Data Saturation
Data saturation is the point in qualitative data collection and analysis at which additional data stop producing new information relevant to the research question, so that further interviews or observations would add little. It is widely used to justify qualitative sample sizes, but it fits some qualitative approaches far better than others.
For a dissertation, the practical question is not “how many interviews make saturation?” There is no universal number. The question is:
Is saturation the right way to judge whether my sample is adequate, and if it is, how will I show the examiner that I reached it?
This page explains what saturation means, the different types, when it is and is not an appropriate criterion, what the evidence says about sample sizes, how to assess and report it, and what to use instead when saturation does not fit your approach.
On this page:
- Data saturation explained simply
- What data saturation is
- Types of saturation
- When saturation is the right criterion
- How many interviews? What the evidence says
- How to assess and report saturation
- Information power: an alternative rationale
- Data saturation in business research
- Application example
- Advantages and limitations
- Common mistakes
- Data saturation in the age of AI and digital research
- Dissertation example
- Exam tip
Data Saturation Explained Simply
Imagine you are interviewing hotel managers about why they struggle to retain kitchen staff. The first interviews raise pay, shift patterns, workload and the cost of travel. By the eighth interview you notice that managers mention the same issues again and again. Nothing new is appearing.
That is the everyday meaning of saturation: you keep hearing what you have already heard. But two cautions follow. First, hearing the same issues is not the same as understanding them; you may know that “shift patterns” matter without yet knowing why they matter differently to small and large hotels. Second, whether “nothing new” is even the right test depends on what kind of analysis you are doing.
What Is Data Saturation?
The idea originated in grounded theory, where Glaser and Strauss (1967) described theoretical saturation: the point at which further data no longer develop the properties of a theoretical category. In grounded theory, saturation is tied to theoretical sampling, in which each new participant is chosen to develop the emerging theory, not simply to add numbers.
Since then, the term has spread across almost all qualitative research and has acquired several meanings. Saunders et al. (2018) identified four distinct approaches to saturation, which differ in how far they rely on inductive or deductive logic and in whether they focus on data collection, data analysis or theory building. Their main conclusion is that saturation should be used in a way that is consistent with the research question, the theoretical position and the analytic framework of the study, and that stretching it too widely makes the concept lose its meaning.
In practice, students usually mean one of two things by “data saturation”: that no new codes or themes are appearing, or that the existing themes are fully understood. These are different, and they need different amounts of data.
Types of Saturation
The table below separates the forms of saturation that dissertation students most often encounter. The labels are used inconsistently in the literature, so it is more important to state which form you mean than which label you use.
| Type | What it means | Typical setting | What you would show |
|---|---|---|---|
| Theoretical saturation | Further data no longer develop the properties of the theoretical categories (Glaser and Strauss, 1967) | Grounded theory with theoretical sampling | How sampling was directed by the emerging categories and why the categories were judged fully developed |
| Code saturation | No additional issues are identified and the codebook stabilises (Hennink et al., 2017) | Thematic analysis using a codebook; applied research | When new codes stopped appearing, for example in a saturation log |
| Meaning saturation | The issues are fully understood and no further dimensions or nuances emerge (Hennink et al., 2017) | Studies that need depth, not only a list of issues | That each main theme has been explored across different participants and contexts |
| Saturation of predefined categories | No new ideas appear within categories specified in advance by a theory (Francis et al., 2010) | Theory-based interview studies with a deductive framework | A stated initial sample and stopping rule, applied category by category |
The distinction between code and meaning saturation matters for sample size. In one study of 25 interviews, Hennink et al. (2017) found that the range of issues was identified within nine interviews, but 16 to 24 interviews were needed to develop a richly textured understanding of them.
When Is Saturation the Right Criterion?
Saturation assumes that there is a finite set of issues “in” the data that additional data will eventually stop revealing. That assumption suits some qualitative approaches and conflicts with others. Before planning a saturation log, decide whether the concept fits your design.
Figure: Is saturation the right criterion for your study? A guide, not a rule, and the author’s synthesis based on Saunders et al. (2018), Braun and Clarke (2021) and Malterud et al. (2016).
