A polling average combines selected survey results to summarize the evidence. Its value depends on the polls included and the rules used to combine them.
An average can make a noisy sequence easier to read. It does not transform surveys into an official vote count.
There is no single universal average
An aggregator may give recent polls more influence, apply quality criteria, limit repeated releases from one pollster, or adjust for methodological differences.
Two averages can therefore differ while drawing from overlapping sources. AAPOR explains that aggregators make their own inclusion and weighting choices. Source: AAPOR's poll aggregation guide.
Before comparing displayed numbers, read the methodology and the latest update time.
A simple example
Suppose three hypothetical surveys estimate Candidate A at 46%, 48%, and 49%. An equally weighted mean is approximately 47.7%.
That arithmetic example is not a model of a real race. If the 49% estimate is newer, an aggregator emphasizing recency might produce a different figure. If two polls share a sample or design, treating all three as fully independent could overstate how much separate evidence exists.
What averaging cannot fix
If several surveys miss the same group or make a similar turnout mistake, the average can retain that systematic error.
A smoother line can appear more certain than the underlying evidence deserves. Averaging reduces some random variation; it does not guarantee correct coverage of the electorate.
Average versus forecast
An average summarizes selected polling. A forecast may incorporate the average plus assumptions about historical errors, turnout, race conditions, and correlations.
A vote estimate is therefore different from a modeled win probability. Read the label before interpreting the number.
A reader's checklist
Check the last update, the included polls, the population definitions, the handling of older surveys, and any adjustments. Ask how much fresh independent evidence supports the latest movement.
Treat a change of a few tenths of a percentage point cautiously. Extra decimal places are not proof of extra predictive accuracy.
Related reading: win probability vs. vote share and reading 2026 polls.