Reference

Frameworks: Grounded Theory and Theories of Action

Part of the Primary Research Synthesis skill · loaded on demand from SKILL.md

This file is the intellectual foundation for primary-research-synthesis. Read it when you need to apply the method with judgment rather than follow the workflow mechanically: when a coding decision is contested, when a client asks "why did you do it this way", or when the interview programme has gone off-script and you need to reason from first principles.

Contents

  1. Why consulting research needs a method at all
  2. Grounded Theory as published (Glaser and Strauss, 1967)
  3. The constant comparative method, stage by stage
  4. Theoretical sampling and theoretical saturation
  5. Strauss and Corbin's coding procedure (1990, 1998)
  6. The Glaser objection and what to take from it
  7. Theories of action (Argyris and Schön, 1974)
  8. Model I and Model II
  9. Reading the espoused / in-use gap in interview data
  10. Calibration evidence on sample size
  11. How the two frameworks fit together in an engagement
  12. Common misreadings of both frameworks
  13. Source list

1. Why consulting research needs a method at all

An interview programme produces two failure modes that look like success.

The first is the confirmation write-up. The team goes in with a hypothesis (from problem-framing or hypothesis-stress-testing), hears twenty people, and writes a synthesis that restates the hypothesis with the best six quotes attached. The quotes are real; the finding is not. Nobody counted how many interviewees disagreed, and the interviewer's questions steered the answers.

The second is the transcript digest. The team summarises each interview faithfully, lists "key themes" that are really the interview guide's section headings, and hands the client a document that tells them what they already knew was on the agenda. Nothing new emerged because the structure was fixed before the data arrived.

Grounded Theory was built against exactly these failures in academic sociology: theory verified by cherry-picked data on one side, description without explanation on the other. Argyris and Schön's theories of action were built against a third failure specific to organisations: taking what people say about their own behaviour as evidence of that behaviour. A consultant who applies both does not just summarise interviews; they produce an explanation of what is going on that the client did not already have, and can say how much evidence sits behind each part of it.

2. Grounded Theory as published (Glaser and Strauss, 1967)

Barney Glaser and Anselm Strauss published The Discovery of Grounded Theory: Strategies for Qualitative Research (Aldine, Chicago) in 1967. The book came out of their fieldwork on dying in hospitals (Awareness of Dying, 1965). Its central claim is that theory should be discovered from data, systematically obtained and analysed, rather than deduced from prior grand theory and then tested.

The method they describe has four commitments that matter for consulting work:

Simultaneous collection and analysis. You do not collect thirty interviews and then start analysis. You code the first two or three, form provisional categories, and let those categories shape who you talk to next and what you ask. In practice this means the interview guide is a living document and the coding starts on day one.

Comparison as the engine. Every new incident in the data is compared with previous incidents coded to the same category, and with other categories. The comparison forces you to articulate what a category actually means, what its properties are, and under what conditions it varies. This is the "constant comparative method" (section 3).

Categories with properties and dimensions. A category is not a label. "Discount approval" is a label. A category has properties (who approves, how fast, how often refused, what the requester expects) and each property has a dimensional range (seconds to days; never refused to usually refused). Until you can state properties and dimensions, you have a tag, not a category.

Theory that explains, not describes. The output is an account of how the categories relate: what conditions produce the phenomenon, what strategies people use in response, and what follows. A list of themes with frequencies is description; Grounded Theory asks for the relationships.

Glaser and Strauss also distinguished substantive theory (about one area: this company's pricing behaviour) from formal theory (about a general class: how control systems get gamed). Consulting work produces substantive theory. Do not overreach.

3. The constant comparative method, stage by stage

Chapter 5 of the 1967 book sets out four stages. They are not strictly sequential; the authors say the stages blend into each other, but each stage dominates a period of the work.

