Comparing Category Sets Across Two Survey Waves¶
A tracking study's whole value is being able to say "this theme grew" or "this complaint dropped off" between waves. That comparison only works if the categories mean the same thing in both waves - and since each survey's Text Analytics results are their own independent category set, that alignment doesn't happen automatically.
Why Category Sets Can Drift Between Waves¶
Category generation analyzes whatever responses exist for that specific question at that specific time (see What Text Analytics Actually Analyzes). Run generation independently on Wave 1 and Wave 2, and you can end up with two category sets that name similar themes slightly differently, split a theme differently, or capture a theme in one wave that didn't show up distinctly enough to earn its own category in the other. None of that is a malfunction - it's what happens when two runs are treated as unrelated data sets, which is technically what they are.
Keeping Waves Comparable¶
Generate once, then reuse the definitions. Run category generation on your first wave, refine the category set until you're happy with it (see Editing and Refining Categories in Text Analytics), and then manually build the same category names and definitions for later waves rather than letting each wave generate its own independent set. This is the single most reliable way to keep results genuinely comparable.
Avoid regenerating fresh categories each wave. It's tempting to just run AI Auto-Generate Categories again on each new wave for convenience, but doing so treats each wave as a standalone analysis rather than a comparable point in a series - see Coding Consistency: Why Re-Generating Categories Can Produce a Different Set Each Time for why this specifically happens.
Document your category definitions somewhere durable. A short written definition for each category - kept outside the tool, in whatever your team already uses for project documentation - makes it much easier to apply the exact same standard when you build categories for the next wave, especially if a different person on your team does the coding that time.
Bringing the Comparison Into a Report¶
Once both waves are classified against the same category definitions, use Combined Survey (see Cross-Tabs and Combined Survey Analysis in Reporting) to compare category frequencies across waves in one report, rather than manually reconciling two separate exports.
FAQ¶
Can Text Analytics automatically apply Wave 1's categories to Wave 2 for me?
Category sets are built per question per survey - applying a prior wave's categories to a new wave is a manual step (recreating the same categories) rather than an automatic carryover.
What if a genuinely new theme shows up in Wave 2 that wasn't present in Wave 1?
Add it as a new category in Wave 2 - a tracking study's category set can evolve over time as long as you're deliberate about it and document when a category was added, rather than letting it happen by accident through independent regeneration.
Does reusing category definitions across waves cost extra credits?
Manually building categories doesn't cost credits - only category generation and classification runs do. Classification still runs (and costs credits) for each wave's own responses.
Is there a way to merge two waves' data into one Text Analytics run instead of comparing separately?
Text Analytics classifies per survey - Combined Survey in Reporting is the tool for bringing multiple surveys' results together for comparison after each has been classified individually.
For the underlying reason category sets differ across separate generation runs, see Coding Consistency: Why Re-Generating Categories Can Produce a Different Set Each Time.