Poll aggregation

Poll aggregation combines results from many separate polls into one composite estimate, on the premise that averaging reduces the random noise in any single poll and can partly offset bias specific to individual pollsters, sometimes called house effects.

Simple aggregators average recent polls; more sophisticated ones weight each poll by sample size, historical pollster accuracy, recency, and methodology, and some feed the aggregate into a broader forecasting model that adds fundamentals such as economic indicators or incumbency and estimates uncertainty around the final number. This differs from a single poll's margin of error, which describes only that poll's own sampling error and says nothing about bias shared across polls that use similar methods, for instance a common likely voter model or survey weighting approach.

Aggregation is generally more accurate than any single poll and is the basis for most public election forecasts and opinion trackers. It cannot, however, correct for an error common to most pollsters in a given cycle, industry-wide misses tied to a shared blind spot have occurred, and decisions about which polls to include or exclude, and how much weight to give partisan-sponsored polls, can shift the average meaningfully. Aggregation is distinct from an exit poll, which surveys voters after they have voted rather than combining pre-election estimates.