Transgender Polling Archive

How the wording changes the answer

The largest source of variation in this archive is not time and not the public. It is how the question was written. This page puts questions on the same policy side by side so the effect is visible instead of averaged away.

The clearest case: medical care for minors, early 2025

Three national polls, ten weeks apart, on the same policy. Items naming a government ban or criminal penalties for doctors return far lower restrictive numbers than items asking whether minors should have access. Roughly a third of Americans want the state to prohibit this; roughly two-thirds are uneasy about minors receiving it. Both are true, and citing either alone misleads.

What this does not show. These three readings differ in polling house, survey mode, sample frame, don’t-know treatment and field date as well as wording. They are not a split-ballot experiment, so the gap cannot be attributed to wording alone. What it does establish is that spreads of this size are normal for non-equivalent questions — which is why no single number can be reported as “public opinion” on this policy.

What the measurement literature says

Large, but not abnormal

Converted to a comparable effect size, the 33→71 gap is Cohen’s h ≈ 0.78 — roughly 1.9× the pooled average across 138 political framing experiments (Amsalem & Zoizner 2022, British Journal of Political Science). Large, at the conventional threshold, but within the range the literature leads you to expect from non-equivalent items.

Format alone can move 10–14 points

Agree/disagree items yield on average 10–14 percentage points more endorsement than forced-choice items on the same view, and offering an explicit “don’t know” can move the base by about 10 (Krosnick & Presser 2010). Format and bookkeeping differences can reach 30–40 points before any framing effect at all.

Methods and caveats →