Short answer: Theoretical sampling is the grounded theory practice of deciding whom to interview, or what data to collect next, based on the concepts emerging from analysis so far. Instead of fixing the sample in advance, you collect, code and compare in rounds, and each round's gaps tell you where to sample next, until your categories are theoretically saturated.
Last updated: November 2, 2026
\n\nTheoretical sampling is the part of grounded theory that most methods chapters claim and fewest show. A sentence like \"participants were recruited through theoretical sampling\" is common; a record of which concept triggered which recruitment decision is rare. This guide explains what theoretical sampling in grounded theory is, how it differs from purposive sampling, how to run it step by step with interview data, and how to document it so a reviewer can follow your decisions.
\n\nTheoretical sampling is the process of data collection for generating theory, in which the researcher jointly collects, codes and analyses data and decides what to collect next and where to find it, so that the theory can develop. The definition comes from Barney Glaser and Anselm Strauss's The Discovery of Grounded Theory (1967) and has changed little since.
\n\nThe key word is theoretical. You are not sampling people to represent a population. You are sampling incidents, conditions and variations of a concept. If your emerging category is \"workarounds when the official process fails\", the next participant is chosen because they are likely to show a workaround under a different condition, not because they fill a demographic cell.
\n\nTheoretical sampling is one of four linked practices in grounded theory, alongside the constant comparative method, memo writing and theoretical saturation. Comparison shows where a category is thin; memos record why; theoretical sampling goes and gets the data that fills the gap; saturation tells you when to stop.
\n\nThe two are easy to confuse because both choose participants deliberately. The difference is what drives the choice and when it is made.
\n\n| Criterion | Purposive sampling | Theoretical sampling | Convenience sampling |
|---|---|---|---|
| What drives selection | Criteria set by the research question | Concepts emerging from analysis | Access and availability |
| When decided | Before fieldwork | During fieldwork, round by round | As people become available |
| Unit sampled | People or cases | Incidents, conditions, settings or questions | People |
| Requires analysis during collection | No | Yes | No |
| Stopping rule | Planned n or data saturation | Theoretical saturation | Budget or time |
| Best fit | Thematic analysis, IPA, case studies | Grounded theory, theory-building studies | Pilots only |
Most grounded theory studies start purposively and move to theoretical sampling once first categories appear. Helen Scott of Grounded Theory Online argued in a September 2025 post that this early phase is better called open sampling: broad, unstructured data collection within the substantive area until the main concern and a possible core category emerge. Narrowing too early, she argues, slows theory development. For the wider set of sampling strategies, see our guide to qualitative sampling strategies.
\n\nName the version you follow in your methods section. Reviewers read \"theoretical sampling\" differently depending on tradition.
\n\nThe steps below assume an interview study. They draw on Ligita, Harvey, Wicking, Nurjannah and Francis's practical example in Qualitative Research Journal (2020), which describes concrete sampling moves from their own grounded theory study.
\n\nRecruit people who can speak to the substantive area. Keep the guide broad. A small first round is enough; what matters is analysing it before the next one.
\n\nOpen-code the first round and compare incidents with incidents. Write memos on each tentative category: what it is, under what conditions it appears, and what you do not yet know about it. See open, axial and selective coding for the coding side.
\n\nFor each thin category, write the question that the next data must answer. For example: \"Do workarounds look different when the manager knows about them?\" That question, not a quota, defines whom you seek.
\n\nLigita et al. describe moves you can combine:
\nReturning to earlier participants with new questions can also work if your ethics approval allows recontact.
\n\nRun the next small round, code it against existing categories, and update memos. Look actively for negative cases that do not fit; they are often the most useful data you will collect.
\n\nStop when new data no longer add properties or dimensions to your core categories and the relationships between them are well explained. This is stricter than data saturation. Weng Marc Lim argues in the Journal of Global Marketing (published online November 2025) that relying on data saturation alone can close data collection too early, and that theoretical saturation often requires around twice as many interviews. Treat that as one author's heuristic, not a rule. Our post on data saturation covers the distinction in more depth.
\n\nImagine a PhD study on how early-career researchers handle research-data-sharing requirements. This is an illustrative scenario, not a real study.
\nThe sample ends at around 25 people. No one fixed that number in advance; the analysis did.
