A Codebook for Open-Ended Questions: Complete Example
A codebook for open-ended questions is an instruction manual for turning text into categories. It tells an analyst what each code means, when to apply it and when to leave it off. A list of attractive theme names is only the beginning. The difficult work is deciding where neighbouring themes meet and making those decisions available to anyone who codes the next batch.
Decide what this codebook covers
This guide uses a fictional customer-service survey asking, “What stood out about your most recent support experience?” The objective is to describe the experience, including positive and negative mentions. It is not to decide whether an agent met an internal service target or to infer the customer's overall satisfaction from a short phrase.
The unit is one complete answer. A respondent may mention several aspects, and each aspect is recorded once per answer. The codebook covers this one question and one wave. If you reuse it for another question, check whether its definitions and residual categories still make sense.
Keep theme separate from direction. “Quick reply” and “slow reply” both mention waiting. If the report needs praise versus criticism, introduce an explicit second dimension with its own rules. Otherwise a theme frequency will mix the two, and the label must make that clear.
Use fields that resolve real decisions
A stable code is the identifier stored in the data. A label is the readable name in tables. The description defines the concept; include and exclude describe its boundaries. The examples field gives short illustrations. In the companion CSV, several examples can be separated with a vertical bar.
Do not rename IDs to change a dashboard heading. Change the label while keeping the identifier stable when the meaning is unchanged. If the meaning changes, record the version and determine whether historic results require recoding. A version note should include the question, date, change and affected rows, outside the upload table if necessary.
Complete example codebook
Download the six-column codebook CSV. These six topical codes are uploadable. The residual statuses described below are handled separately.
| ID | Label and definition | Include | Exclude | Example |
|---|---|---|---|---|
| wait | Waiting: time before a reply or service | Queues, callbacks, immediate replies | Time spent solving after contact | “I waited all afternoon.” |
| access | Contact access: reaching the appropriate channel or person | Broken chat, unavailable phone, transfers | Waiting after reaching a working channel | “The chat button did nothing.” |
| manner | Staff manner: interpersonal treatment | Courtesy, empathy, dismissiveness | Technical competence without treatment | “She listened without rushing me.” |
| knowledge | Staff knowledge: understanding and explanation | Expertise, inaccurate advice, clear explanations | Courtesy alone; outcome alone | “He explained the settings clearly.” |
| resolution | Resolution: whether the issue was fixed | Fixed, unresolved, recurring issue | Waiting with no outcome stated | “The same fault returned.” |
| effort | Customer effort: work needed to pursue the case | Repeating details, sending documents, repeated contacts | Passive waiting; one transfer alone | “I had to explain it again.” |
Residual and missing-answer rules
Use other when an answer has meaningful content but none of these topical codes applies. “The logo on your emails is ugly” is outside this support-experience frame. Keep the original text and inspect repeated other themes before freezing the codebook.
Use non_response for blank text, a refusal to answer or a placeholder such as “prefer not to say.” An answer saying “I have not contacted support” is not automatically a refusal: it describes ineligibility for this question. Preserve a separate eligibility flag in the survey data. In a simple project without such a flag, document explicitly how you handle it rather than inventing a service experience.
The app reserves other and non_response as computed statuses, so they are not rows in the uploadable CSV. For manual coding you can store them in a status column. Never combine other with an applicable theme simply because part of a sentence is unusual. If some text fits a theme, assign the supported themes and record any uncovered detail in notes.
Multiple-code and overlap rules
Apply every explicitly supported theme. A transfer by itself receives access. A transfer followed by having to repeat the entire story receives access and effort. “They fixed it immediately” receives resolution; add wait only when the wording also describes the wait for service. These local rules avoid analysts quietly expanding different categories.
Do not infer staff knowledge from successful resolution. A case can be fixed without the customer mentioning expertise. Likewise, an unresolved case does not prove rudeness. Read the answer in context, but avoid filling gaps with a plausible story.
Worked example: ten coding decisions
The answers below are invented to demonstrate the boundaries.
| ID | Answer | Decision |
|---|---|---|
| S01 | “I waited forty minutes for someone to answer.” | wait |
| S02 | “The chat button would not open.” | access |
| S03 | “She was patient and made me feel heard.” | manner |
| S04 | “He explained exactly why the setting was wrong.” | knowledge |
| S05 | “The problem is finally fixed.” | resolution |
| S06 | “I had to send the same documents three times.” | effort |
| S07 | “I was transferred twice and repeated my story each time.” | access; effort |
| S08 | “A quick reply, but the fault still happens.” | wait; resolution |
| S09 | “Please change the logo on your emails.” | other |
| S10 | “Prefer not to say.” | non_response |
S08 demonstrates why a theme label cannot stand in for a sentiment label. The respondent praises one aspect and criticises another. S04 concerns explanation, not necessarily resolution: the answer does not say whether the problem disappeared. When the evidence stops, the code assignment should stop too.
Pilot, revise and freeze
Give the draft and a mixed batch of answers to the person who will use it. Ask them to record uncertain decisions before discussing the intended answer. Compare the reasoning behind disagreements. “We read it differently” often points to a missing inclusion rule, overlapping categories or a unit that was never specified.
Revise the rules and rerun affected examples. Keep a short decision log for recurring edge cases, then incorporate stable decisions into the codebook itself. A long private notes document that contradicts the published definitions makes handover difficult. Freeze a version for production coding, while allowing a documented correction if new evidence exposes a gap.
Before comparing waves, check semantic continuity. A larger count after splitting one category into three is not automatically a real-world change. Recode comparable historic material where feasible, or clearly mark the break in the series.
Checklist for a usable codebook
- State the question, purpose, unit and allowed number of codes.
- Give each code a unique stable ID and a precise definition.
- Include boundaries and examples for overlapping themes.
- Separate other content, missing answers and eligibility.
- Pilot difficult cases and preserve the final version with the data.
FAQ
Is a codeframe the same as a codebook?
People use the terms differently. A codeframe often means the category structure; a codebook usually includes instructions and definitions. Specify the fields and rules you deliver instead of relying on the label alone.
How many codes should I use?
Use enough to answer the research question at a useful level of detail. Merge categories that lead to the same interpretation and cannot be distinguished consistently; split broad categories when the distinction supports a concrete analysis.
Must categories be mutually exclusive?
They need clear boundaries, but a whole answer can contain several distinct themes. State whether exclusivity applies to a single meaning unit or to the complete response.
What if a new theme appears late?
Log it, assess whether it changes the analysis and review earlier answers if you add it. Do not silently give late respondents access to a category that earlier respondents could never receive.
Survey Coding by Decision Models (Beta) accepts CSV/XLSX answers with an uploaded codebook or a template, shows probabilities per code and supports review of uncertain answers. Export CSV/XLSX or CSV with SPSS syntax; up to 500 answers per project are free. Open Survey Coding.