Introduction and Overview

You have spent six modules building the analytic apparatus for a credible qualitative study. You have a research question (Lesson 2), a sampling plan (Lesson 3), data you trust (Lesson 4), themes and a codebook (Lesson 5), and an analysis framework with memos (Lesson 6). You are now sitting in front of twenty transcripts, a codebook with thirty-odd codes attached to a few hundred coded extracts, and the next question every working qualitative researcher faces at this point: how do these cases differ, and what do the differences mean?

Lesson 7 is the answer to that question, in two parts that Bernard, Wutich, and Ryan treat as a single methodological arc. Bernard, Wutich, and Ryan (2017) devote Chapter 9 to comparing attributes of variables — the matrix-display tradition associated with Miles, Huberman, and Saldaña — and Chapter 10 to grounded theory — the methodological tradition that built itself around the discipline of constant comparison. The two chapters belong together because they are doing the same intellectual work with different vocabularies. The matrix tradition systematizes comparison by laying cases and codes on a grid and reading the grid. The grounded theory tradition systematizes comparison by iterating between data and concepts until a substantive theory emerges. Both are engines for moving from a stack of coded transcripts to a defensible explanatory account.

This section is about the first tradition. Matrices are deceptively simple. A grid with codes on one axis and cases on the other looks like the kind of summary a student might draw on a napkin. The intellectual move is what is on the grid — not raw quotes but second-order analytic compressions — and what the analyst does with the grid: read it across rows for patterns of variation in a single phenomenon across cases, read it down columns for the internal coherence of a single case, and read its empty cells for what is informatively absent. Miles, Huberman, and Saldaña (2014) call matrix building “the workhorse of qualitative analysis” for a reason: every other comparison technique in Bernard, Wutich, and Ryan either feeds a matrix or is fed by one.

1.1 Where Matrix Analysis Comes From

The matrix-display tradition is associated above all with Matthew Miles and A. Michael Huberman, whose 1984 textbook Qualitative Data Analysis: A Sourcebook of New Methods was the first widely adopted operational manual for handling large qualitative datasets in education and policy research. Miles and Huberman were both program evaluators, and the matrix style bears the marks of that origin: they were dealing with multi-site studies in which the analyst had thirty schools or twenty clinics or fifteen reform programs and needed a defensible way to compare them. Narrative case-by-case reporting did not scale. Matrix displays did.

The third edition of the textbook, published in 2014 after Huberman's death, was revised by Johnny Saldaña and is now the canonical reference: Miles, M. B., Huberman, A. M., & Saldaña, J. (2014). Qualitative Data Analysis: A Methods Sourcebook (3rd ed.). SAGE. The 2014 edition reorganizes the original taxonomy around two display families — matrices and networks — and is the version Bernard, Wutich, and Ryan (2017) cite when they discuss Chapter 9 techniques. If your institution gives you access to one qualitative methods sourcebook beyond this course, it should be this one.

The deep claim Miles, Huberman, and Saldaña make is that display drives analysis. The form in which you arrange your data on the page determines what patterns you can see. Running text hides comparisons; a 2D grid makes them visible. The corollary is that producing a good display is itself analytic work — not a presentation step after the analysis is done but a step in which the analysis happens. The cells of the matrix are the unit of analytic compression: you are deciding what counts as the content of cell (code X, case Y), and that decision is an interpretation.

1.2 The Four Matrix Types You Will Actually Use

Miles, Huberman, and Saldaña (2014) catalogue dozens of matrix variants. Most working qualitative researchers in health rely on four. Understanding what each one is for is the difference between using matrices as a real analytic tool and using them as a way to make a methods section look more rigorous.

1.2.1 Conceptually Ordered Matrices

A conceptually ordered matrix arranges its rows and columns by theoretical construct. Rows might be the dimensions of a phenomenon (for loneliness: triggers, embodied features, meanings assigned, coping moves, structural critique). Columns might be theoretical contrasts (for example, situational vs. existential loneliness as candidate types). The cells contain compressed evidence about how each construct shows up in each theoretical slot. The matrix is the analyst's working theory written down in grid form.

Conceptually ordered matrices are most useful when you already have an analytic framework (often from Lesson 6) and you want to populate it with evidence. They are less useful at the very start of analysis, when the frameworks have not yet stabilized.

