Constant Comparative Method: the engine of grounded theory

Incident to incident. Early on, each coded incident is compared with other incidents: is this the same thing? a variation? something new? This is where categories are born and where their properties (the ways instances differ) first show themselves.

Incident to category. As categories stabilise, each new incident is compared against the category’s accumulated content — which either fits, stretches the category’s dimensions, or breaks it into two. The category definition is rewritten as it absorbs cases; a category that never changes definition is not being compared against, only filed into.

Category to category. Mature analysis compares categories with each other: which are properties of which, which co-occur, which are rivals for the same incidents. This is where integration happens and redundancy is removed.

A useful check at any moment: can you name the last comparison that changed something — a definition, a boundary, a memo? If not, the constant has lapsed.

The method was introduced by Glaser in The constant comparative method of qualitative analysis and folded into The Discovery of Grounded Theory (Aldine, 1967) as the core analytic procedure — four stages, joint with coding, aimed at generating theory rather than testing propositions.

For use as a standalone technique, Boeije’s A purposeful approach to the constant comparative method is the practical modernisation: it specifies five comparison types (within a single interview; between interviews within a group; between groups; within couples or units; between data types) with the question each answers — a plan you can put in a methods section.

That codification matters because “constant comparison was used” has become qualitative boilerplate; Boeije’s scheme is the difference between naming the method and doing it inspectably.

• Inside grounded theory, comparison drives theoretical sampling and saturation: what to collect next is whatever the current comparisons cannot settle.
• In thematic and framework studies, comparison is the quality mechanism: themes built comparatively have defended boundaries instead of accreted contents.
• In longitudinal designs, within-case-over-time comparison is the analysis: what changed between wave one and wave two, per participant, before any cross-case claims.
• Do not bother invoking it for single-pass deductive coding against a fixed frame — there the comparisons are pre-made by the framework, and claiming the method adds nothing but a citation.

A researcher studied how two customer-service teams (one phone, one chat) handled complaints, with 30 recorded interactions and 12 interviews per team. Coding began with the phone team; the category “owning the fault” emerged early — agents explicitly taking responsibility — with properties of timing (immediate vs after diagnosis) and scope (personal vs organisational).

Incident-to-category comparison on the chat team broke the category: chat agents wrote apology formulas that looked like owning but, compared closely, deferred ownership (“sorry you’re experiencing this”) while routing fault elsewhere. The comparison forced a split into owning and acknowledging without owning — and re-reading phone incidents found both there too, previously conflated.

The between-group pass then produced the finding: the split tracked monitoring regimes, not medium. Chat transcripts were quality-scored against a script that rewarded apology language; phone calls were sampled rarely. Where surveillance was tight, acknowledging replaced owning. The category note’s revision history — definition v1 through v4, each with the incident that forced it — went into the methods appendix as the audit trail.

• Filing instead of comparing. Assigning incidents to categories by label-match, without reading the category’s prior contents, is sorting, not analysis.
• Frozen definitions. Category definitions that never change across a study mean comparison was not allowed to bite.
• No record of boundary decisions. If the near-misses and splits are not memoed, the audit trail is assertion.
• Comparing only across cases. Within-case contradictions are the cheapest analytic gold and the most commonly skipped pass.
• Claiming saturation without comparison evidence. “Nothing new emerged” should be demonstrable as a run of comparisons that changed nothing.

Comparison scales quadratically in principle and painfully in practice: by the fortieth transcript, honest incident-to-category comparison against everything prior is impossible without tooling and ruthless delimiting. The method’s own delimitation stage is the answer, and it must actually be used.

The method also underdetermines interpretation: comparison tells you incidents differ, not what the difference means. It structures judgement; it does not replace the theoretical sensitivity that decides which differences matter.

And as a citation, it is degraded currency — so widely name-checked that reviewers discount it. The remedy is specificity: which comparisons, in what order, recorded where.

Classic grounded theory embeds constant comparison in a full theory-generating cycle with theoretical sampling.

Straussian grounded theory structures the comparisons through open, axial, and selective phases.

Qualitative comparative analysis is a different creature — set-theoretic comparison of configured cases — despite the shared word.

Reflexive thematic analysis benefits from comparative discipline at its theme-development phases without adopting the grounded theory frame.

The method is retrieval-hungry: comparing a new incident against a category means seeing that category’s incidents, instantly, with context. Evidano keeps every code application linked to its source quote, so incident-to-category comparison is a click rather than an archaeology dig, and cross-group passes (phone vs chat, wave one vs wave two) are code-frequency and segment comparisons run on demand.

What stays with the analyst is the boundary judgement — same, variation, or new — and the memo that records why. A tool can lay the incidents side by side; the comparison itself is the researcher thinking.