Grounded theory is a method, not just a definition — a specific sequence of moves a researcher makes, in a specific order, that together produce a theory grounded in data rather than imposed on it. This guide walks through that sequence: open coding, axial coding, and selective coding; theoretical sampling and the constant comparative method that drive what gets collected next; memoing, which is where the actual thinking happens; and theoretical saturation, which is how a grounded theory study ends. For the criteria that make a study grounded theory in the first place, see the Grounded Theory dictionary entry; this page is about doing it.
The basic loop: collect, code, compare, sample again
Grounded theory does not separate data collection from analysis into two phases. Instead, the researcher moves through a repeating loop: collect some data, code it, compare the new codes against everything coded so far, let that comparison suggest what to collect next, and repeat — until new data stops changing the emerging categories. Three mechanics make that loop work, and each has its own name in the literature:
- Coding — labeling segments of data (an interview excerpt, a field note, a document passage) with a short analytic tag, then building those labeled segments up into categories. Coding runs in three stages, covered below: open, axial, selective.
- The constant comparative method — every new incident, excerpt, or case is compared against codes and categories already developed, not coded in isolation. Comparison is what keeps categories analytically sharp instead of becoming a loose pile of similar-sounding labels.
- Theoretical sampling — the decision about who or what to sample next is made by what the developing categories need (a case likely to test, extend, or contradict a category), not by a sampling plan fixed in advance. This is the mechanism that connects coding back to data collection and keeps the loop turning.
The loop ends at theoretical saturation: the point at which new data stops adding new properties to the core categories. Saturation is a property of the categories, not a fixed number of interviews — see the saturation section below for what that means in practice.
Open coding: labeling what’s in the data
Open coding is the first pass. The researcher works through the data — line by line, or incident by incident — and attaches a short, descriptive label to each meaningful segment, staying close to what’s actually there rather than jumping to an interpretation. Early open codes are often provisional and numerous; the same underlying idea might get three or four slightly different labels across the first few transcripts before the researcher notices they’re the same thing and starts collapsing them.
A common variant, associated particularly with Charmaz’s constructivist approach, is in vivo coding: using the participant’s own words as the code label rather than the researcher’s paraphrase, which keeps the analysis closer to participants’ meanings during this early stage.
Illustrative example (not drawn from a real study — constructed to show the mechanics): a segment reading “I just stopped telling my supervisor about problems, there was no point” might be open-coded as withholding information from supervisor or, in vivo, as "no point telling them". Neither label yet claims anything about why — that comes later, in axial coding.
Axial coding: relating categories to each other
Axial coding takes the open codes and starts relating them to one another — grouping fragmented open codes into broader categories and specifying how those categories relate: what conditions give rise to a category, what actions or interactions people take in response, and what consequences follow. This conditions–actions/interactions–consequences structure is the “paradigm model” associated with Strauss and Corbin’s version of grounded theory (Basics of Qualitative Research, 1990, later editions 1998/2008/2015) and is the most widely taught account of what axial coding is for.
Continuing the example above: withholding information from supervisor might, once compared against other excerpts, turn out to be one instance of a broader category — say, disengagement after unaddressed feedback — with a condition (feedback given, no visible response from supervisor), an action (participant stops volunteering information), and a consequence (supervisor loses visibility into emerging problems). Axial coding is where the analysis starts making claims about relationships, not just naming what’s present.
Selective coding: integrating around a core category
Selective coding is the final coding stage: the researcher identifies a core category — the central phenomenon the study is actually about — and systematically relates every other category to it, refining and, where necessary, dropping categories that don’t connect to the core story. The output of selective coding is the theory itself: a coherent account, built from the categories and their relationships, of the process or pattern under study. This is also where theoretical memos (see below) usually get pulled together into the connected narrative that becomes the write-up.
