Saturation is a decision rule, not a number

“We stopped at saturation” is incomplete unless the study says what saturated, when, and why.

The short answer is that saturation is a stopping decision with a declared object. It can concern the appearance of new codes, the depth of meaning within those codes, or the integration of a theoretical account. Those are not interchangeable endpoints.

Saunders and colleagues show that saturation is widely used in qualitative research but inconsistently conceptualized and operationalized. Hennink and Kaiser distinguish code saturation from meaning saturation and caution that sample-size guidance depends on population homogeneity, study aims, and the type of saturation assessed. van Rijnsoever’s simulation adds sampling strategy and information structure to the decision.

The combined lesson is practical. A sample-size number can be a planning input. It is not a sufficient stopping rule.

What theoretical entity is actually supposed to achieve saturation?

Before collecting data, define the stopping object:

Stopping object	What it asks	What it does not prove
Code saturation	Are the relevant topics, categories, or codes appearing?	That their meanings are fully understood
Meaning saturation	Has additional data added little conceptual depth, nuance, or explanation?	That every possible code has appeared
Theoretical saturation	Has the emerging theoretical account integrated the relevant variation for its purpose?	That the theory is true in every setting

The labels are not a license to choose the easiest endpoint after the fact. A study focused on mapping topics may need code coverage. A study explaining mechanisms may need interpretive depth. A grounded theory project may use theoretical integration as its stopping logic. Each endpoint creates a different evidence burden.

Synthetic stopping card	Code saturation	Meaning saturation	Theoretical saturation
Primary test	No relevant new code appears	New data adds little conceptual depth	No relevant theoretical relationship remains unintegrated
Sampling implication	Cover the intended topic space	Seek variation that can deepen interpretation	Sample information sources that can challenge the account
Required record	Codebook and case-by-code trail	Analytic memos and comparison trail	Theoretical memos and negative-case decisions
Stop question	Would another case add a relevant category?	Would another case change the explanation?	Would another case alter the theoretical integration?

The same interview can add no new code and still change the meaning of an existing code. Conversely, a study can generate rich stories without covering an important subpopulation. Saturation is therefore an argument about the relationship between data, question, sampling, and analysis.

Why is there no universal target sample size for qualitative interview studies?

Hennink and Kaiser review empirical studies and statistical models that assess sample sizes for saturation. Their evidence converges on a distinction between code and meaning saturation and cautions that guidance depends on population homogeneity, study aims, and the type of saturation.

van Rijnsoever’s simulation makes a related point from another angle. It models a qualitative population as subpopulations containing different information sources and codes. Different sampling scenarios produce different requirements for reaching saturation. The probability of observing codes matters more than the number of codes alone in the simulation.

These findings do not mean sample planning is impossible. They mean that a number must be conditional: conditional on the population, the question, the sampling strategy, the saturation object, and the information structure. A narrow, relatively homogeneous study may require a different design from a multi-perspective study in which important information is distributed across subpopulations.

How does population information structure dictate the necessary discovery sample?

Information structure describes where relevant variation and knowledge are located in the population being sampled. If every participant can speak to every important issue, random selection may cover the field efficiently. If information is distributed across roles, sites, cases, or experiences, the sampling plan needs to reach those sources deliberately.

van Rijnsoever distinguishes random-chance, minimal-information, and maximum-information sampling scenarios. The simulation shows that they produce different sample-size requirements and different tradeoffs. More informative purposive sampling can reach codes efficiently, but it can also produce fewer repetitions per code, which affects validation.

The practical implication is not “always sample for maximum information.” It is to make the sampling strategy part of the saturation claim:

Sampling record	Question to answer
Population map	Which subpopulations or information sources exist?
Access rule	How will each source become eligible for the sample?
Variation rule	Which contrasts are needed for the research question?
Information expectation	Why should the next case add a new code or deepen a meaning?
Repetition need	How much recurrence is needed to challenge or validate the interpretation?

Without this record, a small sample can look efficient when it has simply missed a relevant information source.

Why are code saturation and meaning saturation distinct epistemological milestones?

Code saturation is useful when the study’s first task is to identify the range of topics or categories. Once the codebook stops expanding, the team may have a useful map of the field.

Meaning saturation asks more. It concerns the depth, dimensions, conditions, and relationships within a code. A new case may not create a new label and still show that a familiar label contains two distinct mechanisms, different consequences, or an important exception.

The distinction changes the analysis log. A code-count table is not enough for meaning saturation. The team needs comparison memos, negative cases, boundary notes, and an explanation of why additional data no longer changes the interpretation. The stopping statement should name the analytic work, not only the number of interviews.

Why does theoretical saturation require an explicit substantive theory boundary?

Theoretical saturation is not simply “we have heard the same thing many times.” In van Rijnsoever’s simulation, theoretical saturation is reached after all codes in the modeled population have been observed. In an actual study, the theoretical account also has a declared scope and a relationship to variation in the data.

A useful theoretical stopping note answers:

1. What is the emerging relationship or mechanism?
2. Which cases or information sources could still challenge it?
3. Which negative or deviant cases have been examined?
4. What would count as a change to the theory rather than another example?
5. Why is additional sampling unlikely to alter the account within the declared scope?

The note does not turn a qualitative theory into a universal law. It makes the inference inspectable.

How should researchers conduct a stopping rule audit before the next interview?

At each interim review, ask:

1. What was the declared saturation object?
2. Which relevant codes, meanings, or theoretical relationships changed since the last case?
3. Which information source or subpopulation has not been reached?
4. Does the next case have a reason to add coverage, depth, contrast, or challenge?
5. What evidence would justify stopping?
6. Has the reason for stopping been written before the decision is forgotten?

If the answer to question three is unknown, the study may need a population map before it needs another sample-size argument. If question four has no answer, the team may already be at a defensible stopping point, or it may have lost contact with its research question.

Which three common qualitative saturation claims should researchers reject?

First, do not write “saturation was reached at 12 interviews” without naming the type of saturation and the sample conditions. A number without an object is not a method.

Second, do not treat code saturation as proof that meanings are saturated. New data can deepen a code without adding a new label.

Third, do not assume that the number of codes determines the sample size. Sampling strategy and the probability of observing relevant information also shape the result.

For adjacent decisions, compare the defensible evidence review stopping rule with measurement invariance before comparing language groups.

Where are the methodological boundaries of qualitative saturation rules?

Saunders et al. provide the inconsistency warning and the need to specify the form, timing, and rationale of saturation. Hennink and Kaiser provide the code-versus-meaning distinction and the dependence on population, aims, and saturation type. van Rijnsoever provides the sampling and information-structure simulation. The stopping card and audit are author-owned translations. They do not prescribe a sample size or certify saturation in a live study.

References

1. Saunders, B., Sim, J., Kingstone, T., Baker, S., Waterfield, J., Bartlam, B., Burroughs, H., & Jinks, C. (2018). Saturation in qualitative research: Exploring its conceptualization and operationalization. Quality and Quantity, 52, 1893-1907. DOI
2. Hennink, M., & Kaiser, B. N. (2022). Sample sizes for saturation in qualitative research: A systematic review of empirical tests. Social Science and Medicine, 292, 114523. DOI
3. van Rijnsoever, F. J. (2017). (I Can't Get No) Saturation: A simulation and guidelines for sample sizes in qualitative research. PLOS ONE, 12(7), e0181689. DOI