Saturation is a property of the study, not a number you can borrow

Ask how many interviews are enough and the answer usually arrives as a number: twelve, fifteen, twenty, "until saturation". This second of ten units argues that saturation is a claim about one study's aims, the diversity of its sample and the level at which it is analysed, so a figure that describes one design is treated as a threshold that governs another. The unit before it settled who to ask.

The range is real, narrow, and belongs to someone else's study

Guest, Bunce and Johnson (2006) gave the field its most quoted figure. They analysed sixty in-depth interviews with women in two West African countries and found saturation within the first twelve interviews, with the basic elements of metathemes present as early as six. Of the 114 themes the analysis produced, 70 per cent appeared in the first six interviews and 92 per cent within the first twelve.

Hennink and Kaiser (2022) identified 23 studies that assessed saturation, seventeen with empirical data and six by statistical modelling. The empirical studies saturated within nine to seventeen interviews, or four to eight focus group discussions, particularly where the population was homogeneous and the objectives narrow. Multi-country research, metatheme studies and work on meaning needed larger samples.

A range derived from homogeneous, narrowly focused studies describes homogeneous, narrowly focused studies. It also moves with the method, since four to eight groups is not nine to seventeen interviews.

The analytic target decides, so a stated rule cannot settle the question

Hennink, Kaiser and Marconi (2017) make the dependence on the analytic target visible. They studied 25 in-depth interviews. Code saturation came at nine interviews, the point at which the codebook had stopped growing. Meaning saturation came only at sixteen to twenty-four. A stated rule is still the cleaner discipline. Francis and colleagues (2010) give the strongest version of that case. They propose two decisions taken in advance for theory-based interview studies. The initial analysis sample is set at ten. The stopping criterion is set at three further interviews yielding no new ideas. In the first study, on fourteen general practitioners, the criterion was met for normative beliefs but not for other belief categories or studywise saturation; in the second, saturation came at interview seventeen.

The rule can be stated in advance and checked, and it still does not decide the question. A criterion met for one belief category has shown where the analysis stands, not that the study is complete.

Where a firm number can be defended: counting, modelling and salience

Weller and colleagues (2018) examined 1,147 interviews in 28 sets across eighteen topical domains, defining saturation as less than one new item expected per person. The median sample was 75 interviews, with a range of 15 to 194. Ten respondents produced 95 per cent of the most salient ideas with exhaustive listing, but only 53 per cent with three responses each.

Tran and colleagues (2017) push into estimation. They modelled theme accumulation in open-ended survey data as a mixture of zero-truncated binomial distributions, adapted from ecological research. The data came from a burden-of-treatment survey with 1,053 participants from 34 countries. Doubling an initial sample of between 25 and 200 predicted the number of themes with a mean error between 3 and 1 per cent, and a slope-based rule identified 97.5 per cent of the themes.

That precision concerns a bounded domain of items rather than the depth of a meaning.

Information power replaces the number, and a reporting procedure keeps it inspectable

Malterud, Siersma and Guassora (2016) propose information power in place of saturation. The more relevant information a sample holds, the fewer participants are needed. Five dimensions govern the judgement: the study aim, sample specificity, use of established theory, quality of dialogue and analysis strategy.

Guest, Namey and Chen (2020) show a middle path. Their method reports saturation during and after collection using a base size, a run length and a new information threshold, validated on three thematically coded datasets. Information power replaces the concept; the reporting procedure keeps it and makes it inspectable.

The three datasets differ enough to test the procedure: 40 interviews with African American men about seeking health care, 48 with mothers about medical risk in pregnancy across four interview modes, and 60 with women in Kenya and South Africa at higher risk of HIV acquisition. A stopping rule that holds across that much variation has earned more than one that holds in a single dataset.

Saturation assumes a finite quantity, and reflexive analysis denies that it is one

Braun and Clarke (2021) state the conceptual objection. Saturation as information redundancy fits the coding-reliability versions of thematic analysis, in which codes and themes are found in the data. It does not fit reflexive thematic analysis, where meaning is generated through interpretation rather than excavated from the material. If meaning is interpreted, judgements about how many items are enough are situated and subjective, and cannot be fixed in advance. A saturation figure is then the wrong kind of description for interpretive work.

