Interview saturation: how to log new themes without a magic number
How to track saturation across a set of interviews: log new themes, refinements and exceptions, compare segments and decide when to stop without a universal interview count.
Saturation depends on the data and the research goal
“Nothing new after ten interviews” tells you little. The questions may have been too narrow, the participants too similar, the coding too shallow or new details may have stopped being recorded. A log makes the decision reviewable: what counted as new, and for which group?
Studies define saturation differently: no new codes, adequate development of a theme or a persuasive theoretical explanation. Choose a definition that fits your methodology. A popular interview count offers no universal guarantee of quality.
Record what each careful analysis adds
Define what counts as new
Decide whether a new code, theme property, condition, exception or relationship counts. Write down your criterion before reviewing the interviews so it does not shift to fit the desired outcome.
Update the log after each interview
After coding, add a row with the interview ID, segment, new elements and refinements of existing themes. An entry of zero also needs explanation: did the interview add nothing, or was it too thin to tell?
Compare segments separately
An absence of new themes overall can hide a gap in a small but important group. Record the roles, circumstances and situations represented. Keep segments distinct even if combining them would produce a smoother curve.
Document the decision
Explain what is developed well enough, what remains uncertain and what limits the sample. The team decides whether to continue or stop in light of the question, resources and risks. A formula cannot make that decision.
The eighth interview adds no code but refines a condition
This is a fictional training example. The names, circumstances and outcomes are included solely to explain the method.
I use the workaround too, but only when the request can’t wait until morning.
“Workaround” does not need a new code. What is new is the condition of urgency. If you count only new code names, that refinement disappears and you could declare saturation too early.
A theme may be well developed in one segment and supported by only one case in another. This training example shows why the log needs the segment and type of addition alongside a yes-or-no marker.
| Interview | New element | Decision for the series |
|---|---|---|
| A-07 | New code | Check in the next interviews |
| A-08 | Urgency condition | Refine the theme boundary |
| B-02 | Exception | Keep segments separate |
| A-09 | Nothing new | Check quality and coverage |
A row in the saturation log
Use anonymized identifiers and retain a reference to the codebook version.
Research question: Definition of saturation: Interview / anonymized ID: Segment and relevant context: Data quality and completeness: New codes: New theme properties: New relationships: Exceptions: What was confirmed without change: Codebook version: What to ask next: Decision on continuing the series:
Check the reasoning behind the decision
- What counts as new was defined in advance.
- Refinements and exceptions have been preserved.
- Coverage was assessed across relevant segments.
- The log reflects the limitations and interview quality.
A closer look
How many interviews are usually enough?
There is no universal number. It depends on the question, sample, methodology, data quality and definition of saturation.
Does zero new codes mean we can stop?
Not necessarily. New properties, relationships and exceptions may still emerge. Also check interview quality and coverage of important segments.