It is the first question almost every qualitative researcher asks, usually to a supervisor who answers "until you reach saturation". That is true in the way that "spend less than you earn" is true: correct, unhelpful, and impossible to act on before you start.
The practical problem is that saturation is defined after the fact. You cannot know you have stopped hearing new things until you have stopped hearing new things, and by then you have either over-collected — burning months and goodwill — or under-collected, and no amount of careful analysis will fix it.
What saturation actually means
Saturation was never a single idea. In grounded theory, where the term originates, it refers to theoretical saturation: you stop when additional cases stop refining the categories in your emerging theory. That is a claim about a theory, not about a dataset, and it presumes you are sampling theoretically — choosing each next participant to test the model you are building.
Most projects using the word mean something looser: code saturation, the point at which new interviews stop producing new codes. That is a legitimate criterion, but it is a much weaker one, and it arrives earlier than people expect. Codes stop appearing well before meaning stops deepening.
Code saturation tells you when you have heard the range. It does not tell you when you understand it.
Numbers that hold up in practice
The empirical work here is more useful than the theory. Studies that have gone back and analysed when new information actually stopped appearing tend to converge on a narrow band for homogeneous samples with a focused question:
- 9–12 interviews for a homogeneous group and a narrow research question — most core codes appear in the first six
- 15–25 for a study with two or three distinct participant groups you intend to compare
- 20–30 for a heterogeneous sample, a broad question, or a doctoral study that needs to demonstrate depth
- 4–6 focus groups where groups rather than individuals are the unit of analysis
These are planning figures, not guarantees. A study with three contrasting sites needs enough in each site to say anything about that site — the total is the sum of the strata, not a number you pick once.
What to actually do
Set a minimum and a stopping rule before you begin, and write both into your protocol. The minimum is what you commit to collecting. The stopping rule states the conditions under which you would collect more: for instance, that you will run three further interviews after the point at which two consecutive interviews produce no new codes, and stop if those three produce none either.
Then keep a saturation log. After each interview, record the new codes it generated. It takes five minutes, and it converts an unfalsifiable claim into a table you can put in an appendix. When an examiner asks how you knew you had enough, you show them the curve flattening.
Writing it up
Say what you did and why, not what the textbook says. "We planned a minimum of 15 interviews based on the homogeneity of the sample and the focus of the research question, and applied a stopping rule of three consecutive interviews yielding no new codes. Saturation on this criterion was reached at interview 17; two further interviews were conducted and confirmed it." That sentence is defensible. "Data were collected until saturation was reached" is not.
