Grounded theory: a practical guide

Grounded theory develops a theory from the data rather than testing a theory against them. You begin without a hypothesis, code closely, and build upwards until you have an explanatory account of the process you were studying.

Its output is distinctive. Thematic analysis produces themes describing what participants said; grounded theory produces a theory — an account of how something works, what conditions bring it about, and what consequences follow. That difference in output is what should drive the choice of method.

The method originated in the 1960s as a deliberate corrective. Sociology at the time was dominated by grand theorising tested through survey work, and Glaser and Strauss argued that this produced theory disconnected from the settings it claimed to describe. Their alternative was to derive theory from close, systematic engagement with data — hence “grounded”. Understanding that origin helps explain features that otherwise look arbitrary, particularly the suspicion of an early literature review, which was aimed at preventing existing frameworks from being imposed on data that might not fit them.

Grounded theory split into distinct traditions after its founders disagreed, and examiners expect you to know which one you followed.

The Straussian version is the most prescriptive and therefore the easiest to defend in a viva, because each step has a stated procedure. The constructivist version is the most widely used in contemporary social science and is the most comfortable fit for a researcher who cannot plausibly claim to have approached the topic with no prior knowledge.

Two features distinguish genuine grounded theory, and a study lacking either is thematic analysis under a different name.

Open coding

Work line by line or incident by incident, naming what is happening in the data. Codes stay close to the material and are often gerunds — “managing uncertainty”, “deferring to expertise” — because grounded theory is interested in action and process rather than in topics. Expect a large number of codes at this stage; several hundred from a handful of transcripts is normal.

Axial coding

Group codes into categories and, crucially, work out how the categories relate. The Straussian coding paradigm offers a scaffold for this: for each category, identify the conditions that give rise to it, the actions and interactions involved, and the consequences that follow. Not every analysis needs the paradigm applied mechanically, but the relational question — how does this connect to that? — is what separates axial coding from sorting.

Selective coding

Identify the core category: the one that accounts for the most variation and to which the others can be related. Everything else is then integrated around it, and the resulting account is the theory. If no core category emerges, the analysis is not finished, and reporting a list of parallel categories instead is the point at which many grounded theory studies quietly become thematic analyses.

Software helps with the mechanics but does not do the analysis. NVivo, MAXQDA and Dedoose all manage codes, retrieve extracts and store memos efficiently, which matters once you have several hundred codes across twenty transcripts. None of them identifies a core category for you, and a coding tree that looks tidy in software can still be a description rather than a theory.

These are the two engines of the method, and both are procedural rather than conceptual — you either did them or you did not.

Theoretical sampling

After the first few interviews you analyse, notice a gap or an ambiguity in the emerging categories, and choose the next participants specifically to address it. If a category about “managing uncertainty” is developing but every participant so far has been experienced, you deliberately recruit novices next to see whether the category holds.

This means your sampling strategy cannot be fully specified in advance, which has implications for ethics applications. Write the protocol to allow it: state that sampling will be theoretical, that later participants will be selected on the basis of emerging analysis, and give the broad pool you will draw from.

In practice, theoretical sampling rarely means recruiting entirely new participants each round. It may mean returning to earlier participants with sharper questions, seeking documents that bear on a developing category, or observing a setting you had not planned to visit. What defines it is not the source but the reasoning: the choice follows from what the analysis needs, and that reasoning is recorded.

Constant comparison

Every new piece of data is compared against the codes and categories already developed, and every code against the others. The comparisons are what force categories to become precise: encountering a case that does not fit obliges you either to redefine the category or to identify the condition under which it does not apply.

Theoretical saturation is reached when further data no longer add to the properties of your categories. It is a property of the categories, not of the number of interviews, and the distinction matters when you come to defend it.

Sample sizes in published grounded theory studies commonly fall between 20 and 40 participants, but citing a number is not a justification. What defends the decision is describing the point at which categories stopped developing and showing the evidence — typically that the last several interviews contributed nothing new to any core category.

Memos are analytic notes written throughout, and in grounded theory they are not optional. They are where the theory is actually developed; the codes are only its raw material.

Write them immediately, date them, and do not edit them for style. Their value lies in capturing thinking as it develops, including the ideas later abandoned — which is precisely what lets you reconstruct and defend the analytic path in a viva.

There is no minimum, but a study producing a handful of memos across several months has probably not been doing the analytic work the method requires. Many researchers find they write more memo text than they eventually write findings text, and that is a reasonable sign the analysis is developing rather than merely accumulating.

A study explores how newly qualified practitioners manage situations beyond their competence. Three extracts from different participants, and the codes assigned during initial coding.

Moving to a category

Every code involves a tension between needing information and managing how the need appears to others. Grouping them produces a candidate category: managing visible uncertainty. Note that it is stated as a process, not a topic — “uncertainty” alone would describe a subject area, while “managing visible uncertainty” names something participants are doing.

Developing its properties

What theoretical sampling does next

Every participant so far has been newly qualified. To test whether the category is specific to inexperience or persists, the next round deliberately recruits practitioners with ten or more years in post. If they describe the same management of visible uncertainty, the category is broader than first assumed and the theory must account for what sustains it. If they do not, the conditions under which it operates become part of the theory.

1. Collecting all the data before coding any of it

This removes theoretical sampling entirely, and with it the defining procedure of the method.

2. Reporting categories with no core category

A list of parallel categories is a thematic analysis. Grounded theory requires integration around a core.

3. Mixing schools

Name your tradition and apply it consistently. Combining Glaser and Corbin invites exactly the question you cannot answer.

4. Claiming saturation without evidence

Say which categories saturated and what the last interviews added.

5. Codes that are topics rather than processes

Topic codes produce a description. Process codes produce a theory.

6. No memos

Without them there is no record of how the theory was built, and no defence of it.

Structure the findings around the theory rather than around the interview schedule. A results chapter organised by question order reveals that the analysis stayed descriptive; one organised around the core category and its relationships demonstrates that a theory was actually built.