For the complete documentation index, see llms.txt.Skip to main content
Media Reporting of Polls
Read 1 of 5
How journalists misreport polls, the incentives that drive bad poll coverage, and how to be a critical consumer of polling news.
Journalists routinely make several errors when reporting polls. They treat leads within the margin of error as meaningful differences. They report subgroup results (women voters, Latino voters) without noting the much larger margin of error for subgroups. They cherry-pick outlier polls that show dramatic results while ignoring the consensus. They describe snapshot polls as predictions of election outcomes.
The incentive structure explains why: 'Race is a toss-up' is a boring headline. 'Smith surges to 3-point lead!' generates clicks. Media organizations also commission polls partly for the coverage they generate, creating an incentive to find newsworthy results rather than confirming what everyone already knows.
A few simple rules make you a better poll consumer. First, look at poll averages (RealClearPolitics, FiveThirtyEight), not individual polls. Second, check the margin of error: if the lead is smaller than the margin, it is a statistical tie. Third, check who conducted the poll: partisan pollsters produce partisan results. Fourth, read the actual questions, because wording matters enormously. Fifth, remember that polls measure current opinion, not future election results. A poll six months before the election is a snapshot, not a forecast.
Imagine a national poll of 1,000 likely voters reporting Candidate A 48%, Candidate B 45%. The headline reads 'A leads by 3.' Is that lead real?
The margin of error for a single proportion at n = 1,000 is roughly ±3 points at 95% confidence. But the quantity that matters is the margin on the DIFFERENCE between two candidates, which is larger — close to ±6 points here. A 3-point gap sitting inside a ±6 swing is a statistical tie, not a lead.
Subgroups are worse. Break the same 1,000-person sample into demographic slices and each slice is tiny, so its margin of error balloons:
| Group | Sample size | Approx. margin of error |
|---|---|---|
| Full sample | 1,000 | ±3 |
| Voters under 30 | 150 | ±8 |
| Latino voters | 100 | ±10 |
| Black women under 30 | 25 | ±20 |
A story trumpeting 'A gains 12 points among Latino voters!' is often reporting noise inside a ±10 band. This is also why poll averages beat single polls: averaging ten independent polls of 1,000 behaves roughly like one poll of 10,000, shrinking the margin toward ±1. The discipline is just arithmetic — but it overturns most of the dramatic poll headlines you will ever read.