Stage 1: Comparing incidents applicable to each category. Code each incident (a sentence, an anecdote, an observed act) into as many categories as it fits. While coding, compare the incident with previous incidents in the same category. The first few comparisons generate the theoretical properties of the category. Write memos as you go: the moment you notice a conflict between incidents, stop coding and write out what the conflict is.

Stage 2: Integrating categories and their properties. As coding proceeds, comparison shifts from incident-with-incident to incident-with-the-properties-of-the-category. The categories begin to connect. This is where you notice that "manager approval takes minutes" and "managers approve nearly everything" are two properties of one thing (approval as a formality), not two separate findings.

Stage 3: Delimiting the theory. Two things happen. The theory solidifies: fewer modifications are needed as new incidents arrive, and those modifications are refinements rather than restructurings. And the category list is reduced: categories that do not connect to the emerging core are dropped or merged. The consultant's version of this is cutting the codebook from 60 codes to 25 and being able to say why.

Stage 4: Writing the theory. The memos, now organised around the delimited categories, become the write-up. The coded data provides the illustrations. In consulting, this is the insight statements and the theme narrative, with quotes and counts attached.

The mechanism that makes this work is that comparison forces explicitness. If you can only say "this quote is about pricing", you cannot compare it with another quote about pricing. If you must say "this is an instance of a rep pre-empting a discount request before the customer asked", you have a property you can look for elsewhere, count, and find exceptions to.

4. Theoretical sampling and theoretical saturation

Theoretical sampling is the process of choosing the next data source on the basis of what the emerging categories need, rather than on a sampling frame fixed in advance. If the first five interviews are all sales reps and they all describe the same escalation behaviour, the next interview should be a manager who receives those escalations, then a customer who observes them from outside, then a rep at a branch where margins held up. You are sampling to develop the category (find its dimensions and boundary conditions), not to represent a population.

Consulting engagements rarely allow fully open sampling; the interview list is often agreed with the client at the start. The workable compromise is to fix 60 to 70 percent of the list up front for coverage, and hold back the remainder to be allocated after the first coding pass.

Theoretical saturation is the stopping rule. Glaser and Strauss (1967, p. 61) define it as the point at which "no additional data are being found whereby the sociologist can develop properties of the category". Note what this says and does not say. It is saturation of a category, not of the dataset: the core categories saturate at different points and the peripheral ones may never be developed. And it is about properties, not about codes: a new interview may add no new code but still add a new condition under which an existing code applies, and that counts as new data.

Practical consequences:

  • Track saturation per theme, not per project.
  • Track two things: new codes per interview (cheap, mechanical) and new properties or conditions on existing top themes (requires judgment; record in the memo).
  • Segment the sample. Saturation among sales reps says nothing about saturation among customers.
  • Saturation is a judgment you must defend, so write down the criterion before you reach it.

5. Strauss and Corbin's coding procedure (1990, 1998)

Anselm Strauss and Juliet Corbin published Basics of Qualitative Research: Grounded Theory Procedures and Techniques (Sage) in 1990, with a second edition in 1998 and later editions by Corbin alone (2008, 2015). The book turned the 1967 method into a teachable procedure with three coding types. They are best read as three questions you ask of the data in sequence, and then in cycles.

Open coding: what is this an instance of? Break the data into discrete incidents, name each one with a concept, and group concepts into categories. Strauss and Corbin recommend line-by-line coding early on, precisely because it slows you down enough to see things the interview guide did not anticipate. Codes at this stage should be close to the data, often using the interviewee's own words (in-vivo codes: "a price and a pen"). The output is a long, messy code list with definitions. Length is fine here; the discipline is in defining each code so a second coder could apply it.

