Review and Clean Survey Responses

Response review is where researchers identify test submissions, incomplete cases, failed checks, duplicate entries, and unusual open-text responses. Keep the raw export intact, create a separate analytic copy, and document every exclusion rule outside the original data file.

Start from Export. Download the response file, then compare it against the survey structure, Flow paths, quota rules, and any quality checks that were part of the study design.

1. Step 1: Export a backup. Download the current data from Export Research Survey Responses before cleaning.

   Step 1: Export a backup

2. Step 2: Identify test responses. Compare timestamps, draft/public source fields, known pilot submissions, and notes from Preview and Pilot Test a Survey.

   Step 2: Identify test responses

3. Step 3: Check quality fields. Review attention checks, validation fields, completion status, timing, duplicate identifiers, and open text that suggests a failed task.

   Step 3: Check quality fields

4. Step 4: Check experimental fields. Confirm condition assignments, quota-relevant variables, randomized order fields, and treatment exposure fields are present.

   Step 4: Check experimental fields

5. Step 5: Apply planned exclusions. Use criteria from your pre-analysis plan or codebook instead of making ad hoc cuts.

   Step 5: Apply planned exclusions

6. Step 6: Save a cleaning log. Record which rows were excluded, why, who made the decision, and when the cleaned dataset was created.

   Step 6: Save a cleaning log

• Dataset version: export date, survey version, and whether the file came from manual Export or the API.
• Rows removed: test responses, duplicates, incomplete cases, and the row ids or response ids affected.
• Check outcomes: failed attention, comprehension, manipulation, or validation checks, plus whether those failures are exclusions or diagnostic flags.
• Routing status: ineligible, quota-full, screened-out, or incomplete paths.
• Any personally identifying fields handled under privacy rules.
• Fielding changes: mid-field wording, Flow, quota, link, or end screen changes already noted in the codebook.

• Do not overwrite the raw export with recoded or filtered data.
• Do not remove respondents only because their answer is inconvenient for the hypothesis.
• Do not change option labels after launch without documenting how old and new labels map.
• Do not rely on memory for pilot submissions; mark or log test responses when the pilot happens.
• Do not discard condition assignment or randomized order columns, because they explain what each respondent saw.

• Survey Data Quality Checks
• Survey Variable Names and Recodes
• Pull Survey Responses With the API