\n\nDimbath, Ernst-Heidenreich and Roche argue in Forum: Qualitative Social Research (2018) that theoretical sampling often breaks down when sampling decisions go undocumented, and propose \"progression cards\", built on memoing, to record them. The Qualitati log below is a lighter version you can keep in a spreadsheet and paste into an appendix:
\n\n| Field | What to write | Example |
|---|---|---|
| Round and dates | Round number and fieldwork dates | Round 2, two weeks in March |
| Trigger category | The category or relationship that is thin | Strategic minimal compliance |
| Memo reference | The memo that identified the gap | Memo 14 |
| Sampling question | What the next data must answer | Does it appear where sharing norms are strong? |
| Move | New questions, new characteristics, new group or new setting | New characteristics + new question |
| Who or what was sampled | Number and type of participants or data | Five researchers, open-science fields |
| Result | What changed in the category | Split into two conditions: competition vs workload |
| Next decision | Continue, redirect or stop | Redirect to data stewards |
Checklist before you write \"theoretical sampling\" in a methods section:
\nQualitati is a European, budget-friendly qualitative research platform for universities and research companies, with GDPR and data privacy as central priorities. Three parts support theoretical sampling in grounded theory, without doing the sampling for you:
\nThe decision of whom to sample next, and why, stays with the researcher. You can start free with 30 monthly credits, no credit card required, or view transparent pricing. If you are moving from desktop QDA software, see Qualitati vs NVivo.
\n\nHuman-review note: this guide summarises the cited methodological sources; check your discipline's and supervisor's expectations before finalising a methods chapter.
\n\nPhD students and research labs running grounded theory or other theory-building interview studies, and research firms that need a documented rationale for an iterative sample.
\n\nIf you are describing experiences with reflexive thematic analysis, testing predefined hypotheses, or need a fixed sample for comparison across pre-set groups, purposive or stratified sampling is a better fit. See thematic analysis vs grounded theory.
\n\nIt means choosing your next participants or data based on what your analysis so far still needs to explain, rather than on a sample plan fixed at the start.
\n\nNo. Both are deliberate, but purposive sampling uses criteria set before fieldwork, while theoretical sampling uses concepts emerging during analysis. Many studies start purposive and become theoretical.
\n\nThere is no fixed number. You stop at theoretical saturation, when new data no longer add properties to your core categories. Report your rounds and stopping reasoning rather than a target n.
\n\nYes, in other theory-building designs, but say what you borrowed. Without constant comparison and memos, it is hard to show that sampling followed the theory.
\n\nNot on their own. AI can speed up interviews, transcription and coding so rounds are shorter, but deciding which concept to pursue and whom to recruit depends on judgement and fieldwork access.
\n\nTheoretical sampling in grounded theory is simple to state and demanding to do: analyse before you recruit again, let thin categories decide the next round, and stop when the theory, not the codebook, is saturated. Keep a decision log, and your methods section will show the process instead of naming it. To run interviews in short, analysable rounds, start free with 30 credits or view transparent pricing.
\n","related":[{"slug":"llm-survey-respondents-semantic-baselines-2026","title":"Do LLM Survey Respondents Beat Simple Baselines? (2026 Test)","description":"Do LLM survey respondents beat simple semantic baselines? A 2026 test of three LLMs on two surveys found their edge vanished with only 20% of answers known.","category":"Methodology","date":"2026-11-02","readTime":"7 min read","author":"Qualitati Research Team"},{"slug":"llm-vs-student-coders-open-ended-survey-responses-2026","title":"Can LLMs Code Open-Ended Survey Answers Better Than Humans?","description":"Can LLMs code open-ended survey responses? A 2026 study of 1,419 answers found Claude and GPT-5.2 beat trained student coders, with kappa up to 0.72.","category":"Methodology","date":"2026-10-30","readTime":"7 min read","author":"Qualitati Research Team"},{"slug":"smart-coder-framework-llm-qualitative-coding-2026","title":"Does Chain-of-Thought Help AI Qualitative Coding? A 2026 Test","description":"Does chain-of-thought help AI qualitative coding? A 2026 study of 72 interviews found prompt calibration mattered most. Key results, table and workflow.","category":"Methodology","date":"2026-10-29","readTime":"7 min read","author":"Qualitati Research Team"}]}