Axial coding: how do these categories relate? Reassemble the data by relating categories along an axis. Strauss and Corbin provide a "coding paradigm" to structure this. Its elements are:

  • the phenomenon: the central idea or event the category is about
  • causal conditions: what gives rise to it
  • context: the specific set of conditions within which it sits
  • intervening conditions: broader conditions that facilitate or constrain the strategies
  • action/interaction strategies: what people do in response to the phenomenon
  • consequences: what follows from those actions

For consulting, the paradigm is useful because it converts a theme into a causal story a client can act on. "Discount escalation" as a theme is inert. "Discount escalation (phenomenon) occurs because reps face large orders with thin authority (causal condition), in branches where the bonus rewards margin euros (context), with managers under the same bonus (intervening condition); reps respond by pre-clearing discounts by phone (strategy); the result is a matrix that routes rather than controls (consequence)" is something you can change.

Selective coding: what is the core category, and how does everything else relate to it? Choose the category that has the most explanatory power, that appears most often, and that the other categories connect to. Write the storyline: a few paragraphs that state the core category and relate the major categories to it. Then check every major category against the storyline; those that do not fit either indicate the core is wrong or belong in a separate finding. In consulting terms this is the governing insight, and it is the sentence that will become the title of the synthesis.

The three types are not a waterfall. You open-code the first interviews, sketch axial relationships, go back and open-code more with those relationships in mind, and only settle the core late. Corbin, in later editions, is explicit that the procedure should not be applied rigidly.

6. The Glaser objection and what to take from it

Glaser responded to the 1990 book with Basics of Grounded Theory Analysis: Emergence vs Forcing (Sociology Press, 1992). His objection was that the coding paradigm forces data into a preconceived structure (conditions, strategies, consequences) instead of letting the relationships emerge. Charmaz (2006, Constructing Grounded Theory) later treated axial coding as optional and highly structured.

The consultant does not need to pick a side, but should take the objection seriously in one specific way: the paradigm is a checklist for what a complete causal account contains, not a template to fill in. If the data does not show consequences, do not invent them. If the most important relationship is between two conditions rather than between a condition and a strategy, report that. The paradigm is there to make sure you asked all the questions, not to dictate the answers.

The second thing to take from Glaser is his insistence that the analyst should not read the literature (or, in our case, the client's own prior diagnosis) too deeply before coding. In consulting you cannot avoid the hypothesis; hypothesis-stress-testing may have handed you one. The discipline is to code the data first and map to the hypothesis second, and to record every incident that contradicts it.

7. Theories of action (Argyris and Schön, 1974)

Chris Argyris and Donald Schön published Theory in Practice: Increasing Professional Effectiveness (Jossey-Bass) in 1974, followed by Organizational Learning: A Theory of Action Perspective (Addison-Wesley, 1978). Their starting point is that people hold "theories of action": mental maps of the form "in situation S, to achieve consequence C, do action A". Two kinds exist in everyone:

Espoused theory is "the theory of action to which he gives allegiance, and which, upon request, he communicates to others". It is what a person will tell you they do, and often sincerely believe they do.

Theory-in-use is "the theory that actually governs his actions". It is inferred from behaviour, not stated. People are frequently unaware of the gap between the two, and when the gap is pointed out they tend to defend the espoused theory rather than examine the theory-in-use.

The critical claim for research synthesis: an interview, by itself, mostly elicits espoused theory. When you ask a branch manager how discounts are approved, you get the espoused process. When you ask for the last three approvals in detail, or you watch one happen, you begin to see theory-in-use. The two sources of data are not redundant; they measure different things, and the gap between them is often the finding.

Argyris and Schön also distinguished single-loop learning (detect error, correct action, leave the governing values untouched: the thermostat) from double-loop learning (detect error, question the governing values that produced the action). A client who responds to margin leakage by tightening the discount matrix is doing single-loop learning. A client who asks why managers approve everything, and discovers that the bonus and the culture reward not losing orders, is doing double-loop learning. Research synthesis that only reports the espoused theory can only support single-loop responses.

