1. Start with a decision
We begin with the decision the research is meant to inform, then state one primary learning question in falsifiable terms. A useful question names the audience, stimulus, comparison, outcome, and any critical guardrail. Secondary outcomes are labeled before fielding so they cannot quietly replace a disappointing primary result.
2. Define the intended population
Every study defines eligibility, geography, language, age, recruitment source, and exclusions. “U.S. and Canadian consumers” is not adequate if the questionnaire is English-only or the Canadian sample omits Quebec. Reports distinguish the people invited, those who started, the achieved analytic sample, and the population to which conclusions may reasonably extend.
3. Freeze the instrument
Before launch, the study record stores exact question wording, answer choices, order, randomization, stimulus files, device rules, attention and fraud checks, quality exclusions, minimum cell sizes, stopping rule, and planned comparisons. Changes after fielding begins receive a dated change note.
4. Protect participants
Participation is voluntary and limited to adults. A study-specific notice appears before data collection. We minimize collected fields, separate any reward contact details from response data, restrict access, set retention before collection, and publish only aggregates. Sensitive topics, minors, precise location, and unrestricted personal narratives are outside the initial research scope.
5. Field without ad pressure
Questionnaires, consent, confirmation, reward, login, error, and redirect pages are an ad-free zone. Compensation, when offered, is tied only to legitimate participation—not viewing, engaging with, or clicking advertising. Recruitment sources and incentive terms are disclosed in the report.
6. Apply documented quality rules
Quality checks can include impossible completion timing, contradictory eligibility, duplicate or automated behavior, failed instructed-response items, and nonsensical open text. We report counts by exclusion reason. Rules are not changed merely because they produce a preferred answer.
7. Analyze with uncertainty intact
Reports show unweighted bases and explain weighting, missing-data treatment, multiple-comparison controls, derived metrics, and suppression rules. Confidence intervals and significance tests are reported only when their assumptions are defensible. Descriptive differences are not automatically called causal or culturally explanatory.
8. Publish a complete evidence bundle
Each legitimate study is designed to produce a methods page, sample disclosure, field dates, aggregate findings, limitations, human-reviewed analysis, a useful visualization or benchmark, and an aggregate dataset or codebook when disclosure risk permits. Cells below the publication threshold are suppressed.
Separation rule: sponsors may review factual descriptions of their product or campaign before publication, but they may not select respondents, remove valid unfavorable results, rewrite independent conclusions, or receive participant identities.