Braun and Clarke (2021) argue that data, code, thematic and meaning saturation fit coding-reliability and codebook forms of thematic analysis, which treat meaning as something discovered in the data, but do not fit reflexive thematic analysis. In reflexive thematic analysis, meaning is generated through the researcher’s interpretation, so the decision about when to stop collecting data is situated and subjective and cannot be fully settled in advance. They encourage researchers using that approach to justify sample size in ways consistent with interpretive practice instead of claiming saturation.
The practical consequence for a dissertation is simple: if your methodology chapter says you used reflexive thematic analysis, do not also say you continued until data saturation was reached. Examiners familiar with the method may see the two statements as inconsistent. Use the information power rationale described below, or explain how the depth and relevance of the data supported your interpretation.
How Many Interviews? What the Evidence Says
Empirical studies have tested when saturation occurs in real datasets. They are useful for planning a range, but none provides a number that applies to every study.
| Study | What was tested | Finding | Important condition |
|---|---|---|---|
| Guest et al. (2006) | 60 in-depth interviews with women in two West African countries | Saturation within the first 12 interviews; basic elements of the main themes present by six | One fairly homogeneous group; results may not transfer to more varied samples |
| Hennink et al. (2017) | 25 in-depth interviews | Code saturation at nine interviews; meaning saturation needed 16 to 24 | Depth of understanding needs more data than identifying issues |
| Hennink and Kaiser (2022) | Systematic review of 23 studies (17 using empirical data, 6 using statistical modelling) | Empirical studies reached saturation within 9 to 17 interviews or 4 to 8 focus groups | Applies best to relatively homogeneous populations and narrowly defined objectives; diverse populations and broad aims need more |
Three points follow for a dissertation. First, these ranges describe when saturation was observed under particular conditions; they are not targets to quote as proof that your own sample was sufficient. Second, the more varied your participants and the broader your research question, the larger the sample you should plan. Third, it is reasonable to plan a range in advance (for example, 12 to 20 interviews for a narrow question with a fairly similar group of participants) and then decide where to stop on the basis of your analysis, reporting how you made that decision.
The non-probability sampling page gives a brief introduction to sample size in qualitative studies; this page explains how to justify the number you choose.
How to Assess and Report Saturation
Where saturation fits your approach, the strongest dissertations make the judgement visible instead of simply asserting it. Two published methods help.
A stopping rule set in advance. For theory-based interview studies, Francis et al. (2010) proposed specifying two parameters before data collection: an initial analysis sample (they used 10 interviews) and a stopping criterion, that is, the number of further consecutive interviews without new ideas before data collection stops (they used three). Stating these parameters in advance prevents the researcher from declaring saturation whenever data collection becomes inconvenient.
A base, a run and a threshold. For inductive thematic analysis, Guest et al. (2020) proposed assessing saturation with three elements: a base size (the initial set of interviews whose themes form the comparison point), a run length (the number of subsequent interviews examined together) and a new information threshold (the proportion of new themes in a run below which saturation is judged to have been reached). New themes in each run are expressed relative to the themes in the base.
In either case, a simple saturation log lets you show the process. The example below is illustrative, not real data.
| Interview | New codes identified | Cumulative codes | Notes on meaning |
|---|---|---|---|
| 1–4 | 21 | 21 | Pay, shift patterns, workload, travel costs, management style |
| 5–6 | 4 | 25 | New: training opportunities, seasonal contracts |
| 7–8 | 1 | 26 | New: agency staff; shift patterns now understood differently in small and large hotels |
| 9–10 | 0 | 26 | No new codes; seasonal contracts still explored only by two participants |
| 11–14 | 0 | 26 | No new codes; seasonal contracts now explored across hotel types |
The log shows two things an examiner looks for. Code saturation appears around interviews 9 and 10, but the researcher continued until the less-explored theme had been understood across different types of hotel. The count of codes also depends on how finely the codebook is defined, so the log should be read alongside the notes on meaning, not as a precise measurement.
