A codebook is not a list of code names. It is the document that lets someone other than you apply your codes to your data and land in roughly the same place. If it cannot do that, it is a table of contents pretending to be a method.
What a code definition needs
Every code in a defensible codebook carries four things. Skip any of them and you will feel it during analysis, not before.
- A name — short, and describing the content rather than your interpretation of it
- A definition — one or two sentences stating what the code captures
- Inclusion criteria — what qualifies, stated positively
- Exclusion criteria — what looks like it qualifies but does not, and where it goes instead
- An anchor example — a real extract from your data that is a clear, uncontroversial instance
The exclusion criteria are the part people skip and the part that does the work. Codes fail at their boundaries, not at their centres. If you have a code for 'distrust of clinicians' and another for 'frustration with the system', every extract that sits between them is a coin flip until you write down which side it falls on and why.
The four failures
Nearly every codebook that arrives here in trouble has at least one of these.
Codes that are really themes. 'Barriers to access' is not a code, it is a bucket with five codes inside it. If a code applies to a third of your extracts, it is not discriminating between anything, and it will produce a findings section that says very little at length.
Codes that are really summaries. 'Participant describes a difficult experience with the appointment system' is a sentence about one extract, not a category that recurs. A code that only ever applies once is a note, and notes belong in memos.
Interpretation smuggled into the name. 'Internalised stigma' asserts a theoretical claim in the label. Code what was said — 'self-blame for condition' — and make the interpretive move explicitly, in a memo, where it can be examined.
Silent drift. You refine what a code means at interview 12 and never revisit interviews 1 to 11. This is the most common and the most damaging, because the dataset is now coded by two different schemes wearing the same labels.
Version your codebook the way you would version code. If a definition changes, the change gets a date, a reason, and a pass back over everything already coded.
The memo habit
Keep a memo alongside the codebook recording why it changed. Two codes merged because the distinction was not holding — write down the extract that made that clear. A code split because it was doing two jobs — record which. These memos are what turn 'themes emerged' into a documented analytical process, and they are what you will reach for when an examiner asks how a particular theme came about.
Before you code the full set
Code three or four transcripts, then stop. Review the codebook against what actually happened: which codes never fired, which fired constantly, where you hesitated. Fix it then. The cost of restructuring a codebook after four transcripts is an afternoon; after thirty it is a month, which is why so many people press on with a scheme they already know is wrong.
