How Polling Averages Work: Complete Methodology Guide
Understanding how polls are averaged, weighted, and interpreted
Why Averages Beat Individual Polls
Individual polls have a margin of error of ±3-4 percentage points, meaning a single poll can miss the true number by 6-8 points in either direction. Polling averages aggregate multiple polls, reducing random error and providing a more accurate picture.
In 2022, the FiveThirtyEight national polling average missed the Republican vote margin by 1.3 points. Individual polls ranged from R+8 to D+2 in the final week — a 10-point spread. The average provided a much narrower range.
Key insight: aggregation is most valuable when individual polls use different methodologies. Live phone, online panels, text surveys, and IVR (robocall) polls each have different biases. An average of these captures more signal than any single approach.
How Polls Are Weighted
| Factor | Effect on Weight |
|---|---|
| Pollster grade (A+ vs. C-) | A+ gets 3-5x weight vs. C- |
| Sample size | Larger n reduces weight penalty |
| Recency | Older polls decay exponentially |
| Partisan vs. nonpartisan | Internal polls discounted 30-50% |
| House effect adjustment | Known R-lean pollsters adjusted downward |
| Likely voters vs. registered | LV model polls weighted slightly higher near election |
House Effects: What They Mean
A "house effect" is a systematic lean in a pollster’s results compared to the field average. If Rasmussen consistently shows Republicans 3 points higher than other pollsters, its house effect is R+3. Polling averages adjust for known house effects to reduce bias.
Examples of persistent house effects: Rasmussen (R+2-3), Emerson College (R+1-2), YouGov (D+0-1), Quinnipiac (D+0-1). House effects don’t mean the pollster is wrong — they reflect different methodological choices (likely voter models, question wording, mode).