Information Power: An Alternative Rationale
Information power is a way of judging qualitative sample size by how much relevant information the sample holds, instead of by whether new codes have stopped appearing. Malterud et al. (2016) proposed that the more information power a sample has, the fewer participants are needed, and that information power depends on five elements.
| Element | Smaller sample may be adequate when… | Larger sample is likely to be needed when… |
|---|---|---|
| Aim of the study | The aim is narrow | The aim is broad |
| Sample specificity | Participants have characteristics highly specific to the aim | Participants are only loosely related to the aim |
| Use of established theory | The study is supported by established theory | There is little theoretical background |
| Quality of dialogue | Interviews are strong and focused | Dialogue is weak or unfocused |
| Analysis strategy | In-depth analysis of a few cases | Exploratory cross-case analysis |
Information power is useful in two situations: when your approach does not suit saturation (for example, reflexive thematic analysis), and when you must state a planned sample size in a research proposal before any data exist. It does not produce a number. It gives you a structured argument for why a given number is reasonable, which you then revisit during analysis.
Data Saturation in Business Research
Business and management dissertations often interview managers, entrepreneurs or other professionals who are difficult to reach. Access, not analysis, frequently limits the sample. In that situation, claiming saturation after a handful of interviews is risky, because the sample may have stopped growing for practical reasons rather than analytical ones.
Two habits make business research more convincing. First, be specific about the population: “HR managers in UK retail firms with more than 250 employees” supports a smaller sample better than “managers”. This is the sample specificity element of information power. Second, report the reasons for stopping honestly. If access ended data collection, say so, describe what the data could and could not support, and narrow the conclusions accordingly. Focus groups with staff or customers follow the same logic; the review by Hennink and Kaiser (2022) found saturation within four to eight focus groups under favourable conditions.
Application of Data Saturation: an Example
A master’s student studies how small online retailers decide whether to adopt AI chatbots for customer service. The design is a codebook thematic analysis of semi-structured interviews with owners of retailers with fewer than 50 employees.
In the proposal, she plans 12 to 18 interviews, justifying the range with information power (a focused aim, a specific group of decision-makers, and an interview guide based on technology adoption literature) and with the empirical ranges reported by Hennink and Kaiser (2022). She sets a stopping rule in advance: after an initial 10 interviews, she will stop when three consecutive interviews add no new codes, following the approach of Francis et al. (2010).
New codes stop appearing after interview 11. However, her notes show that one theme, concern about customer data, has been discussed in depth only by retailers selling clothing. She recruits two retailers from other product categories to explore it before stopping at 15 interviews. In the methodology chapter she reports the planned range, the stopping rule, the log, and why she continued beyond code saturation.
Advantages and Limitations of Data Saturation
| Advantages | Limitations |
|---|---|
| Widely recognised by examiners and journals as a sample-size rationale | Used with many inconsistent meanings, so a bare claim is weak evidence |
| Links sample size to the analysis, not to an arbitrary number | Often cannot be known in advance, which makes proposals harder to write |
| Can be documented with logs and stopping rules | Code counts depend on how the codebook is defined |
| Encourages analysis alongside data collection | Conflicts with the assumptions of reflexive and strongly interpretive approaches (Braun and Clarke, 2021) |
Common Mistakes When Using Data Saturation
- Claiming saturation without any evidence of how it was judged.
- Quoting “12 interviews” or another published figure as proof that your own sample was sufficient.
- Claiming saturation in a study that used reflexive thematic analysis.
- Collecting all interviews before starting analysis, then claiming saturation was reached during data collection.
- Treating code saturation as if it meant the themes were fully understood.
- Stopping because access ran out and describing it as saturation.
- Using “theoretical saturation” in a study that did not use grounded theory or theoretical sampling.
Data Saturation in the Age of AI and Digital Research
AI tools make it easy to code transcripts quickly and to count how many new codes each interview adds, which makes saturation curves look precise. The precision can be misleading. The number of new codes depends on how the tool, or the researcher, defines a code, and automated coding may split one idea into several codes or merge distinct ideas into one. A flat curve produced by an AI tool is therefore evidence about the tool’s coding as much as about the data.
Comparisons of human and AI-generated themes also suggest that the two can identify different themes from the same interviews, each missing some that the other finds (Hamilton et al., 2023). If an AI tool reports that no new themes are emerging, that is a prompt to check the transcripts, not a conclusion. Meaning saturation in particular requires the researcher’s own judgement about whether each theme is understood.