1.2.2 Time-Ordered Matrices

A time-ordered matrix organizes its columns chronologically. The rows are cases or codes; the columns are time periods, life-stages, phases of an intervention, or sequence positions in a process. For the loneliness dataset, a time-ordered matrix might use columns like “before the loss/transition,” “in the immediate aftermath,” “one year out,” “current.” Each cell contains a compressed account of what loneliness was like in that period for that case.

Time-ordered matrices are the right choice when your phenomenon has a developmental, sequential, or processual structure. They are how you recover a trajectory from a corpus of cross-sectional interviews that nonetheless contain retrospective accounts of process.

1.2.3 Role-Ordered Matrices

A role-ordered matrix groups cases by social position or role. Rows might be codes; columns might be participant roles (patients, providers, family caregivers, system administrators). The point is to expose how the same phenomenon appears differently depending on who is reporting it. Role-ordered matrices are workhorses of implementation science: they make visible the gap between the program-as-designed and the program-as-experienced by quickly stacking the same questions across the people occupying different positions in the same system.

In a single-role dataset like our loneliness study (every participant is in the role of “person experiencing loneliness”), role-ordering is less directly available, but it can still be used by treating life-stage or sampling subgroup as a quasi-role: students vs. retirees, immigrants vs. native-born, partnered vs. unpartnered.

1.2.4 Two-Dimensional Code-by-Case Matrices

The workhorse of workhorses. Rows are codes from your codebook; columns are cases (participants, transcripts, sites). Each cell contains a compressed analytic summary of what that case said about that code — not a raw quote but a digested version of the content. Variants of this design dominate practical qualitative health research because they sit directly on top of the artefact every coded transcript produces: a tagged corpus that can be queried code-by-case.

Your Week 7 capstone matrix — if you choose Option A — will be a two-dimensional code-by-case matrix. The rest of this section walks through the construction and reading of one using the loneliness corpus.

1.3 A Worked Code-by-Case Matrix from the Loneliness Corpus

To make the matrix concept operational, let us build a small example matrix using six codes and eight participants. The codes have been chosen because they recur across the corpus and because they map onto theoretically interesting dimensions of loneliness identified in the existing literature (Cacioppo & Patrick, 2008; Weiss, 1973; Hawkley & Cacioppo, 2010). The eight participants have been chosen to maximize variation across age, life-stage, immigration status, and circumstance.

The codes are:

• Trigger: a specific event, time of day, or context that the participant identifies as reliably bringing on loneliness.
• Embodied feature: a physical sensation participants attach to loneliness (hunger, ache, weight, fatigue, tightness).
• Coping move: an action the participant takes when loneliness arrives.
• What helped that surprised them: something the participant did not expect to help that did — an analytically rich category because it surfaces unmarked or culturally invisible support.
• Structural critique: a statement implicating institutions, policies, or social conditions in producing the loneliness.
• Identity stake: the relationship between the participant's loneliness and their sense of who they are (state vs. trait, failing vs. ordinary, shame vs. acceptance).

The matrix below contains compressed cell content. In a real working matrix, each cell would carry a transcript line reference (e.g., P01: ll. 21–25) so the analyst can return to the underlying evidence. The compression style here is descriptive, not interpretive — the interpretation happens when the analyst reads the matrix.

1.4 Reading the Matrix

Producing the matrix is half the work. Reading it is the other half. There are three reading moves a disciplined analyst makes systematically, and a fourth that distinguishes excellent matrix work from competent matrix work.

1.4.1 Read Across Rows: Cross-Case Variation in One Code

A row tells you how a single code varies across the corpus. Reading the “Trigger” row shows a recurring pattern: triggers cluster around specific times (evenings, weekends, Sunday at 9pm, late at night) and around specific contrasts (a phone call ending, a celebration without the absent person, a holiday performed without its original setting). The row also shows variation: Maya's trigger is environmental (the SkyTrain), Linda's is an object (Bill's chair), Aarav's is bicultural (after a phone call that highlights the displacement), Frank's is interpersonal (visitors leaving). Reading down a row generates analytic candidates — here, the possibility that loneliness triggers can be sorted into three sub-types: environmental cues, relational contrasts, and temporal markers. That sub-typing is a finding the row produces.

