Predicting elections is getting harder. What did the midterms teach us?
Hosted by Regina Barber
- 12m
- Transcript
Summary
Written by a model from the transcript; not yet reviewed.
Political scientists and forecasters face a difficult challenge predicting midterm election outcomes, balancing historical fundamentals with modern polling data. In this episode, host Regina Barber speaks with George Mason University political scientist Jack Remetta about how his Senate forecasting model performed following the midterms. They discuss the mechanics of blending polling with state-level data, the hurdle of plummeting response rates, and how prediction markets influence electoral behavior and candidate decisions.
- Forecasters blend historical fundamentals like partisan lean with recent polling data, though the exact mix is subjective. 7:16
- Traditional polling response rates have plummeted from double digits down to less than one percent in many modern surveys. 5:54
- Reasonable analysts can build different forecasting models that arrive at different answers without a single deterministic model existing. 10:33
- Prediction markets aggregate information well, but they can also influence candidates to enter or exit races. 10:59
Chapters
- 0:00Opening Sponsor Messages
The episode begins with a series of sponsor announcements and brief introductory remarks.
- 0:59Jack Remetta’s Forecast Setup
Jack Remetta, a political scientist, prepares his Senate forecasting model the night before the election.
- 1:40Modeling Process and Data Sources
The hosts explain how the forecast incorporates unemployment data, generic ballot, and Senate polls.
- 4:55Forecast Accuracy and Missed Races
Jack discusses which races were predicted correctly and where the model fell short, noting Democratic overperformance.
- 6:00Challenges of Election Forecasting
The conversation turns to the assumptions, incumbency effects, and national versus local influences in modeling.
- 7:02Polling vs Fundamentals in Models
Jack explains the balance between historical fundamentals and real‑time polling, and the issue of non‑response bias.
- 11:23Prediction Markets and Their Influence
The hosts explore how prediction markets aggregate information and the potential impact on candidate decisions.
- 13:00Nationalization of Senate Races
Jack reflects on how the 2026 midterms showed a nationalized trend, limiting candidates’ ability to distance themselves from national politics.
- 13:52Closing Remarks and Credits
The episode wraps up with congratulations, thanks, and production credits before the final sponsor message.
Highlights
- 0:54
Running the Final Senate Modelstory
The guest sits in his apartment the night before the midterms, uploading data from databases like FRED and the New York Times to run his final senate forecast.
- 4:08
Evaluating the Model Resultsinsight
The guest evaluates his model after the election, noting that it successfully predicted competitive races like Texas and Ohio while missing the exact margins.
- 7:02
Two Ways to Forecast Electionsstrategy
Forecasting relies on two primary approaches: examining historical fundamentals like state partisan lean or incorporating recent, though sometimes biased, polling data.
- 9:06
Plummeting Survey Response Ratesdata
Survey response rates have dropped from historical double digits of 20 to 50 percent down to less than 1 percent in modern online opt-in surveys.
- 10:14
Multiple Models Are Validinsight
Different academic forecasters can build valid models that yield varying predictions, demonstrating that there is no single correct forecasting model.
- 10:59
The Influence of Prediction Marketsdebate
While prediction markets effectively aggregate information, they can negatively influence early electoral behavior when candidates respond to markets driven by faulty data.
- 12:43
The Nationalization of Senate Racesinsight
The midterms demonstrated that senate races are increasingly driven by national tides rather than local or idiosyncratic state factors.
Quotes
“We called most of the competitive races correctly.”
Unnamed speaker · 4:55 “I think polling worked this cycle would be my claim.”
Unnamed speaker · 8:45 “There's a lot of evidence from economics, there are some big papers showing these sorts of things.”
Unnamed speaker · 11:35 “So this year, and I don't want to overstate it, but my takeaway from this year is, things are looking very nationalized.”
Unnamed speaker · 13:00
Sponsored segments
- Open Door 0:00–0:22
- Whole Foods Market 2:52–3:19
- Schwab 3:20–3:36
- Subaru 3:37–4:04
- Midi Health 14:29–14:44
- EasyCater 14:45–14:58
About this episode
Famously, polls got the 2016 election wrong. And they’d been off for multiple elections since. So, Monday night, political scientist Jack Rametta nervously refreshed his Senate forecasting model https://jackrametta.com/senate-forecast/ for the final time. Would it be right this time? Or would it yet again cast doubt on whether modeling, statistics and forecasting can still help us understand what motivates people to vote? Now that the results are in, so is Jack – to tell us how he did, and what his and other researchers’ predictions tell us about the evolving science of polling and forecasting. Interested in more science behind the news? Email us your question at shortwave@npr.org mailto:shortwave@npr.org. Support public media with NPR+ and enjoy perks for over 25 podcasts like this one. This show’s perks include sponsor-free listening. Learn more at plus.npr.org http://plus.npr.org. See pcm.adswizz.com https://pcm.adswizz.com for information about our collection and use of personal data for sponsorship and to manage your podcast sponsorship preferences. NPR Privacy Policy https://www.npr.org/about-npr/179878450/privacy-policy
Transcript
2,530 words · 4 speakers
Speaker names marked are inferred from speaking time and the show’s known hosts — they are not voice-verified.
Topics
- NPR
- University of California Davis
- George Mason University
- Harvard
- Brennan Center for Justice
- Kalshi
- Silver Bulletin
- New York Times
- Fred database
Page updated · transcript and summary are machine-generated