8. Model I and Model II

Argyris and Schön found that most people's theories-in-use, across cultures and organisations, follow one pattern they called Model I. Its governing variables (the values the person is trying to satisfy) are:

  1. Achieve the purpose as the actor defines it (define goals unilaterally and pursue them)
  2. Win, do not lose
  3. Suppress negative feelings (in oneself and others)
  4. Emphasise rationality (stay objective, minimise emotion)

The action strategies that follow are unilateral control, self-protection, and protection of others through withholding. The consequences are defensive relationships, low freedom of choice, and reduced production of valid information. Organisations built on Model I do not surface uncomfortable truths, and gaps between espoused and in-use theory persist because examining them feels like losing.

Model II is the alternative they proposed. Its governing variables are:

  1. Valid information
  2. Free and informed choice
  3. Internal commitment to the choice and constant monitoring of its implementation

For the researcher, Model I explains why the interview data looks the way it does. It predicts that interviewees will present their own behaviour as consistent with policy, attribute deviations to others or to circumstances, and frame the discussion in rational, procedural terms while the actual drivers are about not losing. It also predicts that the espoused/in-use gap will be largest on exactly the topics the client most needs to understand, because those are the ones where the organisation has the most to protect.

9. Reading the espoused / in-use gap in interview data

Three data types are needed to see the gap. Interview statements about policy and intention give the espoused theory. Detailed accounts of specific recent incidents, observations, and system records give the theory-in-use. The comparison gives the gap.

A workable procedure:

  1. In every interview, ask at least one "what is the policy / how should it work" question and at least one "walk me through the last time" question on the same topic. Code the answers separately (espoused codes and in-use codes).
  2. Wherever possible, add a non-interview source for the in-use side: an observation, a log, a sample of documents.
  3. For each major theme, write the espoused theory as a sentence, the theory-in-use as a sentence, and the evidence for each.
  4. Classify the gap. Useful classes: unaware (people genuinely believe the espoused theory), known and tolerated (people know the gap and consider it reasonable), known and concealed (people know and would not say so on the record), policy is wrong (theory-in-use is actually better than the espoused theory, which is common).
  5. Identify the governing variable that sustains the gap. In most cases it will be one of the Model I values: not losing, not surfacing conflict, staying "rational".

The classification matters because the intervention differs. An unaware gap closes with information. A tolerated gap closes only if the incentive or the governing value changes. A concealed gap will not close through a workshop. A "policy is wrong" gap should change the policy, not the behaviour.

10. Calibration evidence on sample size

Two studies are commonly cited to calibrate how many interviews a programme needs. Use them as priors, not rules.

Guest, Bunce and Johnson (2006, Field Methods 18(1)) analysed 60 interviews with a relatively homogeneous population on a narrow topic and found that 73 percent of all codes (80 of 109) had appeared within the first six interviews and 92 percent (100 of 109) within the first twelve. Their conclusion, that twelve interviews suffice for a homogeneous group and a focused question, is the source of the widely quoted "twelve".

Hennink, Kaiser and Marconi (2017, Qualitative Health Research 27(4)) separated code saturation (no new codes; the range of issues is identified) from meaning saturation (no new understanding of the issues; their dimensions are developed). In their data code saturation arrived at around nine interviews and meaning saturation needed sixteen to twenty-four. This maps directly onto the Glaser and Strauss definition: saturation is about properties, not labels.

For consulting programmes the practical consequences are: plan for roughly ten to fifteen interviews per distinct segment where the question is focused; expect the "we have heard it all" feeling at six to nine, and treat it as code saturation only; keep going until the top three themes stop acquiring new conditions and exceptions; and never claim saturation for a segment with fewer than six interviews.

11. How the two frameworks fit together in an engagement

Grounded Theory tells you how to turn raw talk into an explanation with known evidential weight. Theories of action tell you what kind of data to collect so the explanation is about behaviour rather than about what people say about behaviour. Run together:

  • Interview guide design draws on Argyris and Schön: pair each policy question with an incident question.
  • Open coding follows Strauss and Corbin, with codes tagged as espoused or in-use.
  • Axial coding uses the paradigm as a completeness check and builds the causal account.
  • Selective coding picks the core category; in organisational research the core is frequently the mechanism that sustains the largest espoused/in-use gap.
  • The gap table is a distinct deliverable because clients act on it differently from the theme hierarchy.
  • Saturation tracking gives the partner a defensible answer to "how do we know twenty was enough".