Open, axial, and selective coding compared
| Stage | What it does | Typical question the researcher is asking |
|---|---|---|
| Open coding | Labels discrete segments of data with descriptive codes, staying close to the data | “What is this an instance of?” |
| Axial coding | Groups open codes into categories and specifies conditions, actions/interactions, and consequences linking them | “How do these categories relate to each other?” |
| Selective coding | Identifies the core category and integrates all other categories around it into a coherent theory | “What is this study, as a whole, a theory of?” |
For a broader treatment of coding mechanics that applies beyond grounded theory specifically — first-cycle vs. second-cycle coding, inter-coder reliability, coding software workflows — see Coding Qualitative Interview Data: A Worked-Example Guide.
Theoretical sampling: letting categories drive who’s next
Theoretical sampling is the sampling logic specific to grounded theory, and it is easy to confuse with purposive sampling because both involve deliberately choosing participants rather than sampling at random. The difference is timing and driver: a purposive sample is typically specified, in full, before data collection starts, based on characteristics the researcher believes matter. A theoretical sample is built incrementally, one decision at a time, during analysis — each next case is chosen because the categories developed so far suggest it will test, extend, or challenge them. A researcher might not know who the fourth or fifth participant needs to be until the first three have been coded and compared. See Purposive Sampling vs. Convenience Sampling for how theoretical sampling differs from sampling approaches decided up front.
The constant comparative method
The constant comparative method is the analytic engine that makes theoretical sampling possible: it’s the discipline of comparing every new incident against the codes and categories already built, rather than coding each new transcript in isolation and comparing only at the end. Glaser and Strauss’s original four-step summary (The Discovery of Grounded Theory, 1967) is still the clearest statement of the sequence:
- Compare incidents applicable to each category, as they emerge.
- Integrate categories and their properties.
- Delimit the theory — narrowing focus as categories stabilize and the theory takes shape.
- Write the theory, grounded in and traceable back to the comparisons that produced it.
In practice, constant comparison happens at several levels simultaneously: incident-to-incident (does this excerpt fit an existing code, or does it need a new one?), incident-to-category (does this case still fit the category as currently defined, or does the category need to be split or redefined?), and category-to-category (how do these categories relate once several are established?).
Memoing: where the analytic thinking gets written down
A memo is a piece of analytic writing the researcher produces about the data and the emerging categories — not a research note about logistics, but a record of the researcher’s own thinking as it develops: why a code was created, what a category seems to mean, what a comparison revealed, what a category might be a variant of. Memos are written throughout the study, from the first open codes through selective coding, and they accumulate into raw material for the eventual write-up. Grounded theory methodologists generally treat memoing as non-optional: skipping it is one of the more common ways a study drifts from grounded theory into simple thematic description, because the analytic reasoning behind category decisions never gets captured or interrogated. Memos are typically dated and kept alongside (not mixed into) the coded data, precisely so the researcher can trace how a category’s definition changed over the course of the study.
Theoretical saturation: how a grounded theory study ends
Saturation is reached when new data collection stops adding new properties, dimensions, or relationships to the core categories — not when a target number of interviews has been reached. This distinction matters because “data saturation” is sometimes used loosely, across qualitative methods generally, to mean roughly “we stopped hearing new themes.” In grounded theory specifically, saturation is tied to the categories built through coding and constant comparison, and it’s assessed category by category: a study can have some categories that saturated early (well-supported, stable, no new properties appearing) and others still developing, which is itself a signal for where theoretical sampling should focus next. There is no fixed sample size that guarantees saturation — it depends on the scope of the phenomenon, the diversity of the sample, and how tightly the research question is bounded, which is why grounded theory studies report saturation as a methodological judgment rather than cite it via a predetermined participant count.
Glaser, Strauss and Corbin, and Charmaz: three versions of grounded theory
“Grounded theory” is not one settled procedure — it split, early and consequentially, into variants that disagree on how structured the coding process should be and what the researcher’s relationship to the data is. The disagreement between Glaser and Strauss after their joint 1967 book is well documented in the methods literature and is worth knowing before picking a variant, because reviewers and methodology instructors often expect a study to name which one it’s using and stay consistent with it.
