An Election Day Special With Nate Silver

A Bloomberg Tech Minute reports that getting hired at SpaceX can involve Elon Musk reviewing a candidate’s résumé. Lauren Grush of Bloomberg Businessweek says the company culture is defined by Musk, and applicants must show loyalty to him, concrete achievements and hands-on experience; recruiters value raw, innate intelligence more than advanced degrees. Recruiters, advisers and interviewees say the space-exploration mission attracts candidates, while Musk repeatedly invokes that dream. Job ads warn that employees must work extended hours and weekends as needed. The major compensation perk is equity, either share-purchase rights or restricted stock units that vest over time.

Joe Weisenthal and Tracy Alloway open by revisiting Joe’s long-standing preference not to be political reporters. They say the distinction is becoming impossible because politics increasingly determines markets and the economy, a conclusion reinforced by their earlier discussion with James Robinson in Chicago. Recording on October 29 for a November 3 Election Day special, they introduce Nate Silver, founder of Silver Bulletin and the former FiveThirtyEight figure, to assess President Trump’s unpopularity, the midterms and what polling can actually tell them.

Nate Silver puts Trump’s approval at about 37%, possibly the worst rating for a president at a midterm since Harry Truman in 1946. He contrasts that with 2018, when Trump was somewhat less unpopular and Republicans lost 41 House seats. Democrats have favorable historical midterm trends, economic anxiety, opposition to the Iran war, a lame-duck president and strong early voting, plus good 2025 results in New Jersey and Virginia. But winning the Senate requires improbable victories in states such as Texas, Kansas, Alaska and South Carolina. Silver says Democrats likely win the House and some governorships, but models must take a probabilistic leap of faith on those red states.

The hosts ask whether the election is truly a referendum on Trump or instead a collection of local races. Silver’s strongest evidence that Trump is the central character is that Republican campaigns have stopped featuring him in advertisements and podcasts, behavior suggesting candidates want distance. Trump, however, keeps trying to make himself central. Silver says Ken Paxton’s Texas endorsement and Trump’s decision to go to war with Iran may have helped Democrats, and that without them the Senate might be a toss-up rather than lean Democratic. Trump’s 37% approval consists of only about 19% strong approval, creating permission for Republicans to break away before 2028.

Silver rejects the popular claim that pollsters have consistently overestimated Democrats since 2016. Poll errors correlate across similar states, though less strongly in midterms because different candidates appear on each ballot. Trump was underestimated in 2016, 2020 and, more modestly, 2024, but that pattern did not occur when he was absent from the ballot: Democrats were massively underestimated in the 2025 New Jersey and Virginia governor races. Silver’s explanation is that Trump may mobilize low-turnout voters who are hard to reach and get screened out as unlikely voters, while those voters may not turn out for candidates such as Roger Marshall or Ken Paxton.

Silver sees comparatively strong Democratic early voting, including turnout among Black voters concerned about disenfranchisement and among people who voted in no more than one of the previous four elections. He sees little Democratic-to-Republican crossover, with perhaps some movement in the opposite direction. Asked why consumer sentiment is poor despite solid spending and economic data, he says the sentiment surveys themselves may be unreliable. He points to the University of Michigan’s shift online and a sample he believes contains too many Democrats. Unlike election polls, consumer surveys receive no election-night reality check, so their pessimism may reflect methodology rather than an economic contradiction.

The hosts propose using AI-generated replicas of different Americans to solve declining survey response rates. Silver says AI agents cannot substitute for polling because polls remain one of the few sources of ground truth, and synthetic respondents reproduce stereotypes: he says 99% of an AI-generated Hispanic sample might love soccer. He also rejects overly broad demographic categories such as “the Hispanic vote,” which hides differences among younger and older voters, recent immigrants, Cuban Americans, Puerto Ricans, Mexican Americans and Dominicans. In South Carolina, he illustrates why turnout is hard to model: if Black voters turned out at 2024 rates while white voters turned out at 2022 rates, the topline could shift by about ten points.

Silver explains that a gerrymander can become a “dummymander” if it produces fewer seats than the previous map. Florida’s map, drawn by Ron DeSantis for a normal presidential-year environment, may be poorly suited to a Democratic wave: if Democrats run the table, they could win more seats than under the old map. Texas Republicans similarly assumed South Texas Hispanics had become reliably Republican rather than remaining swing voters, potentially leaving Democratic incumbents stronger than expected. Silver says mapmakers had to accept risk to maximize seats, and that gerrymanders become much more useful for Republicans only if the polling environment shifts less blue; his model puts the House loss risk for Republicans at about 5%.

Silver says governor’s races receive too little attention and projects that Democrats could eventually control the governorships representing nearly two-thirds of Americans. Greg Abbott’s Texas race matters partly because he has been unusually supportive of data centers, though even he is less so now. For the Democratic Party’s direction, Silver highlights Michigan, where progressive Abdul El-Sayed beat the more moderate Haley Stevens in the primary. A strong El-Sayed win would aid progressives; a narrow or underperforming victory would fuel criticism. Kansas offers the contrasting test: Adam Hamilton’s moderate message of cheaper gas and integrity could defeat Roger Marshall, allowing centrists to claim the nomination strategy worked.

Silver says prediction markets combine polls, sentiment and faster reactions to scandals, but their Republican-leaning clientele and occasional overconfidence can distort prices. His Flipper model and Polymarket were about 99.5% correlated in individual states, although his overall Democratic Senate probability was higher; he says a fair test would prevent markets from seeing models. He also describes AI as dramatically improving his work: GPT or Claude can code complex NFL playoff tiebreakers in roughly 15 minutes instead of two and a half days and produce faster code, though they can introduce hidden bugs. He says AI is most valuable when paired with his own modeling expertise and careful review.