The published record shows the claim made rather than demonstrated

Vasileiou and colleagues (2018) analysed 214 single-interview-per-participant studies across three health journals over fifteen years. Sample size justifications were limited, unrelated to the number of interviews and related to the journal. Samples were often called small even where saturation was claimed. Saunders and colleagues (2018) identify four approaches to saturation that differ in logic and in their emphasis on collection, analysis or theorising. A stopping rule should also name who is outside the count: a participant interviewed through an interpreter, or one who never reaches the clinic the sample was drawn from, is not in the twelve. The same rule should say what it means for a sample that includes older participants. They argue that saturation is a predictive statement about the unobserved, a judgement that produces equivocation.

Contested question: a transferable target, a reportable measurement, or a conceptual error?

Three positions are held in print. The first treats saturation as a measurement that can be reported and, for bounded questions, estimated. Weller and colleagues (2018) count items in a domain. Tran and colleagues (2017) attach error rates to a modelled estimate. Guest, Namey and Chen (2020) give a procedure for reporting saturation as it develops.

The second treats saturation as a conditional judgement, legitimate only when defined against the study. Hennink and Kaiser (2022) bound their range to homogeneous populations and narrow objectives; Saunders and colleagues (2018) require the operationalisation to match the research question and analytic framework; Malterud and colleagues (2016) replace the number with five dimensions of information.

The third treats saturation as conceptually wrong for interpretive work. Braun and Clarke (2021) hold that meaning is generated by interpretation and cannot be exhausted, so the metaphor of redundancy fails. Varpio and colleagues (2017) argue that a term used as a proxy for quality can damage the thing it was meant to protect, the cobra effect in which the fix worsens the problem.

What would settle it is a study the literature does not yet contain. It would apply the competing criteria to the same dataset, and show whether the rules select the same stopping point.

The three positions are unevenly evidenced. The measurement position has the strongest record: a count across 1,147 interviews in 28 sets (Weller et al., 2018) and a model fitted to 1,053 responses (Tran et al., 2017). The conditional position rests on a review of 23 saturation studies (Hennink and Kaiser, 2022) and two rule-governed studies (Francis et al., 2010), while the conceptual position is argued from theory rather than data (Braun and Clarke, 2021). While the dispute is open, a designer should fix the analytic level and the criterion before collection. The strongest objection is that one review and two studies cannot bound a rule, because saturation moves with aim and diversity.

Where the argument is heading

The direction of travel is away from the borrowed number and towards a stated, study-specific justification. Khan, Sheikh and MacEachen (2026) propose reflexive sufficiency, in which enoughness is grounded in the study's philosophy and purpose and answered by argument rather than arithmetic, while Guest, Namey and Chen (2020) report saturation in measurable form. Hsu and colleagues (2026) give the move a name, putting data adequacy rather than saturation at the centre.