Digital data raise a separate issue. Online comments, reviews or forum posts can be collected in very large quantities, so “running out” of new information is no longer limited by recruitment. In that situation, the question shifts from how many data items to collect to how the data were selected and whom they represent, which is a sampling question (see sampling bias).
Large language models can also generate synthetic interview answers. These cannot contribute to saturation in a study of real participants, because they do not provide new information about those participants’ experiences. If AI tools were used for coding or for checking saturation, state which tool, for what task, and how its outputs were checked against the transcripts.
Dissertation Example
The following paragraph shows how sample adequacy can be justified in a methodology chapter for a codebook thematic analysis. The study is hypothetical.
“A planned range of 12 to 18 semi-structured interviews was set on the basis of information power (Malterud et al., 2016): the aim was narrowly focused on chatbot adoption decisions, participants were owners directly responsible for those decisions, and the interview guide drew on established technology adoption literature. Saturation was assessed using a stopping rule specified in advance: an initial analysis sample of 10 interviews, after which data collection would stop once three consecutive interviews produced no new codes (Francis et al., 2010). No new codes were identified after interview 11. Two further interviews were conducted to explore concerns about customer data, which had been discussed in depth by only one type of retailer, giving a final sample of 15. A saturation log is provided in Appendix C. Because the sample consisted of small online retailers in one country, the findings describe the considerations of this group and are not intended for statistical generalisation.”
This wording names the type of saturation, shows that the decision rule existed before the data, provides evidence, explains why data collection continued beyond code saturation and limits the claims to the sample.
Exam Tip
If you are asked to evaluate a qualitative study’s sample size, do not ask only whether saturation was claimed. Ask which kind of saturation was meant, whether it fits the analytic approach, how it was judged and documented, and whether the study needed code saturation or meaning saturation to answer its question. A study that explains its rationale without claiming saturation can be stronger than one that claims saturation without evidence.
Not sure how to justify the sample size in your qualitative study?
Dudovskiy Research Assistant can review your topic, research questions and planned qualitative approach and help you decide whether saturation, information power or another rationale fits your design, and how to word it in your methodology chapter.
John Dudovskiy
References
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, 13(2), 201–216. https://doi.org/10.1080/2159676X.2019.1704846
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, 25(10), 1229–1245. https://doi.org/10.1080/08870440903194015
Glaser, B.G. and Strauss, A.L. (1967). The Discovery of Grounded Theory: Strategies for Qualitative Research. Chicago: Aldine.
Guest, G., Bunce, A. and Johnson, L. (2006). How many interviews are enough? An experiment with data saturation and variability. Field Methods, 18(1), 59–82. https://doi.org/10.1177/1525822X05279903
Guest, G., Namey, E. and Chen, M. (2020). A simple method to assess and report thematic saturation in qualitative research. PLOS ONE, 15(5), e0232076. https://doi.org/10.1371/journal.pone.0232076
Hamilton, L., Elliott, D., Quick, A., Smith, S. and Choplin, V. (2023). Exploring the use of AI in qualitative analysis: a comparative study of guaranteed income data. International Journal of Qualitative Methods, 22. https://doi.org/10.1177/16094069231201504
Hennink, M. and Kaiser, B.N. (2022). Sample sizes for saturation in qualitative research: a systematic review of empirical tests. Social Science & Medicine, 292, 114523. https://doi.org/10.1016/j.socscimed.2021.114523
Hennink, M.M., Kaiser, B.N. and Marconi, V.C. (2017). Code saturation versus meaning saturation: how many interviews are enough? Qualitative Health Research, 27(4), 591–608. https://doi.org/10.1177/1049732316665344
Malterud, K., Siersma, V.D. and Guassora, A.D. (2016). Sample size in qualitative interview studies: guided by information power. Qualitative Health Research, 26(13), 1753–1760. https://doi.org/10.1177/1049732315617444
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, 52(4), 1893–1907. https://doi.org/10.1007/s11135-017-0574-8
Last reviewed: November 2026.