Pre-post surveys: Same instrument, linked responses
Two survey links do not automatically produce paired data. A paired comparison requires knowing which baseline and follow-up observations belong to the same person. Matching is a design decision, not a cleanup task to postpone until analysis. [1]
Keep the measurement comparable
Use the same approved instrument version, response options, scoring rules, and planned administration approach at both times. Also consider timing. A baseline collected after part of the intervention is no longer an unexposed baseline for that component.
A change from paper to online administration or from private completion to completion in front of a supervisor can introduce differences beyond time. When a change is unavoidable, document it and consider its implications rather than assuming the measurements are identical.
Decide how responses will be linked
Hypothetical example. Twenty-four staff members are invited to a workshop. Twenty complete the baseline survey, 18 complete follow-up, and only 15 responses can be linked reliably. The main paired analysis has 15 pairs, not 38 observations treated as independent people.
| Linkage situation | What the data can support |
|---|---|
| Same participants, reliable links | Within-person change among complete pairs. |
| Different participants at each time | A comparison of the two groups, with attention to composition. |
| Possible overlap, but no reliable links | Descriptive comparisons and a revised analysis plan; independence is not established. |
Never match by spreadsheet row order, similar scores, age, or a guess about identity.
A participant-retained random code can sometimes provide linkage without putting names in the analysis file. A self-generated code is another option, but it can be forgotten, entered inconsistently, or duplicated. Avoid assuming that a code built from personal facts is private simply because it omits a name. Pilot the approved approach for reproducibility and collisions.
Coded information is not automatically anonymous. A separate link file or a combination of indirect identifiers can permit identification. HHS de-identification guidance makes clear that removing direct names is not the whole privacy question. [2]
Protect the design when responses are missing
Plan reminders, an adequate response window, and clear code instructions within your approval requirements. Keep contact information separate from analysis data when linkage or reminders require it. Do not silently discard unmatched responses: report the numbers available at each time and the number used in the paired analysis.
A complete-pair analysis describes people with both measurements. Whether those people differ from those missing follow-up is part of the interpretation, not something a matching code can solve.
Your next decision
Test the whole survey process with fictional responses before launch. Confirm the score direction, export structure, code consistency, and missing-value handling. See Setting up your project data for data layout, Planning for dropout and missing data for missingness, and Paired t-test or Wilcoxon? Analyzing pre-post data for paired analysis.