1.4.2 Read Down Columns: Within-Case Coherence

A column tells you how the codes hang together for one person. Reading down P15 Amira's column shows that her triggers (anniversary of destruction, bathroom moments), her embodied features (sleep loss, hidden crying), her coping moves (cooking mother's recipes, Syrian women's group), her structural critique (services treat trauma and loneliness separately), and her identity stake (wahda teaching acceptance) form a coherent picture. The loneliness she describes is refugee loneliness as a specific configuration of trauma, displacement, language, and unrepeatable memory. Reading down Chen's column shows a different configuration: achievement-bound loneliness, in which the trap is partly self-built and partly structural.

Column reading is how you preserve the integrity of the case while still doing comparison. The matrix is not just a cross-case engine; it is also a within-case engine, because the columns force you to keep each person whole.

1.4.3 Read the Empty Cells: Informative Absence

An empty cell is information. Helen's “what helped (surprising)” cell is blank not because Helen had no surprises but because the interviewer did not ask Helen that question. That is a data-collection note that goes into the limitations section of your paper. Differently, the relative thinness of the “structural critique” row for some younger participants reflects a real pattern: younger participants are less likely to articulate their loneliness as a product of policy or institutions. That is a substantive finding.

The rule of thumb Bernard, Wutich, and Ryan offer is: when a cell is empty, ask first whether the data are missing (an under-elicitation issue) or whether the case really has nothing in this slot (a substantive issue). The two require different responses. Missing data are addressed in the limitations section; substantive absences are addressed in the findings.

1.4.4 Read for Anomalies

The fourth reading move — the one that separates good matrix work from competent matrix work — is reading for anomalies. An anomaly is a cell whose content does not fit the pattern its row and column would predict. Frank's “identity stake” cell — calm acceptance of companionship with absent persons — is anomalous against the broader pattern of identity-stakes among participants of his age, who more often report acceptance with sadness or resignation rather than calm. Anomalies generate the most productive analytic memos because they force the analyst to either refine the dominant pattern or identify an alternative type. The grounded theorists call this the search for “negative cases,” and we will meet it again in Section 3 as a formal step in the grounded theory procedure.

1.5 Magnitude Coding — When and Why to Count

Bernard, Wutich, and Ryan (2017, Chapter 9) introduce a term that may make readers from interpretivist traditions uncomfortable: magnitude coding. Magnitude coding is the practice of attaching a numerical or ordinal value to a coded extract — for example, marking each mention of a code as “low,” “medium,” or “high” intensity, or simply counting how many transcripts contain a code, or how many times each transcript mentions it. Saldaña (2016) treats magnitude coding as a recognized coding family. Miles, Huberman, and Saldaña (2014) treat it as routine matrix-cell content. Bernard, Wutich, and Ryan (2017) treat it as defensible whenever the question being answered is about commonness, dominance, or change.

The argument for counting in qualitative work is straightforward. If you have decided that a pattern is real because it shows up “often,” you owe the reader a defense of “often.” The least defensible version of “often” is “in my impression.” The most defensible version is “in N of M transcripts.” Counting transforms an impressionistic claim into a checkable one. It does not turn the qualitative study into a quantitative study, because the units being counted are qualitative codes, not pre-specified variables.

The argument against counting — which Bernard, Wutich, and Ryan acknowledge — is that frequencies in qualitative work are artefactual. They reflect what was asked, who was sampled, and how long each interview ran, as much as they reflect the underlying phenomenon. A code that appears in eighteen of twenty transcripts may show up that often because the interview guide asked about it, not because the phenomenon is widespread. So a defensible practice involves reporting counts and being explicit about their interpretive limits.

Three rules of thumb help:

1. Count when commonness matters to the claim; do not count when the claim is about meaning or mechanism. Saying “14 of 20 participants used spatial metaphors to describe loneliness” is a legitimate commonness claim; saying “14 of 20 participants found loneliness shameful” is more fraught, because shame is a meaning-claim that should not be reduced to a count.
2. Report the denominator transparently. “14 of 20” is checkable; “most participants” is not.
3. Acknowledge the artefactual character of the count when the code was directly elicited by the interview guide. A code that emerged from open coding and shows up in many transcripts is a stronger commonness signal than a code that the interviewer asked about in every interview.