12. Common misreadings of both frameworks

These are the errors most often made by teams who know the vocabulary but not the mechanism.

MisreadingWhat it gets wrongCorrection
"Grounded Theory means we go in with no hypothesis"Glaser and Strauss asked for theoretical sensitivity, not ignorance; and consulting engagements arrive with hypothesesCode the data before mapping to the hypothesis; record every contradicting incident; report the hypothesis outcome explicitly
"Themes are the things people talked about most"Confuses frequency with explanatory power; the core category must explain variation, not just recurTest candidate cores against negative cases; a frequent theme that cannot explain the deviant site is a major category, not the core
"Saturation is when we stop hearing new things"That is code saturation; the 1967 definition is about properties of categoriesTrack new properties on top themes separately; expect meaning saturation to arrive later than the "heard it all" feeling
"Open, axial and selective coding are three phases done in order"Strauss and Corbin describe them as cycling; a strict sequence means the sample cannot adaptOpen-code early interviews, sketch axial relationships, return to open coding with those in mind, settle the core late
"The coding paradigm must be filled in for every theme"Glaser's objection: forcing structure onto data that does not contain itUse the paradigm as a completeness check; leave rows blank when the data is silent and say so
"Interviews tell us what people do"Interviews mostly yield espoused theoryPair policy questions with incident questions; add observation or records for the in-use side
"The espoused/in-use gap means people are lying"Argyris and Schön found the gap is usually unconsciousClassify the gap (unaware, tolerated, concealed, policy wrong) before deciding on the response
"Model I is a personality type"It is a pattern of governing values that most people in most organisations shareLook for the governing value in the incident, not in the person; the same person behaves differently where the social cost of losing changes
"Senior interviewees carry more weight"Position to observe, not seniority, determines credibility for a given claimWeight the front line on theory-in-use, leaders on espoused theory, and triangulate across
"Twelve interviews is the rule"Guest et al. studied a homogeneous group on a narrow questionPlan per segment; use twelve as a prior for focused questions in homogeneous groups, and expect more where segments differ

13. Source list

  • Glaser, B. G. and Strauss, A. L. (1967). The Discovery of Grounded Theory: Strategies for Qualitative Research. Chicago: Aldine.
  • Glaser, B. G. and Strauss, A. L. (1965). Awareness of Dying. Chicago: Aldine.
  • Strauss, A. and Corbin, J. (1990; 2nd ed. 1998). Basics of Qualitative Research: Grounded Theory Procedures and Techniques. Thousand Oaks: Sage.
  • Corbin, J. and Strauss, A. (2008; 4th ed. 2015). Basics of Qualitative Research. Thousand Oaks: Sage. (Later editions, with the procedure presented as more flexible.)
  • Glaser, B. G. (1992). Basics of Grounded Theory Analysis: Emergence vs Forcing. Mill Valley: Sociology Press.
  • Charmaz, K. (2006). Constructing Grounded Theory. London: Sage.
  • Argyris, C. and Schön, D. A. (1974). Theory in Practice: Increasing Professional Effectiveness. San Francisco: Jossey-Bass.
  • Argyris, C. and Schön, D. A. (1978). Organizational Learning: A Theory of Action Perspective. Reading, MA: Addison-Wesley.
  • Guest, G., Bunce, A. and Johnson, L. (2006). "How Many Interviews Are Enough? An Experiment with Data Saturation and Variability." Field Methods 18(1): 59-82.
  • Hennink, M. M., Kaiser, B. N. and Marconi, V. C. (2017). "Code Saturation Versus Meaning Saturation: How Many Interviews Are Enough?" Qualitative Health Research 27(4): 591-608.