Presidential polling repeatedly fails, though rarely in the same way twice, because pollsters keep 'fighting the last war'
Source: "Why presidential election polling keeps failing, from 1936 to now | Fox News." October 30, 2026. www.foxnews.com
The Gist
Pollsters have been getting elections wrong for ninety years, and each time the mistake is a little different. They fix the last problem and then get blindsided by a new one. Today's mix of fewer people answering and partisan or even fake polls means the misses are likely to continue.
Conclusion
Election polling has repeatedly failed throughout modern history, each time in a new way, so pollsters (and those who rely on them) should be skeptical of polling's predictive reliability, since fixing the last error does not prevent the next one.
Premises
- Early polling failures came from flawed methodology: Literary Digest's 1936 poll, despite 2.4 million responses, used biased sampling frames (phone books, car registrations) and wrongly predicted a Landon win; the 1948 'Dewey Defeats Truman' error came from relying on outdated polls.
- Even as polling professionalized and campaigns came to depend on celebrity pollsters (Harris, Caddell, Wirthlin, Morris), those pollsters were successful but not infallible, as when Caddell misjudged the public mood before Carter's 'malaise' speech.
- Modern failures keep occurring through different mechanisms: the 2004 exit polls wrongly predicted a Kerry win, Romney's 2012 pollster used flawed turnout weighting that underestimated Obama-coalition turnout, and 2016 national polls predicted a Clinton win that did not happen.
- In the current 2026 midterm cycle, partisan polls overstated progressive support in the primaries, skewing perceptions and causing some candidates to drop out, and one pollster reportedly admitted fabricating surveys.
- Shrinking response rates and partisan polling continue to skew public-opinion data despite technological advances, so the pattern of failure persists.
Assumptions
- The cited failures are representative of polling performance generally, not just the most memorable misses.
- Each failure was caused by a distinct, newly arising flaw rather than a single persistent underlying problem.
- Pollsters respond to the most recent error by correcting for it, leaving them vulnerable to new kinds of error.
- Polling accuracy should be judged by whether it correctly predicts election winners.
- Declining response rates and partisan sponsorship meaningfully degrade data quality.
- Technological advances have not solved the fundamental problem of obtaining representative samples.