1.6 Within-Case and Across-Case as Analytic Perspectives

The matrix tradition makes one more conceptual move that is worth pulling out explicitly: it treats within-case analysis and across-case analysis as two perspectives on the same dataset, both of which are necessary. Within-case analysis preserves the integrity of each participant's account, on the grounds that an interview is a coherent whole and that fragmenting it into codes loses something. Across-case analysis sacrifices wholeness for comparability, on the grounds that the patterns we care about appear only when we juxtapose cases.

Both are right. Matrix work alternates between them. The column read is within-case (Aarav's loneliness is its own configuration). The row read is across-case (triggers cluster into three sub-types). Most rigorous qualitative analyses move back and forth across the term of the analysis, sometimes producing “case-ordered displays” (Miles, Huberman, & Saldaña, 2014) in which cases are deliberately ranked along an analytic dimension to make a pattern visible.

One thing Bernard, Wutich, and Ryan are explicit about: do not allow within-case work to collapse into mere description, and do not allow across-case work to lose the participant. Both failure modes are common. Description-only within-case analysis reads like a series of biographies that never reach a finding. Pattern-only across-case analysis reads like a survey report in qualitative clothing. Good matrix work refuses both.

Introduction and Overview

If matrix analysis is the comparison engine of the displays tradition, grounded theory is the comparison engine of an entire research paradigm. Grounded theory is the most-cited qualitative methodology in health research and the most-misused. Most papers that claim to have used “a grounded theory approach” have not; they have used some version of thematic coding and named it grounded theory because the name carries methodological prestige. Real grounded theory is a more specific and more demanding thing. This section establishes what grounded theory actually is, where it came from, and why three substantially different variants now circulate under the same name. Section 3 will then take you through the operational pipeline.

The reason Bernard, Wutich, and Ryan (2017) place grounded theory next to matrix analysis is that both are organized around the same intellectual move: constant comparison. In matrix work, comparison happens visually, across rows and columns of a finished display. In grounded theory, comparison happens iteratively, between each new piece of data and the analytic categories the researcher has built so far. The grounded theorist is constantly asking: is this new instance the same as the others I've coded under this category, or is it different? If it is different, does the category need to be subdivided? Does the data need a new category? Does an existing category need to be renamed or refined? The discipline of asking those questions continuously is what produces a substantive theory grounded in data, rather than an unmotivated list of themes.

2.1 The Discovery of Grounded Theory (1967)

Barney Glaser and Anselm Strauss published The Discovery of Grounded Theory: Strategies for Qualitative Research in 1967, and the book is best understood as a polemic against the sociology of its moment. Mid-century American sociology was dominated by Talcott Parsons's grand theory, which proposed elaborate conceptual schemes that empirical work was expected to verify, and by Paul Lazarsfeld's quantitative-survey tradition, which tested narrow hypotheses with statistical machinery. Qualitative work — what survived of it — was treated as preliminary to real research: useful for hypothesis-generation, perhaps, but not for theory-building proper.

Glaser and Strauss rejected this hierarchy. Their argument, made on the back of Strauss's ethnographic work on dying in hospitals (Glaser & Strauss, 1965), was that qualitative research could and should generate theory directly from systematic engagement with data. The theory would be “grounded” in the sense that every concept in it could be traced to a piece of empirical evidence that justified including it. Theory would not come down from Parsonian heights; it would come up from the field.

The 1967 book laid out the constituent moves: theoretical sampling (sample to develop categories, not to represent populations), constant comparison (continuously compare each new datum to existing categories), coding in successive levels of abstraction, memo-writing as the engine of conceptual development, and theoretical saturation as the stopping rule. The book is sometimes described as combative in tone; that is fair. Glaser and Strauss were arguing for the legitimacy of a kind of research that mainstream sociology was actively dismissing, and the argument is sharpened accordingly.

