Midterms 2026Answers · Voters and issues
Why are election polls sometimes wrong?
Polls are wrong when the people who answer them don't match the people who actually vote, or when voters change their minds late. Polls are usually off by a few points, and in a close race a few points is enough to call the wrong winner.
Who answers is the biggest problem
Most people don't pick up for unknown numbers anymore. Response rates have fallen a lot over the years, so pollsters start with a tiny share of the people they try to reach. If the people who do answer lean one way compared to the people who don't, the poll drifts off, even with careful weighting.
This is hard to fix because the missing people are, by definition, missing. You can't measure how they differ.
Guessing who will vote
A poll can be right about opinion and still wrong about the result if it guesses turnout badly. Midterms make this harder. Turnout in midterm years is lower than in presidential years, so the electorate is different, and pollsters have to guess how enthusiastic each group is.
Here's what trips them up:
- A group that says it will vote but stays home
- A group that is harder to reach but turns out in force
- Enthusiasm shifting in the final weeks
Late deciders and honest changes of mind
Some voters really do decide in the last week. A poll from ten days out can be accurate and still look wrong on election night because opinion moved.
Errors tend to line up
One reason polling misses make headlines is that errors are often correlated. If a method underestimates one kind of voter in Ohio, it likely does the same in Pennsylvania. So instead of random misses that cancel out, many states can miss in the same direction at once.
State polls are harder than national ones
National polls have bigger samples and plenty of coverage. State and district polls are fewer, cheaper and noisier, and House races in particular have very little polling. I'd trust a district poll much less than a national average.
How I'd use polls anyway
I don't think the answer is to ignore polls. They're still the best tool for measuring opinion at scale. My rules:
- Treat a poll as an estimate with a real error range, not a prediction
- Watch averages and trends
- Expect misses of a few points, and be ready for either direction
- Be most careful in close races
The campaign version
In Midterms 2026, the free iPhone game I made about founding a new party, the projection can shift under you. The Democrats and Republicans run attack ads where you get close, so a lead you had last turn can slip. I like that, because real campaigns also have to plan for opinion moving after the last poll.