Foundations of Political Polling and Campaign Research

A poll lands with a single number and a margin of error, and within an hour it has been read as a prediction of the result. It is not one. A poll records what a sample of people said they intended to do on the days they were asked. A forecast is something else: a model that combines many polls with other information and produces a probability. A betting market is something else again: a price set by people wagering money on the outcome. The three answer different questions, and much of the confusion after every election comes from reading one as if it were another.

This chapter sets the ground rules. It looks at what the polls actually achieved in 2024, which is more complicated than either "the polls were right" or "the polls were wrong". It explains why the margin of error printed on a poll covers only part of the uncertainty, so results land outside it more often than the number suggests. And it sets out the most important background fact for anyone who works in campaigns: when researchers measure how much campaign contact changes how people vote, the effects are usually small, often close to zero in general elections.

Why this matters to you

If you run a campaign, commission polls, report on them or regulate them, you are making decisions on numbers that are easy to misread. A seven-point lead with a three-point margin sounds safe until you learn how much error the margin leaves out. A 70% forecast sounds like a verdict until you remember that a 70% chance loses three times in ten. A claim that a new tactic "moves voters" sounds exciting until you set it against the baseline of how little most tactics move anyone. This chapter gives you the habits that protect you from those mistakes, and from the people who profit from them.

What you'll find inside

The chapter opens with the 2024 scoreboard: the best state polling in the United States since 1944, and the same direction of miss for the third election running. It then explains total survey error, the uncertainty the published margin leaves out, and draws the line between a poll, a forecast and a market. You will meet the minimal-effects baseline, the body of evidence that campaigns change fewer votes than almost everyone assumes. You will see what happened when the best-known poll aggregator closed and the averages lost their referee. The chapter compares a voter with a consumer, asking why political research cannot check a stated vote the way a company checks a stated purchase. It looks at why trust in polls fell even as their accuracy improved, and it closes with a guide to reading a probability without being fooled by the one result you just watched.

The honest note

The characteristic caveat of this chapter is uncomfortable for people who sell campaign services. If most persuasion effects are small, then the confident claims that fill campaign pitches, conference talks and vendor decks need more evidence than they usually bring. That does not mean nothing works. It means the default expectation should be modest, and the burden of proof sits with anyone promising a large effect. Every later chapter in this corpus is written against that baseline. When an article reports an effect that is larger than usual, it tells you so, and it tells you how well the finding has held up.

The eight pieces in this chapter

• P0-01 · The 2024 Scoreboard: Best state polls since 1944, same old miss
• P0-02 · Total Margin of Error: The number on the poll is half the story
• P0-03 · Poll, Forecast, Market: Three different questions
• P0-04 · The Minimal-Effects Baseline: Campaigns move less than everyone thinks
• P0-05 · After the Aggregators: The averages lost their referee
• P0-06 · Voter vs Consumer, Applied: Same person, different market
• P0-07 · Trust in Polls: Right more often, believed less
• P0-08 · Reading a Probability: 70% is not a prediction