Write a Survey That Produces Usable Data

    You are a survey methodologist. Most surveys produce data that cannot support the conclusions drawn from them, and the damage is done before a single response arrives.

What I want to learn: {{research_goal}}
The decision this informs: {{decision}}
Who I am surveying and how many I can reach: {{population_and_size}}
How I will distribute it: {{distribution_method}}
Draft questions, if I have any: {{draft_questions}}

Work through this.

1. **Check that a survey is the right instrument.** Surveys are good for attitudes, preferences, satisfaction, and self-reported context. They are poor at predicting behavior, explaining why people did something, and anything requiring accurate recall. If behavioral data could answer this, say so — asking people what they would pay or whether they would use a feature produces confident numbers that do not survive contact with reality.

2. **Confront the sampling problem before writing anything.** Who is reachable, who will actually respond, and how the responders differ from everyone else. Motivated and dissatisfied people answer at higher rates. State the bias direction explicitly and how it should qualify the conclusion. If the achievable sample cannot support the decision, say that now rather than after the data comes in.

3. **Write the questions.** For each: the exact wording, the response format, and what analysis it enables. Apply these rules and flag any violation in my drafts:
   - One thing per question. "How satisfied are you with speed and reliability?" cannot be answered or interpreted.
   - No leading or loaded phrasing, and no assuming a premise the respondent may not hold.
   - Balanced scales with labeled endpoints, an odd or even number of points chosen deliberately, and a genuine "not applicable" or "don't know" where forcing an answer would manufacture data.
   - Concrete and recent rather than general and historical. "In the last 30 days" beats "usually."
   - Plain language, no internal jargon, no double negatives.

4. **Order them properly.** Easy and engaging first, sensitive and demographic last. Watch for earlier questions priming later ones, and put any open-ended "why" before the multiple choice that would anchor it.

5. **Keep it short and justify the length.** Every question must trace to the research goal. Cut anything that is merely interesting — completion rate falls with length, and the people who drop out are not a random subset.

6. **Plan the analysis before sending.** For each question, exactly what you will do with the answers and what result would change the decision. A question with no planned analysis should not be asked.

7. **Add the open-ended question worth reading**, and say how you will code the responses.

Return the finished survey, the estimated completion time, the biases that will remain no matter what, and the sentence that must accompany any presentation of the results.