This post is part of The Data Lab, a dedicated section of Strength In Numbers for extra charts, general polling wonkery, and methods work that can be a little too nerdy for the normal editorial vertical. If you’re the type of person who wants to know how our polling is done or reads election forecasting methodologies, the Data Lab is for you.
Emails for posts on the Data Lab vertical are off by default, but you can easily toggle them on here. The Data Lab also accepts pitches by contributors.
Most Data Lab posts are free, but the occasional deeper ones are for paying subscribers, whose support makes all of this possible.
This will be the last column I write in 2026 about the way the polls could go wrong. Tomorrow I’ll have an article synthesizing all the different posts I’ve written this cycle about the various mechanisms of polling error, the consequences of weighting choices, and partisan targets that may be pushing surveys in one direction or another. But for now, a final and potentially important post about one remaining source of error.
This last source of error stems from geographic bias in who’s answering polls. While many firms balance their samples by sex, age, race, region, party, and/or vote in some past election, few consider the interplay between these factors. In our Strength In Numbers/Verasight polls, however, the interaction of geography and past presidential vote turns out to matter a lot.
So here’s the problem: After weighting, 15% of the 2024 Trump voters in our latest poll live in the bluest zip codes, which are home to a fifth of American adults. But only about 7% of Trump’s actual votes were cast there. These urban Republicans are far more likely than other Trump voters to back Democrats for Congress, so there is a concern this bias may end up artificially inflating support for Democrats on the House generic ballot.
But by how much?