Methods, limits and how to cite

What is in this archive, how it was built, what it gets wrong, and what it cannot support. The limits below are not boilerplate; each one changes what you may conclude.

What is in scope

A record enters the archive when a survey organization asked a sample of people a question that explicitly names transgender people, gender identity, gender transition, or a trans-specific policy, with field dates between 2014-01-01 and 2026-08-26.

Included: US national polls, US state and local polls, polls of subpopulations (parents, teenagers, teachers, physicians, LGBTQ adults), and — labeled separately — non-US and cross-national surveys. Excluded: questions about LGBT people in aggregate that never name gender identity, and any figure that could not be traced to a source.

The four things to know before citing anything

1. Unverified leads are not data. A record marked unverified lead means the poll is known to exist but its figures were not confirmed against a retrieved source. Some carry no figures at all. They are hidden by default and exist so the gaps in the archive are visible rather than invisible. Do not cite their numbers.

2. Paraphrased wording is not the pollster’s wording. Where a pollster published no topline — increasingly common, as houses release narrative summaries instead — the question text is reconstructed from a write-up and flagged paraphrased wording. Quote those as a description, never as the instrument.

3. A missing subgroup means “not published”, not “not asked”. Party crosstabs appear wherever a pollster published them. Several major houses print none at all, so their absence here says nothing about what the poll measured.

4. Surveys of transgender people are not opinion polls. Records so flagged (the U.S. Transgender Survey, Trevor Project, Washington Post/KFF) are mostly nonprobability community samples. They describe a population; they are never poolable with polls of the public.

Two traps specific to this subject

Wording swamps time

Between-question variation is larger than change over time. Three 2025 polls on care for minors returned 33%, 56% and 71% restrictive. Averaging across a topic produces a number that describes no one’s opinion and hides the actual finding. This is why questions are stored verbatim and why trend series only join near-identical items from the same house.

Calibration: that gap is Cohen’s h ≈ 0.78, about 1.9× the average across 138 framing experiments — large, but within the normal range for non-equivalent items.

The mode transition

The industry moved off live-caller RDD onto online panels across this period. Any trend spanning that shift is partly measuring the shift. Where a change of method falls inside a single series, the Trends page marks it mode change inside series.

In this archive, live-caller samples run more restrictive on sports items than online opt-in ones — so the industry shift would, if anything, depress measured restrictiveness over time rather than inflate it.

How it was built

Extraction agents worked from a single written contract, each assigned a pollster, topic or region, and each forbidden to record a figure it had not retrieved. Output was merged by a build script that deduplicates across agents, canonicalizes pollster names and geographies, and regenerates every artifact from the raw extractions.

A separate verification pass then re-read 77 headline toplines against their primary sources. All headline percentages survived. It also found structural defects — a survey of teenagers filed inside an otherwise all-adult trend, one question series split across four ids, duplicate fieldings, and short chart labels stored as though they were questionnaire text. Those are corrected through a declared corrections file rather than by editing the underlying records, and the affected rows are excluded from this site entirely.

Known gaps

This is everything publicly retrievable within the project’s budget. It is not everything that exists. There is no registry of polls; unreleased internal polling is invisible; and releases from 2014–2017 have had a decade to disappear from the web. The Internet Archive was unreachable throughout the build, which is the single largest cause of loss in the early period.

Absence from this archive is not evidence that a question was never asked.

Some absences were checked and are real: no US poll has a standalone item on the January 2025 passport policy or the “X” gender marker; birth certificates appear nowhere as a standalone item across the entire period; and no general-public polling exists on transgender access to homeless shelters at any point.

Citing

Cite the original pollster, not this site. Every figure links to the release it came from; that release is the citable source and governs where the two disagree. This archive is a finding aid and a consistency check, not a primary source.

Corrections welcome. If a figure here disagrees with a pollster’s own published release, the release is right and this archive is wrong. Records carry the source link precisely so that can be checked. To send a correction, use the corrections page.

Not affiliated with any polling organization. Archive last built September 24, 2026.