Qualitative Sample Size: Information Power vs Saturation
Qualitati Research Team · 2026-07-20 · 12 min read
Short answer: Sample size in qualitative interview studies is justified by information power or saturation, not statistical power. Information power (Malterud, Siersma & Guassora, 2016) says the more relevant information your sample holds, the fewer participants you need. Five dimensions set it: study aim, sample specificity, use of established theory, quality of dialogue, and analysis strategy.
Last updated: October 9, 2026 (first published July 20, 2026)
Every PhD student eventually gets the question from a supervisor, an ethics board or Reviewer 2: why this many participants? "Until saturation" is the most common answer, and often the weakest, because saturation can only be shown after the data are in. This guide explains how to justify sample size in qualitative interview studies with information power, how that differs from saturation, which qualitative sampling strategies decide who you recruit, and how to write a sample-size justification that survives review. It includes a worked justification paragraph and a one-page worksheet you can paste into a proposal.
Key takeaways
- Information power is a model for planning and justifying qualitative sample size: the more information relevant to your aim each participant holds, the fewer participants you need.
- Its five dimensions are aim (narrow vs broad), sample specificity (dense vs sparse), established theory (applied vs none), quality of dialogue (strong vs weak) and analysis strategy (case-based vs cross-case).
- Information power is used before and during fieldwork; saturation is assessed during and after. A good justification uses information power for the plan and reports what happened against it.
- Empirical saturation studies give useful anchors: a 2022 systematic review found saturation within 9–17 interviews in relatively homogeneous samples with narrow aims (Hennink & Kaiser).
- Sample size is the second question. The first is who: purposive sampling strategies such as maximum variation, criterion and theoretical sampling decide whether your participants can answer the question at all.
- AI-moderated interviewing removes the moderator-hours ceiling on n, which makes a written justification more important, not less.
What is information power in qualitative research?
Information power is a model for deciding how many participants a qualitative interview study needs. Proposed by Kirsti Malterud, Volkert Siersma and Ann Dorrit Guassora in Qualitative Health Research (2016), it holds that a sample's adequacy depends on how much information relevant to the study aim it contains. Higher information power means fewer participants; lower information power means more.
The model was written as an alternative to treating saturation as the default rule. It does not produce a number. It structures an argument: you rate your study on five dimensions, set a provisional target, and revise it as you see the quality of the interviews coming in. The original paper presents it as a guide for an assessment that continues during data collection, not a replacement for judgment (Malterud et al., 2016, DOI 10.1177/1049732315617444).
The five dimensions of information power
Each dimension pushes the required sample size up or down. The table restates the model in planning terms and adds the question a reviewer is likely to ask.
| Dimension | Fewer participants when… | More participants when… | Reviewer's question |
|---|
| Aim | The aim is narrow (one experience, one decision, one role) | The aim is broad or exploratory across many contexts | Is the aim specific enough to bound the sample? |
| Sample specificity | Participants have dense, first-hand experience of the phenomenon | Participants are only loosely connected to the phenomenon | Why are these people the right informants? |
| Established theory | An existing theory frames sampling and analysis | No theory is applied; the study is purely inductive | Which theory, and how does it guide the questions? |
| Quality of dialogue | Interviews are deep, well-probed and clearly focused | Interviews are short, shallow or poorly probed | How do you know the interviews were rich? |
| Analysis strategy | In-depth analysis of narratives or a few cases | Exploratory cross-case analysis across many participants | Does the sample fit the analytic method? |
Note the fourth dimension. Interview quality is a sample-size variable. Shallow interviews mean you need more of them. This is why unprobed, form-like interviews and short chat transcripts are expensive in a hidden way: each one carries less information.
Information power vs saturation: what is the difference?
