Nov 3, 2026 · 53 min
Nate Silver tests the limits of election forecasting
An Election Day Special With Nate Silver
The discussion examines how much confidence voters, campaigns, and markets should place in polls, prediction markets, and AI-assisted models.
- 1Trump’s unpopularity creates Republican headwinds, but turnout and correlated polling errors could still reshape the midterm picture.
- 2Synthetic respondents may reproduce assumptions rather than reveal voter intent, making turnout especially difficult for AI polling systems to predict.
- 3Michigan and Kansas races could signal whether Democrats favor progressive ambitions or moderation suited to local electorates.
Don't miss
Nate Silver explains why AI-generated digital twins could reproduce assumptions rather than improve polling, especially when turnout is uncertain.
The brief
Nate Silver joins Joe Weisenthal and Tracy Alloway to assess a favorable Democratic midterm environment, Republican exposure to Donald Trump’s unpopularity, and the uncertainty that turnout can introduce.
Silver argues that polling errors are correlated and that former Trump voters might stay home rather than switch parties, complicating assumptions that polls systematically overestimate Democrats.
The hosts test whether AI-generated digital twins could replace surveys. Silver says synthetic respondents reproduce model assumptions and stereotypes, while turnout remains particularly hard to forecast.
Gerrymandering, advertising, and prediction markets each offer imperfect signals. Silver sees diminishing returns from ads and treats markets as useful but limited mixtures of polls, sentiment, and judgment.
Michigan and Kansas become tests of the Democratic Party’s ideological direction, while generative AI emerges as a tool for auditing code and accelerating models—not replacing human expertise.
What was said on this episode
35 statements · 15 positive · 14 negative · 6 neutral
Politics increasingly affects markets and the economy.
“politics matter, they're probably going to keep mattering more and more.”
Listen at 3:11
Trump’s approval rating is approximately 37%.
“Trump's approval rating is at about 37%.”
Listen at 4:12
Democrats will probably win the House and gain governorships and legislatures.
“they'll probably win the House. they'll probably pick up some governorships and state legislatures”
Listen at 5:05
Democrats are expected to win the Senate; failure would discredit pollsters.
“it would be a really bad night for pollsters if Democrats don't win the Senate.”
Listen at 5:56
Trump remains central to the midterm election.
“he's the central character”
Listen at 7:18
Polling errors tend to be correlated across similar states.
“polling errors are correlated”
Listen at 7:53
Polls underestimated Trump in 2016, 2020, and slightly in 2024.
“we have had systematic polling errors that underestimated Trump in 2016, 2020. 2024, a little bit more modest, but still direction the same”
Listen at 8:33
Trump gained many young male voters in the 2024 election.
“Trump got a lot of young men last year who flipped to 2024”
Listen at 9:32
Early voting indicators appear favorable for Democrats.
“They seem to be seeing... Good numbers for Democrats, better numbers in the polls.”
Listen at 10:24
Trump’s broken promises on gas prices and wars contribute to his unpopularity.
“the basic story is that he has broken promises, most prominently on cheaper gas and on no more stupid wars.”
Listen at 11:33
A lame-duck president’s political support can collapse.
“when you're a lame duck, the bottom can drop out potentially.”
Listen at 12:24
Silver’s model gives Democrats roughly 70% or higher Senate-win probability.
“in our model, they're at 70 high 70s now to win the Senate”
Listen at 12:57
Democrats lead by approximately nine points on the generic congressional ballot.
“If the national environment is as polls say it is, Democrats are ahead by about nine points on the generic ballot”
Listen at 14:53
The national environment creates a substantial disadvantage for Republicans.
“there's too much of a headwind for Republicans”
Listen at 15:05
The University of Michigan survey’s online redesign overrepresents Democrats.
“the University of Michigan survey changed the composition to online in a way that just has like way too many Democrats in the sample.”
Listen at 20:40
Pollsters have incentives that discourage systematic polling errors.
“pollsters are rational actors. And therefore, although poles are imprecise instruments, there is a lot of incentive not to systematically miss.”
Listen at 23:00
AI agents cannot replace polling as a source of original election data.
“polling cannot be outsourced to AI agents”
Listen at 24:25
Synthetic Hispanic respondents generated by models exhibit stereotyped preferences.
“99% of their synthetic. Hispanic respondents love soccer.”
Listen at 24:43
Different racial turnout rates could shift South Carolina results by about ten points.
“that shifts the top lines by like 10 points”
Listen at 26:14
South Carolina, Louisiana, and Mississippi are relatively insensitive to national political shifts.
“South Carolina, but especially like Louisiana, Mississippi have very low elasticity.”
Listen at 27:14
A strong Democratic performance could make Florida’s new map yield more Democratic seats.
“if Democrats kind of run the table in Florida, then they could actually wind up with more seats than they would have under the old map.”
Listen at 28:07
Silver’s model assigns a 5% chance that Republicans win the House.
“There is a 5% chance”
Listen at 28:49
Democrats could represent nearly two-thirds of the country through governorships.
“we could wind up with almost two thirds of the country being represented by a Democrat.”
Listen at 29:41
Silver would not bet against Greg Abbott in the Texas governor’s race.
“I don't think I'd bet against Abbott given the odds that Flipper gives.”
Listen at 29:58
AI could become a major area of congressional action before 2028.
“the one policy area where you could see action before 28 is on AI”
Listen at 30:26
Excessive campaign spending and mail can backfire electorally.
“it wound up backfiring a little bit”
Listen at 35:41
Kansas has become somewhat more Democratic politically.
“Kansas has low key gotten a bit more democratic.”
Listen at 37:34
Roger Marshall’s medical billing practices are politically unflattering.
“Roger Marshall has these. medical billing practices that are very unflattering.”
Listen at 37:45
Election models understate Florida’s Republican strength because of in-migration.
“Missing low on Florida because you have a steady flow of red voters into the state.”
Listen at 39:47
Prediction markets attract a clientele that leans slightly Republican.
“I think the prediction markets have a slightly Republican-leaning clientele”
Listen at 41:06
Prediction markets incorporate scandals faster than polling does.
“if there's a scandal, the New York Times reports on Thursday morning, then it will take several days to show up in the polling”
Listen at 42:28
Prediction markets would perform extremely poorly without access to polls.
“my expectation is that prediction markets would be extremely terrible.”
Listen at 43:02
Silver’s model is 0.995 correlated with Polymarket on individual states.
“we are like literally like 0.995 correlated with polymarket on individual states.”
Listen at 43:41
AI can code complex NFL playoff tiebreakers in about 15 minutes.
“they do it in 15 minutes while you're at the gym”
Listen at 45:54
AI-generated code for the NFL model runs ten times faster.
“its code is 10 times faster.”
Listen at 45:59
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.