Survey Designer & Bias Auditor

Design a survey for my research goal, then audit your own draft like a hostile methodologist. Bad surveys don't fail loudly — they return confident, wrong numbers. Research goal (the decisions this data will inform): {{research_goal}} Population and how I'll reach them: {{population}} Rough length limit: {{length_limit}} **Step 1 — From goal to constructs.** List the 3-6 things I actually need to measure. For each: why it matters to the decision. Kill any question idea that doesn't map to a construct — "interesting to know" is how surveys bloat past the abandonment point. **Step 2 — Draft the survey.** For each item specify: construct measured, question text, response format (and why that format), and required/optional. Order matters: easy/engaging first, sensitive/demographic last, and no early question that primes answers to later ones. **Step 3 — The bias audit.** Check EVERY item against this catalog and show the check: - **Leading/loaded** — "How much do you love our new feature?" assumes love. - **Double-barreled** — "Was the app fast and easy to use?" is two questions; split it. - **Absent 'no opinion'** — forcing an attitude manufactures data. Where's the legitimate opt-out? - **Unbalanced scales** — 3 positive options vs. 1 negative; asymmetric endpoints. - **Vague quantifiers** — "often", "regularly": replace with concrete frequencies ("2-3 times a week"). - **Recall overreach** — "how many times last year" invites fabrication; shrink the window. - **Social desirability** — will people answer aspirationally? Reword to normalize the honest answer ("Many people rarely..."). - **Acquiescence** — agree/disagree batteries drift toward agree; mix in construct-reversed items. Output the audit as a table: item | risks found | fixed wording. **Step 4 — Analysis preview.** For each construct: the exact chart or statistic I'll compute from its responses. If we can't name it now, the question isn't ready. Note where I need a screener question, and the minimum responses for the subgroup cuts I said I'd make. **Step 5 — Pilot instructions.** The 5-person think-aloud test: what to watch for, and the 2 items in this survey most likely to be misread.