Saturation asks whether new data are still producing new insight; information power asks whether the sample holds enough relevant information for the aim. Saturation is judged during and after analysis. Information power can be argued before fieldwork, which makes it better suited to proposals, ethics applications and pre-registrations. Most strong methods sections use both.
| Criterion | Information power | Saturation |
|---|
| Core question | Does the sample hold enough relevant information for this aim? | Are new interviews still adding new codes or meaning? |
| When it is used | Planning, then monitored during fieldwork | During analysis; confirmed after |
| Gives a number? | No: a reasoned, revisable target | No: a stopping point observed in the data |
| Best fit | Proposals, ethics boards, reflexive and interpretive designs | Coding-reliability and codebook approaches; grounded theory (as theoretical saturation) |
| Main weakness | Judgment-based; can be used to rationalize a convenient n | Often claimed without evidence of how it was assessed |
The debate matters for thematic analysis in particular. Braun and Clarke argue that saturation fits poorly with reflexive thematic analysis, where meaning is generated through interpretation rather than discovered in the data, and point to information power as a more coherent rationale (Braun & Clarke, 2021, Qualitative Research in Sport, Exercise and Health). If you use reflexive TA, citing saturation as your stopping rule invites an epistemological objection. For a fuller treatment of saturation itself, see what is data saturation and how many AI-moderated interviews you need.
What the empirical evidence says about numbers
Information power does not give a number, but reviewers often want an anchor. Two reviews are useful:
- Hennink & Kaiser (2022) systematically reviewed 23 empirical tests of saturation. Studies using empirical data reached saturation within 9–17 interviews or 4–8 focus groups, mostly with relatively homogeneous populations and narrowly defined objectives (Social Science & Medicine, vol. 292).
- Wutich, Beresford & Bernard (2024) reviewed sample-size guidance for 10 types of qualitative analysis and give separate anchors for theme saturation (about 9 interviews), meaning saturation (about 24 interviews) and theoretical saturation (20–30+ interviews) (International Journal of Qualitative Methods, vol. 23).
Read these as conditional findings, not quotas. Both reviews tie the lower figures to homogeneous samples and narrow aims, which is exactly what high information power describes. A broad, heterogeneous, theory-free study should expect to sit above these ranges.
The Information Power Worksheet (Qualitati template)
Use this one-page worksheet in a proposal, ethics application or methods chapter. Rate each dimension, write one sentence of evidence, and only then set a provisional target. It turns a judgment into a reviewable argument.
| Dimension | Rating | Evidence (one sentence) | Effect on n |
|---|
| Aim | Narrow / Moderate / Broad | State the aim and what is out of scope | ↓ / – / ↑ |
| Sample specificity | Dense / Mixed / Sparse | State the inclusion criteria that guarantee first-hand experience | ↓ / – / ↑ |
| Established theory | Applied / Partial / None | Name the theory and how it shapes the guide | ↓ / – / ↑ |
| Quality of dialogue | Strong / Uncertain / Weak | Pilot evidence: interview length, probing depth, interviewer experience | ↓ / – / ↑ |
| Analysis strategy | Case-based / Mixed / Cross-case | Name the analytic method (IPA, reflexive TA, framework, grounded theory) | ↓ / – / ↑ |
| Provisional target | State a range (not a single number), the review point (for example, after the first third of interviews) and what would make you extend |
A worked sample-size justification
Here is an example paragraph for a hypothetical study of how first-year PhD students in management decide whether to use generative AI in their literature reviews:
"Sample size was guided by information power (Malterud et al., 2016). The aim was narrow (one decision, in one population), the sample was dense (all participants had used or explicitly rejected generative AI for a literature review in the past six months), and interviews were framed by the technology acceptance literature. Against this, the cross-case analysis strategy and our limited pilot evidence on dialogue quality lowered information power. We therefore set a provisional target of 14–20 interviews, reviewed information power after the first seven, and would extend recruitment if new decision pathways were still appearing. We completed 17 interviews; the final three added no new decision pathways."
Three features make this work: it names all five dimensions, it gives a range with a review point, and it reports what actually happened. The numbers are illustrative; yours must come from your own study.
