Classic Grounded Theory: the Glaserian approach

Grounded theory began as one method, published by Glaser and Strauss in The Discovery of Grounded Theory (Aldine, 1967), and then split. Strauss, with Juliet Corbin, proceduralised it; Charmaz rebuilt it on constructivist foundations; Glaser kept the original and called it classic.

The classic position has three commitments the other versions relax. Emergence over forcing: the problem, the categories, and the theory come from the data, not from a literature review or an interview guide built around prior concepts. All is data: interviews, observations, documents, statistics — whatever bears on the emerging theory is admissible. Conceptualisation over description: the product is a theory about a core category and its relationships, abstract of time, place, and people — not a rich description of a setting.

The split with Strauss is not cosmetic. Glaser’s account of the divergence and of forcing argues that structured coding paradigms produce “full conceptual description” rather than grounded theory. Understanding the argument matters practically, because supervisors and reviewers from different lineages expect different procedures under the same name.

• Use it where the field lacks theory. Its home ground is social processes nobody has yet conceptualised — how people manage a chronic illness at work, how volunteers decide to quit.
• Use it when you can genuinely sample iteratively. The method requires collecting, analysing, then deciding what to collect next. If all data must be gathered before analysis (a fixed ethics protocol, an archived dataset), the engine is missing.
• Do not use it to test or extend an existing framework. That is deductive work wearing the wrong label; directed content analysis or theory-driven coding fits better.
• Do not use it if you need description. A needs assessment or an implementation study wants specifics of this programme, these people — grounded theory abstracts away from exactly that.

A researcher began with an area — families and dementia — and no fixed question. Early interviews with six adult children were open coded; the incidents kept indicating monitoring, testing, and comparing notes with siblings. The emerging main concern was not grief but uncertainty about competence: is Dad still safe to drive, to cook, to sign?

Theoretical sampling followed the theory, not demographics: next came children who lived far away (how is monitoring done remotely?), then a sibling pair in open conflict (what happens when assessments diverge?), then two family GPs, because the category “recruiting authoritative confirmation” had emerged and needed the professional side. Twenty-three interviews and two observed family meetings later, new data were confirming rather than adding — saturation.

The core category was calibrating capability: an ongoing family process of testing, comparing, and re-setting what the parent is trusted to do, with siblings’ divergent calibrations explaining most conflict in the data. The final theory — abstract enough to be examined in other dependency contexts — bore no resemblance to the interview guide the researcher would have written on day one, which is the method working as designed.

• A full literature review first, then “emergence”. The classic sequence delays deep engagement with the specific literature until the theory is formed, precisely to protect openness. (Universities often require otherwise; handle it as a minor, bracketed review and say so.)
• A fixed semi-structured guide across all interviews. If interview twenty asks the same questions as interview one, theoretical sampling never happened.
• Convenience sampling relabelled. “We interviewed until saturation” with a sample recruited in one batch is description of exhaustion, not theoretical saturation.
• Multiple core categories. The theory explains variation around one core; three cores mean selection has not been done.
• Description offered as theory. If the findings cannot be stated without naming the setting, conceptualisation stopped too early.
• Skipping memos. The single most predictive failure. No memos, no theory.

The method’s openness is expensive: it demands tolerance for months without a stated question, which funders, ethics boards, and doctoral committees are often unwilling to extend. Much published “grounded theory” is the compromise that results.

Saturation is a judgement with no external test — Saunders and colleagues’ Saturation in qualitative research: exploring its conceptualization and operationalization is the useful corrective to treating it as a checkbox, distinguishing the theoretical saturation Glaser meant from the data-saturation shorthand it has become.

And classic grounded theory’s claim that the researcher can approach data without preconception is philosophically contested; the constructivist revision exists because many methodologists find the claim untenable. Choosing classic means being ready to defend that stance, not just cite it.

Straussian grounded theory trades emergence for procedure: a structured coding paradigm and validation-oriented canons.

Constructivist grounded theory keeps the iterative engine but treats theory as co-constructed rather than discovered.

The constant comparative method is the analytic engine inside all grounded theory variants, and is often borrowed without the rest of the package.

Interpretative phenomenological analysis also works inductively but seeks the texture of individual experience, not an abstract process theory.

Grounded theory’s bookkeeping is heavy — hundreds of provisional codes, incidents to compare across transcripts, memos to link and sort. Tools that keep every code application attached to its incidents make constant comparison retrievable instead of remembered, and Evidano’s coding does this with a clickable quote behind every code, plus fast inductive first passes over new theoretical-sample batches so the comparison work starts the day the transcript lands.

The judgement calls — what the main concern is, when a core category has earned selection, when saturation is real — are the theory-building itself. Software shortens the distance between data and comparison; it does not, and should not, nominate your core category.