How should I preserve survey routing and answer-option rules?
By Lewsearch
Preserve the approved questionnaire as a set of explicit rules, then check that the synthetic version produces the expected questions, choices and export values for known inputs. Accept the import only when those checks pass. A question shown to the wrong group, an extra answer choice or a changed missing-value code can alter what the results mean.
Write down what must survive the transfer
This guide covers transferring an existing human questionnaire into a synthetic version. For reviewing wording and comprehension before fieldwork, see our questionnaire pretesting guide.
Keep an approved reference copy with question IDs, exact wording, instructions, answer labels and codes. For each question, record:
- Who receives it: the earlier answers or other conditions that make a record eligible, including where each branch ends or rejoins.
- Which choices appear: the full list and any choices carried forward from earlier answers.
- What can be selected: one answer or several, selection limits, and choices such as “None” that cannot accompany another answer.
- How it is presented: fixed or randomized order, anchored choices, and text inserted from earlier answers.
- What the export means: original codes and separate treatment of an answer, a displayed question left unanswered, and a question skipped by design.
Branch rules determine the path through the questionnaire. Qualtrics’ branch logic documentation illustrates how earlier answers control access to later blocks.
Ask the provider to confirm compatibility for each rule actually used. A broad statement that a system supports surveys leaves too much unspecified. Qualtrics’ Synthetic Panels documentation lists routing, carry-forward choices, randomization and recode values separately. That is a useful reason to check these features individually rather than assume they transfer together.
Use known inputs to check the imported version
Give the provider deliberately chosen answers and specify the expected next question, visible choices and saved values. Compare these expectations with the actual execution record and export. Include both permitted inputs and attempts that the original questionnaire would reject.
The Census Bureau’s A2 instrument standard, section A2-3.4, describes verification against specifications, realistic mock inputs and output comparisons. These are useful methods for checking a transfer; the standard governs Census Bureau work.
A hypothetical worked example
Everything in this example is invented for illustration.
Q1 asks which of two brands were purchased in the past 30 days. Its choices are Alpha, code 21; Beta, code 37; None, code 97; and Don’t know, code 98. Multiple brands are allowed. None and Don’t know must each be selected alone.
Q2 asks which purchased brand is preferred. It appears only after at least one brand was selected. It carries forward only those brands, allows one answer and permits a blank answer. Brand codes remain unchanged.
| Test input | Required outcome |
|---|---|
| Q1 = Alpha and Beta; Q2 = Beta | Q2 offers Alpha and Beta; saves 37. |
| Q1 = Alpha; attempt Q2 = Beta | Q2 offers only Alpha; Beta is rejected. |
| Q1 = Alpha and Beta; attempt Q2 = both | Two preference answers are rejected. |
| Attempt Q1 = Alpha and None | Invalid combination is rejected before continuing. |
| Q1 = None, repeated with Don’t know | Q2 is skipped; no preference answer is generated. |
| Q1 = Beta; Q2 left blank | Record Q2 as shown but unanswered, distinct from skipped. |
Suppose 100 hypothetical records contain 60 selecting a brand, 30 selecting None and 10 selecting Don’t know. Q2’s eligible base is 60. If 36 choose Alpha, 18 choose Beta and six leave Q2 unanswered, Alpha represents 36/60 = 60% of eligible records, or 36/54 = 66.7% of answered records. Label whichever base you use. Dividing by all 100 would answer a different question.
Reject or repair an unfaithful import
Stop acceptance if the import broadens eligibility, restores excluded choices, changes codes or fills skipped questions with generated answers. Request a correction and rerun the affected cases, including later questions that depend on them.
If faithful transfer is impossible, ask the research owner to approve a revised questionnaire before running it. Save the changes and assess whether comparisons with the original remain meaningful. Do not quietly simplify the rules to make an import succeed.
Acceptance checklist
Before approving the run, confirm:
- Every eligibility condition and carried-forward choice set matches the reference.
- Invalid combinations and out-of-range selections are rejected as specified.
- Randomization preserves option identity and fixed positions.
- Exported codes and skipped, unanswered and explicit uncertainty categories remain distinguishable.
- Expected and actual test outputs, corrections and final approval are saved together.
For uncertainty categories, see handling “don’t know” answers.
Keep the questionnaire version, processing decisions and reported bases with the results. AAPOR’s disclosure standards cover exact wording and context, data processing and sample sizes. When presenting both sources of evidence, also follow our guide to reporting human and synthetic research together.