Polling sources

Source model

Each country loader targets a defined set of national polling tables, normalises pollster and party names, rejects unrelated table types, and keeps fieldwork dates, samples, results, and source URLs. The country page shows a compiled-source link and links individual pollster names to original reports whenever possible.

The United States tracker uses FiveThirtyEight's historical archives, archived tables from the former public HuffPost Pollster API, and VoteHub's current generic-ballot API. FiveThirtyEight and VoteHub publish their data under the Creative Commons Attribution 4.0 licence. Bismuth Media filters and normalises the records; none of the sources endorses this site.

Compiled tables are useful, but they can contain errors. A value should be checked against the linked original publication before being quoted as authoritative.

Published country register

31 published country trackers

United States archive

The national generic congressional ballot tracker combines an archived FiveThirtyEight generic-ballot dataset, FiveThirtyEight's pollster-ratings archive, archived tables from the former public HuffPost Pollster API, and VoteHub's current polling API. FiveThirtyEight and VoteHub publish their data under the Creative Commons Attribution 4.0 licence. Records are filtered and normalised by Bismuth Media; none of the sources endorses this site.

FiveThirtyEight's detailed archive supplies fieldwork ranges and original report links from 2016 onward. The archived HuffPost tables extend comparable nonpartisan voter polling back through 2012. Earlier rows come from FiveThirtyEight's pollster-ratings archive, which provides the median field date rather than the full fieldwork range and contains campaign-period polling only. The tracker excludes presidential, state, district, primary, adult-only, partisan and internal polling.

Fallback archives

Country pages include checked local polling data so they still render when the internet source is unavailable. The page labels that state as bundled or fallback data. It does not silently claim that a fallback was fetched live.

Corrections

If a result, date, party label, pollster, or source link looks wrong, include the country, visible value, expected value, and original publication in a correction report. Reproducible corrections are prioritised over broad requests to change an average.