Predicting elections is getting harder. What did the midterms teach us?

Summary of Predicting elections is getting harder. What did the midterms teach us?

by NPR

12m•November 5, 2026
Listen to Episode

Overview of Predicting elections is getting harder. What did the midterms teach us?

This NPR Short Wave episode examines how election forecasting performed in the latest midterms, using a conversation with political scientist Jack Rommetta of George Mason University. The episode explores why models sometimes miss margins even when they correctly call winners, how polling has evolved after past failures, and what prediction markets mean for forecasting and electoral behavior. The main takeaway: this cycle, polling and forecasting worked better than in recent years, but elections are still harder to predict than many people assume.

Key Takeaways

  • Forecasts were mostly directionally correct, but not perfectly calibrated.
    • Rommetta’s model correctly called many competitive Senate races.
    • However, the model and many others missed the size of Democratic overperformance in some contests.
  • Polling appears to have been more accurate this cycle than in 2016.
    • Pollsters have improved methods to deal with nonresponse and biased samples.
    • Still, polling remains imperfect and may not stay accurate in future cycles.
  • Election forecasting is a blend of science and judgment.
    • Models combine:
      • fundamentals (state partisanship, incumbency, prior results)
      • polling (more recent but potentially biased information)
    • Deciding how much weight to give each is partly subjective.
  • The political environment looked highly nationalized.
    • Senate races seemed more influenced by the broader national mood than by local quirks.
    • That may reduce how much candidates can separate themselves from national trends.
  • Prediction markets may be useful, but they can also influence the race.
    • They can aggregate information well.
    • But there is concern that campaigns and candidates may respond to market signals, which can affect behavior.

What the Forecasting Model Got Right and Wrong

What it got right

  • Correctly predicted several competitive races.
  • Kept outcomes within a reasonable range of predicted possibilities.
  • Produced a solid overall estimate that Democrats had a good chance of retaking the Senate.

Where it missed

  • It underestimated Democratic margins.
  • It missed some individual races, including:
    • Mississippi
    • Iowa
  • It also expected Kansas differently than the final outcome.

How Election Forecasting Works

Rommetta explains that forecasting generally relies on two inputs:

1. Fundamentals-based modeling

This approach uses more stable, historical factors such as:

  • partisanship of the state
  • incumbency advantage
  • past election performance
  • whether a race is open or has an incumbent

2. Polling-based modeling

This approach uses:

  • current surveys
  • recent voter sentiment
  • the latest public data before Election Day

Why blending them is hard

  • Polls are recent, but can be wrong.
  • Fundamentals are stable, but can miss sudden shifts.
  • Forecasters must decide how much to trust each source, especially when the national environment is moving quickly.

Polling: Fixed or Still Fragile?

The episode draws a contrast between 2016, when polling was widely seen as flawed, and the current midterm cycle, when many polls performed better.

Why polls improved

  • Pollsters are adjusting for:
    • declining response rates
    • sample bias
    • underrepresented voter groups
  • Modern polling often uses more sophisticated weighting and adjustment methods.

Ongoing concern

  • Response rates have fallen dramatically.
  • People who answer surveys may not resemble the broader electorate.
  • So while polling worked well this cycle, it is not guaranteed to keep working.

Prediction Markets: Useful, but Worth Watching Carefully

The conversation also covers prediction markets like Kalshi.

Why they matter

  • They can be effective at aggregating information.
  • Research suggests they often provide useful forecasts.

Why they worry Rommetta

  • Markets may influence candidates’ and campaigns’ decisions.
  • In some primaries, candidates appeared to react to market odds.
  • If markets are partly driven by bad data or fraudulent polling, they can amplify misleading signals.

Main Insight About American Politics

Rommetta’s broader takeaway is that U.S. elections are looking increasingly nationalized.

What that means

  • Local and candidate-specific factors may matter less than the overall national mood.
  • Senate races in particular may be more exposed to national trends than in the past.
  • This makes forecasting somewhat more straightforward in one sense, but also highlights how little room candidates may have to shape their own outcomes.

Notable Quote

“There isn’t like one correct forecasting model.”

This captures one of the episode’s central themes: forecasting is useful, but not absolute. Reasonable analysts can build different models and still reach different conclusions.

Bottom Line

The episode argues that:

  • forecasting is valuable but inherently uncertain
  • polling improved in this cycle
  • models still need refinement
  • national political forces may be overpowering local differences more than before

In short, the midterms taught forecasters that they are getting better at predicting elections—but also that elections remain dynamic, probabilistic, and increasingly nationalized.