“Generating a theory from data means that most hypotheses and concepts not only come from the data, but are systematically worked out in relation to the data during the course of the research.” — Glaser & Strauss, 1967, The Discovery of Grounded Theory, p. 6

The 1967 stance contained an unresolved ambiguity that would later split the methodology. The book emphasized that the analyst should approach the data without preconceived theoretical commitments, allowing categories to emerge. But the book also emphasized systematic procedures — coding levels, axial relationships, paradigm models — that imposed a particular analytic structure on what was emerging. Was grounded theory radically inductive (categories arise wholly from data) or systematically procedural (the analyst applies a specific coding apparatus)? The 1967 book held both stances in tension. The next generation of grounded theorists pulled the two apart.

2.2 The Glaser–Strauss Split

By the late 1980s the two founding authors had developed substantially different visions of what grounded theory should be. The split became public when Strauss, with Juliet Corbin, published Basics of Qualitative Research: Grounded Theory Procedures and Techniques in 1990 (see also Corbin & Strauss, 1990). The Strauss & Corbin book emphasized procedural rigor: it laid out specific coding stages (open, axial, selective), a formal “paradigm model” for axial coding (causal conditions, phenomenon, context, intervening conditions, action strategies, consequences), and a structured route from open codes to a core category.

Glaser objected. In Basics of Grounded Theory Analysis (1992), he argued that Strauss and Corbin had betrayed the original method by imposing a coding apparatus on the data before allowing categories to emerge. The paradigm model, in Glaser's view, was not grounded theory; it was forcing data into a pre-given structure. Grounded theory, Glaser insisted, requires letting categories arise without procedural scaffolding. His phrase — that the categories should “emerge” — became the rallying flag of what is now called the Glaserian or classic grounded theory tradition.

The dispute is methodologically interesting for two reasons. First, it exposed an ambiguity in the original 1967 formulation that no amount of careful reading could resolve. Second, it left the field with two camps both calling their work “grounded theory” while doing substantively different things. Health researchers who claim to use grounded theory often inherit this confusion: are they following Glaser's emergence approach, or Strauss and Corbin's procedural approach, or some unacknowledged hybrid?

2.3 Charmaz's Constructivist Grounded Theory

Kathy Charmaz, a student of Strauss and a medical sociologist whose own substantive work was on chronic illness identity, proposed a third path. Her Constructing Grounded Theory (Charmaz, 2006, second edition 2014) reframed the methodology in explicitly constructivist terms: the theory is not discovered in the data; it is co-constructed by the analyst and the participants through interpretive engagement. This stance is now the dominant variant in contemporary qualitative health research, particularly in nursing, public health, medical sociology, and health-psychology.

What Charmaz changes:

1. Epistemological stance. Charmaz rejects the assumption (Glaserian) that there is a single theory in the data waiting to be found. Multiple defensible readings are possible. The analyst's positionality, theoretical commitments, and interpretive choices shape what is seen. This is the change Charmaz's stance shares with the broader interpretivist and critical turn in social science.
2. Treatment of the literature. Where Glaser insisted on a literature review only after analysis (to prevent contamination), Charmaz treats the literature as an interlocutor that the analyst is always in conversation with — before, during, and after analysis. The analyst's job is not to keep prior theory out but to make its influence visible.
3. Treatment of participants. Charmaz's stance is more attentive to power, voice, and the social context of meaning-making. She is explicit about whose categories appear in the final theory and why, and she resists the assumption that the analyst's categories transcend the participants' own.
4. Coding procedures. Charmaz preserves the open-axial-selective sequence but loosens the paradigm-model requirement of Strauss and Corbin. She emphasizes “focused coding” (selecting the most analytically productive codes) over rigid axial procedures.
5. Memo-writing. Memos take on heightened importance in Charmaz: they are where the analyst writes their interpretive engagement into existence. The line between memo and finding is intentionally blurred.

“We are part of the world we study and the data we collect. We construct our grounded theories through our past and present involvements and interactions with people, perspectives, and research practices.” — Charmaz, 2014, Constructing Grounded Theory (2nd ed.), p. 17

Charmaz's reframing made grounded theory legible to a generation of qualitative health researchers who were sympathetic to its procedural discipline but uncomfortable with the realist epistemology of the Glaserian formulation. Her variant is now the most-cited grounded theory text in the health sciences and the variant most likely to appear in published nursing and public-health papers. Bernard, Wutich, and Ryan (2017) treat all three variants as legitimate but note that the procedural scaffolding (open / axial / selective coding) is shared across all three, even when the philosophical framing differs.

