Gioia Method: inductive theory with a defensible data structure

First-order analysis stays in informant language: codes that preserve participants’ own terms and distinctions, deliberately proliferating — a hundred-plus categories early on is normal and healthy.

Second-order analysis moves to theoretical language: the researcher asks what concepts the first-order categories suggest, whether some are facets of one theme, and — the Gioia signature question — whether there is something here the literature does not already have.

Aggregate dimensions distil the second-order themes into the two-to-four overarching constructs the emergent theory runs on.

The data structure displays all three tiers with their linkages. It is a static snapshot of the analysis — evidence of the interpretive chain — and explicitly not the theory. The theory is the dynamic model: how the dimensions relate in motion, usually drawn as a process figure with arrows earned in the data.

The definitive statement is Gioia, Corley and Hamilton’s Seeking qualitative rigor in inductive research: Notes on the Gioia methodology, which sets out the assumptions — informants as knowledgeable agents whose terms deserve preservation — and the working procedures.

Two stances distinguish it from neighbouring approaches. First, the semi-ignorance discipline: engage the literature, but hold its categories loosely enough that they cannot pre-format the first-order work. Second, the informant-centric evidence rule: second-order claims must be demonstrably built from first-order terms, which is what the data structure exists to show.

The authors have repeatedly cautioned against templating — the method is “a methodology, not a method” in their phrase — and reviewers who require a data structure regardless of fit are misreading the source.

• Fits: organisational processes with articulate insiders. Identity change, strategy formation, sensemaking during upheaval — settings where informants have rich language about their own situation.
• Fits: single or few cases studied deeply, with interviews as the spine and documents or observation as corroboration.
• Forced: phenomena informants cannot narrate — tacit practice, distributed system behaviour — where observation-first designs serve better.
• Forced: theory-testing questions. The method generates concepts; if constructs and hypotheses exist already, this is the wrong instrument and the data structure becomes decoration.

A researcher followed a 200-person software firm through the year it dropped its “engineering-led” identity, interviewing 34 insiders across three waves plus board minutes and all-hands recordings. First-order coding preserved the company’s own vocabulary — “adult supervision”, “the old guard”, “shipping culture” — yielding 118 categories.

Second-order theming consolidated these into themes including identity custodianship (who polices what “we” means), narrative re-anchoring (retelling the founding story to fit the new direction), and licensed nostalgia (sanctioned spaces where the old identity could be mourned). The third had no close analogue in the identity literature — the discrepant finding that justified the study.

The data structure displayed 118 → 11 → 3: aggregate dimensions of identity destabilisation, identity work, and identity re-stabilisation. The grounded model animated them as a cycle in which licensed nostalgia, counterintuitively, accelerated re-stabilisation — mourning rituals let veterans endorse the new identity without disowning their past. Every arrow in the model cited interview waves where the sequence was visible.

• Data structure as theory. Presenting the hierarchy as the finding; the model — relationships in motion — is the finding.
• Second-order themes that are just tidier first-order labels. The tier shift is from informant terms to theoretical constructs; renaming is not abstraction.
• Retro-fitted structures. Building the diagram after writing the findings, as decoration, is visible to experienced reviewers in the disconnect between boxes and text.
• Quote sprinkling. One quote per theme, chosen for vividness, defeats the evidentiary logic; themes need demonstrated breadth across informants.
• Template worship. Forcing every study into 1st/2nd/aggregate tiers regardless of question — the misuse the originators explicitly disown.

The method leans on informants’ ability to narrate their world, so it inherits their blind spots: what organisations cannot say about themselves — power, taboo, the taken-for-granted — enters only if the researcher works against the grain of the interviews.

The data structure’s rhetorical power is also a risk: a clean diagram can make thin analysis look rigorous, and the format’s reviewer appeal rewards exactly that. Rigor lives in the traceability, not the graphic.

And the approach is single-study, concept-generating by design; its theories are candidates for later testing and boundary-setting elsewhere, a modesty the write-up should keep.

Straussian grounded theory supplies a fuller procedural canon; the Gioia approach is looser in coding mechanics and stricter in display.

Constructivist grounded theory shares the interpretive epistemology and the gerund habit, without the data-structure convention.

Reflexive thematic analysis offers theme generation without theory-building ambitions — the right label when a grounded model is not the goal.

The method’s evidence rule — every theme traceable to informant terms — is a linkage problem tools are good at. Evidano codes interview corpora with each application tied to its quote, so first-order categories keep their verbatim grounding, tier-consolidation is done over retrievable evidence, and the final data structure can be audited link by link instead of asserted.

The second-order move — seeing what the categories are a case of, and what the literature lacks — is the researcher’s theoretical contribution. Software maintains the chain of custody; it does not have the disciplinary imagination the middle tier requires.