The sources this unit rests on

• Braun, V. and Clarke, V. (2021). To saturate or not to saturate? Questioning data saturation as a useful concept for thematic analysis and sample-size rationales. Qualitative Research in Sport, Exercise and Health. doi:10.1080/2159676x.2019.1704846 Argues that saturation as information redundancy fits coding-reliability thematic analysis but not reflexive thematic analysis, where meaning is generated by interpretation. Holds that judgements about how many items are enough are situated and subjective.
• Francis, J. J., Johnston, M., Robertson, C., Glidewell, L., Entwistle, V., Eccles, M. P. and Grimshaw, J. M. (2010). What is an adequate sample size? Operationalising data saturation for theory-based interview studies. Psychology & Health. doi:10.1080/08870440903194015 Proposes an initial analysis sample of ten interviews plus a stopping criterion of three further interviews yielding no new ideas, applied in two theory-based studies. In the first the criterion was met for one belief category but not the study as a whole; in the second saturation came at interview seventeen.
• Guest, G., Bunce, A. and Johnson, L. (2006). How many interviews are enough? Field Methods, 18(1), 59-82. doi:10.1177/1525822x05279903 Analysed sixty in-depth interviews with women in two West African countries and found saturation within the first twelve interviews, with metatheme elements present as early as six. The finding belongs to a tightly bounded topic and a relatively homogeneous group.
• Guest, G., Namey, E. and Chen, M. (2020). A simple method to assess and report thematic saturation in qualitative research. PLOS ONE. doi:10.1371/journal.pone.0232076 Presents a method for assessing and reporting saturation through a base size, a run length and a new information threshold, validated by bootstrapping on three thematically coded datasets. Argues saturation can be reported as it develops rather than asserted at the end.
• Hennink, M. and Kaiser, B. N. (2022). Sample sizes for saturation in qualitative research: a systematic review of empirical tests. Social Science & Medicine. doi:10.1016/j.socscimed.2021.114523 Reviewed four databases and identified 23 studies that assessed saturation empirically or by statistical modelling. Empirical studies saturated within nine to seventeen interviews or four to eight focus groups, particularly with homogeneous populations and narrow objectives, while multi-country work, metathemes and meaning needed larger samples.
• Hennink, M. M., Kaiser, B. N. and Marconi, V. C. (2017). Code saturation versus meaning saturation. Qualitative Health Research. doi:10.1177/1049732316665344 Compared two ways of assessing saturation in 25 in-depth interviews, reaching code saturation at nine interviews but meaning saturation only at sixteen to twenty-four. Distinguishes hearing it all from understanding it all, so one dataset can carry two defensible stopping points.
• Hsu, C., Glass, D., Cruz, S., Corage Baden, A. and Senturia, K. (2026). Beyond saturation: a qualitative framework for operationalizing respondent sampling (Q-FORS) for data adequacy. Social Science & Medicine, 390, 118781. doi:10.1016/j.socscimed.2025.118781 Reviews the sample-size literature and argues that saturation cannot serve as a general standard for dataset completeness, proposing data adequacy instead. Q-FORS is a practical framework for interviews and focus groups.
• Khan, T. H., Sheikh, M. S. and MacEachen, E. (2026). Revisiting sample size and saturation dilemma in qualitative research: a philosophical and practical reflection on reflexive sufficiency. International Journal of Qualitative Methods. doi:10.1177/16094069261446123 Proposes reflexive sufficiency, in which enoughness is grounded in the study's philosophy and purpose rather than a fixed figure. Argues the question should be answered by argument rather than arithmetic.
• Malterud, K., Siersma, V. D. and Guassora, A. D. (2016). Sample size in qualitative interview studies. Qualitative Health Research, 26(13), 1753-1760. doi:10.1177/1049732315617444 Proposes information power, the principle that the more relevant information a sample holds the fewer participants are needed, judged across five dimensions: aim, sample specificity, established theory, dialogue quality and analysis strategy.
• Saunders, B., Sim, J., Kingstone, T., Baker, S., Waterfield, J., Bartlam, B., Burroughs, H. and Jinks, C. (2018). Saturation in qualitative research: exploring its conceptualization and operationalization. Quality & Quantity. doi:10.1007/s11135-017-0574-8 Identifies four approaches to saturation that differ in logic and in their emphasis on collection, analysis and theorising. Argues saturation is a predictive statement about the unobserved, a judgement that produces equivocation.
• Tran, V. T., Porcher, R., Tran, V. C. and Ravaud, P. (2017). Predicting data saturation in qualitative surveys with mathematical models from ecological research. Journal of Clinical Epidemiology. doi:10.1016/j.jclinepi.2016.10.001 Modelled theme accumulation in open-ended survey data as a mixture of zero-truncated binomial distributions adapted from ecological research, using a burden-of-treatment survey with 1,053 participants from 34 countries. Doubling an initial sample of 25 to 200 predicted the number of themes with a mean error of 3 to 1 per cent, and a slope-based rule identified 97.5 per cent of themes.
• Varpio, L., Ajjawi, R., Monrouxe, L. V., O'Brien, B. C. and Rees, C. E. (2017). Shedding the cobra effect: problematising thematic emergence, triangulation, saturation and member checking. Medical Education. doi:10.1111/medu.13124 Problematises four taken-for-granted qualitative terms, saturation among them, arguing that the strategies used to win acceptance for qualitative work have brought unintended consequences. A term used as a proxy for quality can worsen the problem it was meant to fix.
• Vasileiou, K., Barnett, J., Thorpe, S. and Young, T. (2018). Characterising and justifying sample size sufficiency in interview-based studies: systematic analysis of qualitative health research over a 15-year period. BMC Medical Research Methodology. doi:10.1186/s12874-018-0594-7 Analysed 214 single-interview-per-participant studies across three health journals over fifteen years. Found sample size justifications were limited, unrelated to the number of interviews and related to the journal, with sample sizes often called small even where saturation was claimed.
• Weller, S. C., Vickers, B., Bernard, H. R., Blackburn, A. M., Borgatti, S., Gravlee, C. C. and Johnson, J. C. (2018). Open-ended interview questions and saturation. PLOS ONE. doi:10.1371/journal.pone.0198606 Examined 28 sets of interviews, 1,147 interviews in all, across eighteen topical domains, defining saturation as less than one new item expected per person. Found a median sample of 75 interviews, a range of 15 to 194, and that ten respondents produced 95 per cent of the most salient ideas with exhaustive listing but 53 per cent with three responses each.