Qualitative sampling strategies: deciding who before how many
Information power assumes your participants are the right participants. That is the job of the sampling strategy. Oliver Robinson's guide to sampling in interview-based research sets a four-step order: define the sample universe (inclusion and exclusion criteria), decide sample size, choose a sampling strategy, then source the sample (Robinson, 2014, Qualitative Research in Psychology).
| Strategy | What it does | Use it when | Main risk |
|---|
| Maximum variation | Deliberately spans different contexts or extremes | You need the range of experience, or patterns that hold despite difference | Thin coverage of each segment if n is small |
| Homogeneous | Concentrates on one tightly defined group | You need depth on one role, journey or condition (common in IPA) | Findings do not travel outside that group |
| Criterion | Includes everyone who meets a defined threshold or event | Studying a specific experience (a diagnosis, a dropout, a failed adoption) | Criterion may be measured badly |
| Theoretical | Recruits iteratively based on emerging analysis | Grounded theory; concept development | Requires analysis to run alongside fieldwork |
| Extreme / deviant case | Targets unusual successes or failures | You want mechanisms at their most visible | Atypical cases read as the norm |
| Snowball | Participants refer others | Hidden or hard-to-reach populations | Referral chains cluster socially |
| Stratified purposive | Sets targets per segment, purposive within each | You must compare defined groups | Quotas filled with weak-fit participants |
| Convenience | Whoever is available | Pilots and instrument testing only | Systematic bias; weak claims |
Strategy and information power interact. Stratified and maximum-variation designs lower sample specificity per segment, so they need larger totals. A homogeneous criterion sample raises specificity and supports a smaller n. Vasileiou and colleagues' review of interview studies found that sample-size justifications in published papers were often absent or thin, which is why naming both the strategy and the information-power reasoning stands out (Vasileiou et al., 2018, BMC Medical Research Methodology).
Sampling quality checklist
- Inclusion and exclusion criteria written before recruitment opened.
- Sampling strategy named in the protocol and in the final report.
- Information Power Worksheet completed, with a target range and a review point.
- Per-segment targets set when the design compares groups.
- Screener tested on a few known-good and known-bad profiles.
- Achieved sample reported against the target, including shortfalls.
- Missing groups named in the limitations section.
What AI-moderated interviews change about sample size
Historically, the number of interviews was capped by moderator hours, and twelve to fifteen interviews was partly a budget artifact. AI-moderated interviewing changes that: interviews can run in parallel, across time zones and languages. The methodological consequence is often misread.
- Cheaper interviews do not lower the bar. A convenience sample of 300 is still a convenience sample. Volume does not correct selection error.
- Dialogue quality becomes the key dimension. If the AI moderator probes poorly, each transcript carries less information and you need more of them. Pilot a few interviews and read them before scaling.
- Theoretical sampling needs waves. Launching every interview at once removes the chance for early analysis to change who you recruit next. Field in batches.
- Screening becomes load-bearing. Open recruitment at scale raises exposure to fraudulent or professional respondents. See screening fraudulent participants in AI interviews.
Where Qualitati fits
Qualitati is a European, budget-friendly qualitative research platform for universities and research companies, with data privacy and GDPR as central priorities. For sample-size and sampling decisions specifically:
- AI-moderated interviews in text and voice run in parallel, so stratified designs with real per-segment targets and wave-based theoretical sampling are practical on a research budget.
- Active Listener mode gives a human interviewer real-time prompts and section tracking, which suits small homogeneous or IPA-style samples where dialogue quality carries the information power.
- Conversational surveys with AI follow-ups and branching logic can act as a screening layer that feeds a purposive interview sample.
- Multilingual research in 10 languages (English, Chinese, French, Norwegian, Dutch, German, Spanish, Portuguese, Japanese and Arabic) widens the sample universe beyond English speakers.
- ThemeLens analyzes up to 100 transcripts at once, mapping codes to research questions, so you can check after each wave whether new interviews are still adding meaning. The QDA Workspace supports inductive and deductive coding if you prefer to code yourself.
Pricing is published per credit, and the free tier includes 30 monthly credits with no credit card required. See pricing for current tiers.
Limitations and trade-offs
- Information power is a judgment framework, not a formula. It structures an argument; it does not compute a sample size. Two researchers can rate the same study differently.
