Pre-registered study with public methods, data, and verification

We ran a large, pre-registered online study asking US adults whether they support a proposed Right to Honest Government: a legal duty for major government decisions to be explained honestly on the public record, with evidence, clear reasoning, and accountability when the reasoning is dishonest.

Figures are generated from the frozen S5 dataset (rthg-s5-dataset.jsonl) and published output (rthg-s5.json). S5 results are locked. The S6 instrument is fielding with Verasight as a nationally representative replication. When that wave is published, this page will carry the new figures and keep S5 available as the archived confirmatory snapshot (a stable archive URL will be added then; do not treat the current homepage as that archive yet).

Companion finding: one sentence, near-universal agreement

This finding comes from a different set of tests, not the ballot study above. While we were testing early versions of our websites, we showed people one sentence and asked whether they agreed with it, before they read anything else: “The American people have a right to honest government service.” Of the 1,174 US adults who answered between July 14 and August 16, 2026, 99.4% agreed. On a scale of 1 to 7, the average answer was 6.81, and 85% picked 7, strongly agree. Republicans agreed unanimously (244 of 244), Democrats 99.5% (425 of 427), and independents 99.0% (296 of 299).

Two things to be straight about. These participants were recruited online and chose to take part, so they are not a random sample of the country and this kind of survey cannot claim a margin of error. “1,174 Americans surveyed” is the accurate way to cite it, and “99.4% of Americans” is not. The Verasight wave now in the field asks the same sentence of a nationally representative panel (pre-registered at osf.io/x5ve6), and that is the study that could let the finding be stated more strongly.

Transparency and independent verification

This site provides the survey instrument, frozen de-identified data, analysis code, and pre-registration documents needed to audit reported estimates and, separately, to field a replication using the same instrument and protocol. The full set of materials includes the exact questions respondents answered, the exclusion rules committed to before data inspection, and the scripts that produce the published numbers.

Study design

The study opens by asking respondents to rate 13 substantive items in Section B (11 governance principles plus a cross-partisan-vocabulary restatement and a reverse-coded honesty check) on a 1–7 scale, alongside one instructed attention check that is not counted in the agreement index. This comes before respondents see the proposed law or how it would be enforced. The idea is to learn whether people share the principles that the law is built on, and to give them a clear picture of what the law is trying to achieve before they vote on it. The principles are also part of the law text when placed on a state ballot initiative. In the headline sample, 99.5% of respondents rated the principles at a mean of 5 or higher across all 13 substantive items (pre-registered hypothesis H1).

The study then works through the law in pieces: the ballot text, who reviews the government's reasoning on the public record, and how a citizen review panel would be structured. Each section has its own response items so that people rate one piece at a time. The ballot vote includes a conditional option (“Yes, conditional on the mechanism: I support the principle, but my vote depends on how the review process is structured”), which is reported separately per the pre-registered support definition.

Section B item order is randomized within a pre-registered stratified scheme. The neutral sponsor label (“American Institutions Study 2026”) is the only framing shown to respondents before debrief. Design, exclusion rules, and analysis plan were pre-registered on OSF (osf.io/br3u2) before data inspection. De-identified responses and published summary statistics are linked by a dataset hash; a mismatch indicates alteration.

Verification

npm install
npm run reproduce

Study materials

• rthg-s5-dataset.jsonl — frozen, de-identified responses (salted pseudonyms, no Prolific IDs).
• rthg-s5.json — published numbers, dataset hash, and freeze date.
• INSTRUMENT WALKTHROUGH — section-by-section view of the instrument with per-principle Wilson 95% CI, ballot text, mechanism results, and re-vote shift.
• VERIFY — step-by-step reanalysis checklist, dataset hash confirmation, and exclusion-rule application.
• S5 PRE-REGISTRATION — completed wave (OSF: osf.io/br3u2).
• S6 PRE-REGISTRATION — current wave (OSF: osf.io/x5ve6, registered August 11, 2026, before any confirmatory fielding). Full instrument package, with the trimmed s6-short arm in Appendix A and the s6-verasight representative-panel wave in Appendix B.
• S5 → S6 CHANGES — what changed between the completed S5 wave and the current S6 instrument.
• METHODOLOGY (S5) — sample design, instrument sequence, randomization, attention and quality checks, and exclusion rules for the locked wave.
• METHODOLOGY (S6) — the same design rationale for the S6 instrument now fielding with Verasight, and the s6-short variant.
• S5 PACKET / S6 PACKET — everything for one wave (prereg, methodology, data, instrument, code) as a single set.
• RUN YOUR OWN REPLICATION — step-by-step guide to fielding the identical study on your own Prolific account.

Replicate this study

Limitations

This study uses a nonprobability online panel (Prolific), not a probability sample of U.S. adults; classical margins of error do not apply. Wilson intervals summarize precision for reported proportions. Party-identification subgroup tables are descriptive and were not pre-registered as confirmatory tests. Independent methodological review and external replication are underway.