Survey Data Quality & Open-End Coder
Before analysis: exclusion rules for speeders, straightliners, and junk; then a codebook for open-ends with example recodes. Hand off a clean file to Survey Analysis Framework. Does not invent statistics.
Prompt
Act as a Principal UX Researcher who cleans survey data before anyone charts it.
I need exclusion rules and an open-end codebook for this survey.
Survey purpose: [decision this informs]
Questions (closed + open): [paste instrument]
Completion time target: [minutes]
n started / n submitted: [if known]
Segment fields: [e.g., plan type, tenure, role]
Open-ended responses (or a representative sample if the dump is huge):
[Paste]
Optional: paste a few closed-item distributions if you already have them. Do not ask me to invent counts.
Please produce:
1. Data quality rules
For each rule, give the definition, why it matters, and what to do (exclude / keep / flag):
- Speeder (too fast vs. the time target)
- Straightliner / patterned clicking
- Duplicate or copy-paste open-ends
- Nonsense, URL spam, or off-language dumps
- Incomplete but usable (which items can be missing)
List rules I should NOT apply (e.g., dropping people who chose the midpoint).
2. Exclusion log template
Columns: response id, rule, evidence, keep/drop. Fill 3-5 example rows from my paste only. If I did not paste ids, use "Row A" labels.
3. Open-end codebook
For each open question:
- Theme codes (short names)
- Definition in one sentence
- Inclusion / exclusion for the code
- 2 example quotes from my paste (or say insufficient text)
- An "other / unclear" bucket
4. Recode examples
Show 5-10 actual lines from my paste mapped to codes. Do not polish quotes.
5. What I still cannot claim
Sample-size caveats, missing segments, and why this output is not a findings report.
If the paste is too thin for a codebook, say so and ask for more rows. Do not fabricate responses.
Follow-ups: "Merge overlapping codes" or "Export the codebook as a two-column table I can use in a spreadsheet."How to use
- 1Export from your tool with response id and duration if you have them.
- 2Anonymize names and emails before pasting.
- 3Apply the keep/drop rules in the spreadsheet first, then run Survey Analysis Framework on the cleaned set.
- 4If open-ends run to hundreds of rows, paste a random 40-60 first, lock the codebook, then code the rest yourself or in batches.
Tips
- Do not let the model compute significance or "key findings" in this pass.
- A code that appears once is an outlier, not a theme.
- If duration is missing, skip speeder rules rather than guessing.
