Published November 2026 · 9 min read

How to Reuse a Codebook Across Survey Waves (Without Breaking Your Trend Data)

Tracking studies, quarterly pulse surveys, annual member surveys and panel waves all have the same quiet trap. Each wave, someone codes the open ends from scratch, the categories come out slightly different, and the trend chart in the report now compares things that are not the same. AI makes this trap easier to fall into, because generating a fresh codeframe takes seconds. This guide covers how to treat a codebook as a versioned research asset, and the workflow for reusing it wave after wave.

Why Fresh Categories Every Wave Break Your Trends

Imagine a customer tracker's "What could we do better?" question. Wave 1's codeframe has Price. In wave 2 the codeframe is regenerated and splits it into Cost and Value for money. Wave 2's Cost is lower than wave 1's Price. Did pricing complaints fall, or did the definition change? You cannot tell, and neither can your client.

The same thing happens more subtly when category names stay the same but their boundaries shift: a model that put "slow website" under Technical issues last wave and under Speed this wave. A trend is a comparison of the same measure over time. In open-ended coding, the codebook is the measure.

Five Rules for a Tracking Codebook

1. Build it carefully once. Wave 1 (or a pilot) is where you invest in the codeframe: AI suggestions as a starting point, then real researcher review. Every later wave inherits these decisions.
2. Freeze it. From wave 2 onward, apply the saved codebook. Do not regenerate it.
3. Change by adding, not rewriting. If a new theme appears, add a code. Avoid renaming, merging or reordering existing ones, because each of those breaks comparability with earlier waves.
4. Version every change. "v3, added App outages from wave 5" is a sentence your future self, and your client's analyst, will thank you for.
5. Watch the "none" share. The share of answers that fit no code is your early-warning signal. If it climbs from 6% to 14% between waves, something new is being said. Read those answers before you decide to add a code.

The Workflow in SurveyCat

Wave 1: build and save

1. Upload the wave 1 file (CSV, Excel, SPSS or Stata) and select the open-ended questions.
2. Let the AI propose categories for each question, then edit them until they reflect what the study needs to measure. Order matters, as explained below.
3. Run the classification and check the results.
4. On the results page, save each question's categories as a codebook, with a clear name like "Member survey – improvements".

Wave 2 onward: load, check, run

1. Upload the new wave and select the same questions.
2. On the review screen, pick the saved codebook from the dropdown next to the question and click Load. The categories replace the suggestions, and you can still see and edit them before approving.
3. Approve without changes and run.
4. On the results page, compare the none of the above share with last wave's. If it is stable, you are done.

When a new theme shows up

Read the answers coded none. If a clear theme runs through them, add it as a new category at the end of the list, re-run, and save the codebook again under the same name. Saving under an existing name creates a new version (v2, v3…) rather than overwriting, and the codebooks page keeps the full version history. Write the change and the wave it took effect into your project notes.

Saved, versioned codebooks are part of Research Pro. Only category names are saved, never response text, so they stay consistent with SurveyCat's data-retention promises.

Keep Output Columns Stable So Your Syntax Keeps Working

The codebook also determines the column names in your output file, which matters if you have syntax, a do-file or an Excel template that reads them.

• CSV and Excel: columns are named from the category text, for example q5_cat_value_for_money. Same category name, same column name. Renaming a category renames its column.
• SPSS and Stata: variables are numbered by position, q5_c01, q5_c02…, with the category name in the variable label. Same order, same variable names. Inserting or reordering categories shifts the numbers.

This is why rule 3 says add to the end. A code added at the end becomes q5_c07 in SPSS and a new column in Excel. Every existing name stays put, and last wave's syntax runs unchanged. The question variable name matters too: keep it the same across waves (q5, not q5_w2), or rename it before upload.

Stacking Waves for Trend Analysis

With stable column names, combining waves is a simple append: ADD FILES in SPSS, append in Stata, or pasting the rows under each other in Excel, with a wave variable added to each file first. Then the trend is the mean of each 0/1 column by wave:

* Stata: share mentioning each code, by wave, among those who answered
use wave1_coded, clear
append using wave2_coded wave3_coded
tabstat q5_c01-q5_c07 if q5 != "", by(wave) statistics(mean) format(%4.2f)

A code added in wave 3 will be missing for waves 1 and 2 after the append, which is correct: it was not measured then. Show it from its first wave onward in the report, with a footnote.

A Note on Back-Coding

Sometimes a change is unavoidable: a client insists on splitting Price into Cost and Value. The clean fix is to back-code: re-run the earlier waves with the new codebook version so the whole series uses the same definitions. With AI coding, that costs 1 credit per answer and a few minutes per wave, instead of re-reading thousands of verbatims by hand. That makes back-coding a realistic option rather than a footnote apologising for a break in the series.