If Democrats pick up four GOP-held seats but lose Michigan, they’ll get stuck on the same result as if they pick up three seats but hold Michigan: a 50-50 Senate where JD Vance would cast the tie-breaking vote.

Before I show you the data, let me warn you that the polling averages are a little thin in some of these states. The more advanced versions of FLIPR heavily

hedge toward the “fundamentals” in these cases

and don’t necessarily take the polling averages at face value. They would take the “over” on Republican performance in South Dakota, for instance — although the fact that we’re even talking about South Dakota tells you something about how red the terrain is. I would also note that we’re slightly cheating in Alaska by looking at the margin before ranked-choice voting reallocations, which could hurt Mary Peltola.

2

If the current FLIPR polling averages are

exactly

right — they won’t be, but please humor me — Democrats would pick up seven Republican seats while holding all of their own, producing a D 54-R 46 map.

It is worth noting, though, that there are eight states where Democrats currently hold leads of between 1 point and 4.3 points. There are lots of states that just tilt or lean toward them, in other words.

3

There are few “Tilt R” states, however. This looks a lot like the presidential maps of 2012 and 2016, when Barack Obama and Hillary Clinton respectively held narrow leads in most swing states. Needless to say, one of those elections turned out well for Democrats and the other didn’t. It’s

not

like 2020 when Biden had more of a cushion (which it turned out he

very much needed

).

So let’s stress-test that lead. What if the polls overrate Democrats by 2 points in every state? Well then, Kansas and Iowa stay Republican. But James Talarico would prevail by 0.6 points, and Troy Jackson would hold on by 0.7 points in Maine.

4

Democrats still take the Senate 52-48.

With a 3-point polling error, though, Maine and Texas turn into narrow GOP wins, and Democrats would top out at 50 seats. And with a 4-point polling error, they’d also fail to pick up Alaska and lose Michigan, producing a 52-48 GOP Senate.

Still, you’d always rather have a narrow polling lead than a narrow polling deficit. Democrats’ situation is more robust than if you just treat all of the states in the middle of the table as pure toss-ups. Peltola, Talarico and Sherrod Brown, in particular, have led in the vast majority of polls, and those pickups plus North Carolina would suffice to give Democrats a 51-49 winning map. That helps explain why FLIPR gives Democrats around a 3-in-4 chance of flipping the Senate rather than 50-50.

This analysis leaves a lot out, though.

It’s fair to say that FLIPR is

mostly

riding-or-dying with the polls, but not completely so. FLIPR Classic incorporates fundamentals, and FLIPR Deluxe also adds

expert ratings

. The probabilities you see in FLIPR are based on the errors it estimates in these blended

forecasts

and not just the error associated with the polling averages on their own.

5

Also, the sort of uniform, national polling error that I described above

isn’t the only kind of error

that FLIPR’s simulations account for. There’s variation in individual states and districts above and beyond the national trend.

Maine is a hard state to forecast, for instance. You’d figure it would be very blue in an environment like this one, but Susan Collins has

outperformed her polls before

. It wouldn’t be shocking if Democrats have a pretty good night overall but Collins holds on. I

also

wouldn’t be that surprised if Democrats have a mediocre night but Collins loses. That latter scenario is what you might get if the map adheres super closely to statewide partisanship instead of Democrats overperforming in red states, as

polls currently project

. Maybe they flip Maine and hold Michigan fairly easily, but states like Texas and Iowa and certainly Kansas prove to be just out of reach.

FLIPR also simulates the possibility of regional polling errors or polling errors among different demographic groups. In both 2018 and 2022, polls were relatively unbiased nationally, but there were some weird regional swings — like a big GOP overperformance in Florida in 2018 relative to expectations at the time

6

, or in New York state in 2022 during what was otherwise an underwhelming night for Republicans. One such scenario — a regionalized polling error against Democrats in the Midwest — has received outsized attention, understandably because that’s what we saw in 2016. FLIPR does consider exactly this scenario, although it doesn’t privilege it. It also considers possibilities like Democrats underperforming in the South, overperforming among Hispanics,

overperforming

rather than underperforming in the Midwest

7

, and so forth.

So let’s try this. We run 40,000 FLIPR simulations every day. In each sim, FLIPR draws a random number from a

Student’s

t

distribution

to simulate uniform, national polling error. It also draws a bunch of other random numbers, representing other types of correlated (e.g. a systematic error among Hispanics) and uncorrelated errors (e.g. an error localized to North Carolina’s 1st Congressional District).

I’ve divided our latest batch of FLIPR simulations into thirds based solely on that first random draw — uniform national forecast error:

Scenario 1. National error is modest: within roughly 1.3 points of FLIPR’s target.

Scenario 2. A uniform national error against Democrats. On average in this scenario, the forecast overestimates Democrats by 3.5 points, so it’s a pretty good reflection of the “normal polling error” scenario.

Scenario 3: A uniform national error against Republicans. On average in this scenario, FLIPR Deluxe forecasts overestimate Republicans by 3.5 points here.

Let’s see what each scenario looks like. I’ll give you FLIPR’s overall odds for House and Senate control under each one, along with the projected number of seats gained or lost. And I’ll show you the three Senate maps, because maps are fun.

Scenario 1: Forecasts are basically right nationally — but there can be local and regional errors

This post is for paid subscribers

Subscribe

Already a paid subscriber?

Sign in

Previous