Polling and Predictions

A poll is a snapshot, not a prediction. It tells you what a sample of people say right now — not what will happen on election day.

The Mechanics

1. Sample — pollsters contact 800-1,500 people (larger samples = smaller margins of error)
2. Weight — raw responses are adjusted to match the population's demographics (age, race, education, geography)
3. Likely voter screen — not everyone who answers will actually vote; pollsters filter for likely voters using past behavior and stated intention
4. Report — results include the topline numbers AND a margin of error (typically ±2-4 points)

The margin of error means: if a poll shows Candidate A at 48% with a ±3% margin, the true value is likely between 45-51%. If Candidate B is at 46%, that race is a statistical tie — not a lead for A.

Methods

Why Polls Fail

Polls have had some spectacular misses:

• 2016 US — National polls were close (Clinton +3, actual +2), but state polls missed the Midwest. Wisconsin hadn't been polled in weeks.
• 2020 US — Polls overestimated Biden by about 4 points in key states. Likely cause: differential non-response (Trump voters were less likely to answer polls).
• 2015 UK — Polls showed a dead heat; Conservatives won by 6 points. Shy Tory effect? Probably differential turnout.
• 2016 Brexit — Most polls showed Remain narrowly ahead. Leave won 52-48.

The common thread: polls struggle with who actually shows up. Modeling turnout is the hardest part of polling, and small errors compound when races are close.

How to Read Polls Like a Pro

1. Look at the average, not individual polls
2. Check the margin of error — leads within it are meaningless
3. Watch trends, not snapshots
4. Be skeptical of polls from partisan organizations
5. Remember: a 70% probability still means 30% chance of the other outcome