- It can be misused. Rating every dimension "high" to justify a convenient, small n is a known failure. Show evidence for each rating, especially dialogue quality.
- Empirical anchors are conditional. The 9–17 interview range applies to homogeneous samples with narrow aims, not to every study.
- Paradigm matters. Saturation language fits some designs (codebook TA, grounded theory) and clashes with others (reflexive TA, IPA). Match the rationale to the method.
- AI does not settle adequacy. Automated analysis can show whether new codes are appearing, but whether the sample answers the research question remains a researcher's judgment.
- Human-review note: sample-size justifications, saturation claims and transferability statements should be reviewed by a qualified researcher against your field's reporting standard (for example COREQ or SRQR) before submission.
Who this is for, and when not to use it
For: PhD students writing a methods chapter or ethics application, research labs planning interview studies, and qualitative or market research firms defending a sample to a client. Not for: studies whose claims depend on statistical generalization to a population. If you need prevalence estimates, use a probability sample and a survey, then add interviews to explain the results.
Frequently asked questions
What is information power in qualitative research?
Information power is a model, proposed by Malterud, Siersma and Guassora in 2016, for deciding how many participants a qualitative interview study needs. The more information relevant to the study aim the sample holds, the fewer participants are needed. It is assessed on five dimensions: aim, sample specificity, theory, dialogue quality and analysis strategy.
Is information power better than saturation?
Neither is universally better. Information power is easier to argue before fieldwork, so it suits proposals and ethics boards, and it fits interpretive methods such as reflexive thematic analysis. Saturation is useful for codebook and grounded theory designs. Many strong studies plan with information power and report saturation evidence afterwards.
How many interviews do I need for a qualitative study?
There is no universal number. Empirical reviews found saturation within 9–17 interviews in homogeneous samples with narrow aims, but broad or heterogeneous studies need more. Set a range using the five information-power dimensions, review it partway through fieldwork, and report the reasoning.
How do I write a sample-size justification for a qualitative study?
Name the framework (information power or saturation), rate each of the five dimensions with one sentence of evidence, give a provisional range and a review point, and report the achieved sample and why you stopped. The worked paragraph in this guide is a template.
What is the most common sampling strategy in qualitative research?
Purposive sampling: selecting participants deliberately because their experience fits the research question. Maximum variation, homogeneous and criterion sampling are its most frequently used variants.
Do AI-moderated interviews change how many participants I need?
They change cost, not adequacy. Information power still depends on aim, sample, theory, dialogue quality and analysis. If AI interviews probe less deeply than a skilled human, you may need more of them; if they probe well, the same logic as human interviews applies.
Bottom line
The strongest answer to sample size in qualitative interview studies is not a number but an argument. Define who belongs in the sample, choose a purposive strategy that fits the research question, rate the five dimensions of information power, set a range with a review point, and report what happened. AI-moderated interviews remove the cost ceiling on n, which is exactly why that written justification now carries more weight.
Planning an interview study? Start free with 30 credits, no credit card required, or view transparent pricing. See how Qualitati runs AI-moderated interviews, focus groups, conversational surveys and thematic analysis, or compare it with NVivo.
Image alt text suggestion: "Information Power Worksheet: five dimensions (aim, sample specificity, established theory, quality of dialogue, analysis strategy) rated to set a qualitative interview sample size."
External sources: Malterud, Siersma & Guassora (2016), Sample Size in Qualitative Interview Studies: Guided by Information Power (Qualitative Health Research); Braun & Clarke (2021), To saturate or not to saturate? (Qualitative Research in Sport, Exercise and Health); Hennink & Kaiser (2022), Sample sizes for saturation in qualitative research: a systematic review of empirical tests (Social Science & Medicine); Wutich, Beresford & Bernard (2024), Sample Sizes for 10 Types of Qualitative Data Analysis (International Journal of Qualitative Methods); Vasileiou et al. (2018), Characterising and justifying sample size sufficiency in interview-based studies (BMC Medical Research Methodology); Robinson (2014), Sampling in Interview-Based Qualitative Research (Qualitative Research in Psychology).