
Nov 12, 2026 · 26 min
Pollsters confront a changing Democratic electorate
How To Poll A Wave, With Nate Cohn
The episode shows how turnout models can miss rapid coalition shifts—and how pollsters are trying to correct for them.
- 1Southern polls underestimated unexpectedly strong Black and Latino turnout in Florida, South Carolina, and Texas.
- 2Historical turnout weighting can lag when electorates change, while recalled-vote weighting can distort results in shifting states.
- 3Nate Cohn describes “synthetic past vote,” a revised method designed to address both kinds of error.
Don't miss
Nate Cohn explains how “synthetic past vote” could improve polling in both rapidly changing states and white working-class states.
The brief
The 2026 Democratic victory was large enough to invite historic comparisons, but the sharper question is whether polling should have seen it coming.
Nate Cohn says Southern polls missed unexpectedly strong Black and Latino turnout, while Upper Midwest errors remain harder to diagnose because turnout data are weaker.
The episode examines a central polling dilemma: historical turnout models can lag behind coalition shifts, but heavy reliance on recalled past votes can also badly distort results.
Cohn details “synthetic past vote,” a revised weighting method that corrects for false claims of voting and missing reports of past vote.
The broader lesson is methodological: pollsters need models that recognize demographic change without allowing assumptions about participation to overwhelm the data.
What was said on this episode
22 statements · 8 positive · 11 negative · 1 mixed · 2 neutral
The 2026 election was the largest presidential repudiation and landslide in half a century.
“This was an extraordinary repudiation of the president. It's the biggest landslide that we've seen in half century.”
Listen at 1:35
Democrats were likely to win seven Senate states Trump carried by over ten points.
“The Democrats are probably going to win, what, seven states that voted for Trump by more than 10 points in the Senate.”
Listen at 1:58
Voter dissatisfaction with the president drove the Democratic victory.
“And the explanation is simple. Voters are fed up with the president. We're trying to put a check on him.”
Listen at 2:18
Democrats are more likely than Republicans to respond to Times polls.
“Democrats are much likelier to respond to our polls than Republicans, which we've seen before.”
Listen at 5:09
Democratic turnout was expected to exceed the polling model's assumptions.
“We could also see before the election that the Democratic turnout was just going to be better than our polls supposed.”
Listen at 5:17
Cohn considered it possible that Times polls would underestimate Democrats.
“I definitely thought it was possible that we could underestimate the Democrats.”
Listen at 5:42
Overestimating Democrats would have been more consequential than underestimating them.
“the cost of overestimating the Democrats, if we had done so, would have greatly outweighed the risk of underestimating the Democrats.”
Listen at 6:41
Black and Latino turnout exceeded Times polling expectations in Southern states.
“The turnout among black and Latino voters in particular was higher than we expected.”
Listen at 9:27
Historical turnout inputs can pull polling models toward outdated electorates.
“when the electorate really changes, that incorporation of how voters have behaved in the past can drag our electorate a little bit towards towards what's happened in the past.”
Listen at 9:49
Latino voters likely comprised a larger Texas electorate share than in the prior presidential election.
“The Latino share of the electorate in Texas is probably higher than it was in the last presidential election.”
Listen at 10:34
Determining Michigan and Minnesota voters may take months.
“It may be months until we know who voted in Michigan and Minnesota.”
Listen at 10:55
Michigan polling errors may reflect nonresponse bias that underestimates Republicans.
“Maybe we'll find that. Maybe we'll think it's, you know, some sort of non-response bias reminiscent of previous elections where the polls have underestimated Republicans.”
Listen at 11:57
The Times turnout model relies heavily on historical data and carries risks.
“Our turnout model depends a lot on history. That is just empirically what works best in our data. It comes with risks.”
Listen at 15:22
Historical reliance prevents the turnout model from fully adapting to electorate changes.
“Our model gets us like halfway to where we would have liked it to be, but it doesn't get us all the way there because we're still relying on history to some extent.”
Listen at 15:37
Times polling has a strong record of detecting major demographic shifts.
“We, I think, have a very good track record, actually, of detecting the major demographic shifts.”
Listen at 16:19
Recall-vote weighting does not improve Times polling in backtests.
“when we try and use it on our election, our previous data, it doesn't help.”
Listen at 18:48
Recall-vote weighting worsened Times polls in 2020.
“In 2020, our polls do not get better with recall vote. They get worse.”
Listen at 19:16
Recall-vote weighting would have incorrectly shown Colin Allred defeating Ted Cruz in Texas.
“we had a poll in Texas in 2024 where we would have shown Colin Allred winning and beating Ted Cruz if we had waited on a recall vote.”
Listen at 20:22
Pollsters should not weight nonvoters who falsely claim they voted.
“there are voters who didn't vote in the last election who tell us that they did and that they didn't. So we shouldn't be waiting on them.”
Listen at 21:33
Synthetic past-vote adjustments improve some polling problems.
“Adjusting for those two things really seems to help with some of the problems I talked about.”
Listen at 22:25
The current weighting method correctly reproduced the 2024 Florida result.
“with the weighting technique we're using now, we get the right result in Florida in 2024”
Listen at 22:31
Cohn says the new method would have produced publishable 2024 results.
“we would have been really proud to release the results that we had with this new method for 2024”
Listen at 23:15
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.