In Reply to: Seven data-driven lessons from the 2025 elections posted by mh on November 06, 2025 at 11:47:08
He makes a few contradictory statements
> Pollsters missed by about as much as they did in 2024 — but in the other direction
> Weighting polls to the 2024 presidential electorate may have artificially punished Democrats
In general, the polls were only about 1% to 3% biased in the swing states (still in the margin of error) for 2024, which makes them very accurate. The slight bias accounts for Nate Silver's prediction that the most likely outcome is a narrow sweep of Trump in the swing states since we tend to not properly model Trump's ability to pull in 1st time and low-propensity voters. The bias against Dems, I believe, is bigger in this midterm.
Any pollster who used the 2024 demographic weight is a moron because midterms after a presidential election are always different from a presidential election. The more dramatic the win, the bigger the swing. At the very least, they should start with the midterm demographics after Trump's first Presidency and assume he does worse.
He also said, don't use the result of just a few polls to judge the quality of a poll then goes on to say he did better because of the results of a handful of polls.
I'm sure their stats are good, but some of their reasoning is obvious or contradicts itself.