| Dimension | Glaserian (classic) | Straussian | Constructivist (Charmaz) |
|---|---|---|---|
| Key text | Glaser, Theoretical Sensitivity (1978) | Strauss & Corbin, Basics of Qualitative Research (1990; later eds. 1998/2008/2015) | Charmaz, Constructing Grounded Theory (2006; 2nd ed. 2014) |
| Coding structure | Looser, emergent; categories should “emerge” from data with minimal forcing | More prescribed: explicit open → axial → selective sequence, axial coding via the conditions–actions/interactions–consequences paradigm model | Keeps coding, memo-writing, and theoretical sampling but treats them as interpretive tools, not a fixed procedure |
| Researcher’s role | Neutral discoverer of a theory latent in the data | Systematic analyst applying a defined procedure | Co-constructor of meaning, in interaction with participants and their own position |
| Prior literature review | Discouraged before data collection, to avoid forcing existing concepts onto the data | More accepting of engaging literature earlier, alongside analysis | Engaged reflexively; the researcher’s disciplinary lens is treated as part of the analysis, not a contaminant to eliminate |
| Underlying stance | Broadly positivist — theory as discovered | Post-positivist / systematic | Constructivist — theory as jointly constructed, not discovered |
None of the three is more “correct” than the others in the abstract; the choice should follow the research question and be stated explicitly and applied consistently, since mixing procedural elements from different variants without acknowledging it is a common and avoidable methodological weakness reviewers flag.
Common mistakes
- Treating grounded theory as a label for “we did qualitative coding.” A study that codes interview data thematically but collects all its data up front, doesn’t sample theoretically, and doesn’t build toward an explanatory theory is doing thematic analysis, not grounded theory, regardless of what it calls itself. See Thematic Analysis: A Step-by-Step Guide to Braun and Clarke’s Six Phases for that method on its own terms.
- Skipping memos. Coding without memoing produces categories with no recorded reasoning behind them, which makes it difficult to defend category decisions later or trace how a category changed.
- Citing a saturation number instead of a saturation judgment. Reporting “saturation was reached after 15 interviews” without describing what stopped changing in the categories reads as a borrowed convention rather than an actual methodological claim.
- Mixing variants without saying so. Using Glaserian language (“categories emerged”) alongside a rigidly Straussian axial-coding paradigm model, without acknowledging the blend, confuses reviewers who know the literature.
- Front-loading the literature review in a Glaserian study. Classic grounded theory specifically cautions against reviewing the substantive literature before coding begins, to avoid forcing existing concepts onto the data — a step that’s fine, even expected, under the Straussian or constructivist variants.
Software for grounded theory coding
Grounded theory’s iterative coding and constant comparison are typically supported with qualitative data analysis software rather than manual methods once a project has more than a handful of transcripts, mainly because the software makes it tractable to retrieve every excerpt tagged with a given code as categories are revised. See Qualitative Data Analysis Software: NVivo vs. ATLAS.ti vs. MAXQDA for a comparison, or the NVivo guide specifically.
Frequently asked questions
Is grounded theory qualitative or quantitative? Qualitative. It builds a theory inductively from non-numerical data such as interviews, observations, or documents, through coding and constant comparison, rather than testing a hypothesis against numerical data.
How many interviews do you need for a grounded theory study? There’s no fixed number — sample size is determined by theoretical saturation, not a target count. Published grounded theory studies commonly range from roughly 15 to 30-plus interviews, but the actual stopping point is when new data stops adding new properties to the core categories, which varies with the scope of the research question.
What’s the difference between open coding and axial coding? Open coding labels discrete segments of data with descriptive codes, staying close to what’s literally present. Axial coding takes those open codes and relates them into broader categories, specifying the conditions, actions or interactions, and consequences that connect them.
Do you need a literature review before starting grounded theory? It depends which variant you’re using. Classic (Glaserian) grounded theory discourages reviewing the substantive literature before coding, to avoid forcing existing concepts onto the data. Straussian and constructivist variants are more accepting of engaging literature earlier.
What is theoretical sampling, and how is it different from purposive sampling? Theoretical sampling is decided incrementally, during analysis, based on what the emerging categories need next. Purposive sampling is typically specified in full before data collection begins, based on characteristics the researcher believes matter in advance.
What is a core category in grounded theory? The central category identified in selective coding, around which every other category is related and integrated to form the study’s theory. It’s the answer to “what is this study, as a whole, a theory of?”