2.4 Other Variants Worth Knowing

Two further variants deserve brief mention because they appear in the health literature and you will encounter them in papers.

Adele Clarke's situational analysis (Clarke, 2003; 2005, 2018) is a post-Charmazian extension that emphasizes the situation rather than the individual as the unit of analysis. Clarke proposes “situational maps” that include human and non-human elements, discursive constructions, and contested positions, and treats grounded theory as part of a broader cartographic strategy for analyzing complex social situations. Situational analysis is most used in studies of contested policy domains and in social-movement research.

Antony Bryant's (2017) pragmatist grounded theory reanchors the methodology in Deweyan pragmatism, treating theory as a tool for action rather than a representation of reality. Bryant's work is influential among researchers in information science and education but less so in health.

You do not need to master these variants for HSCI 841. The reason to know they exist is that you will see them named in published methods sections, and you should not mistake them for the same thing as Charmaz's variant or the original 1967 method. For an accessible single-volume contemporary textbook that synthesizes across these traditions, see Birks and Mills (2015); for a wide-ranging edited reference, see Bryant and Charmaz's Handbook of Grounded Theory (2007).

2.5 Grounded Theory and Theme Identification: Same or Different?

Lesson 5 introduced theme identification — the systematic process of recognizing recurring patterns of meaning in qualitative data. A reasonable student question at this point is: how is grounded theory different from theme identification with a fancier name?

The honest answer is that there is significant overlap. Both are interpretive processes; both involve coding; both require comparison; both produce categories of meaning. But there are real differences, and they matter for what you can claim:

• Theme identification ends with a descriptive account of what the major themes in the dataset are, often supported by exemplary quotes and a frequency count. It is a finding about what is in the data.
• Grounded theory ends with a substantive theory: an explanatory account of how the elements of a phenomenon relate to one another, organized around a core category that captures the main analytic insight. It is a finding about why things in the data are the way they are.

The deeper difference is that grounded theory is committed to theoretical sampling and theoretical saturation as procedural commitments. Theme identification on a fixed dataset (the way your capstone is structured) cannot literally do theoretical sampling, because the data are already collected. Most contemporary health-research grounded theory studies acknowledge this and report a “grounded-theory-informed” or “modified grounded theory” analysis rather than claiming to have done full grounded theory. Bernard, Wutich, and Ryan (2017) accept this practical compromise but ask that researchers be honest about it.

2.6 Why This Matters for Public Health

Grounded theory has been used in public-health research for studies of HIV stigma, chronic disease self-management, end-of-life decision-making, vaccine hesitancy, drug use, mental health help-seeking, and many other areas where the central analytic problem is not measuring how often something happens but understanding the process by which it unfolds. Three exemplary substantive theories generated by grounded theory work:

• Charmaz's own (1991) theory of good days, bad days in chronic illness — the way people with serious chronic conditions organize their identity around the rhythm of disease management.
• Glaser's (1968) theory of awareness contexts in dying — the way patients, families, and clinicians negotiate what is known and unknown about a terminal prognosis.
• Strauss's (1985) theory of illness trajectories — the way chronic illness work is distributed across patient, family, and clinical actors over time.

These are not theme lists. They are explanatory accounts with a core analytic insight, supporting categories, and a structure that lets readers see why the phenomenon takes the shape it does. They are the kind of contribution that grounded theory at its best produces — the kind that a careful matrix analysis on its own would not, because matrix analysis is built for comparison and grounded theory is built for explanation.

Introduction and Overview

Section 2 established what grounded theory is and where its variants come from. This section operationalizes the procedure. By the end you should be able to walk through a grounded-theory analysis from the first line-by-line coding pass to the articulation of a substantive theory anchored in a core category, with a clear understanding of where constant comparison and theoretical sampling enter the workflow. The exposition follows the Strauss & Corbin (1990, 1998) procedural backbone — because that is the version most operationally specifiable — and notes where Charmaz's (2014) constructivist re-reading shifts each step.

The pipeline you are about to walk through has four named stages: open coding, axial coding, selective coding, and the writing-up of the resulting substantive theory. Two cross-cutting practices run throughout: constant comparison, which iterates between data and categories at every step, and theoretical sampling, which decides which case or excerpt to look at next on the basis of what the emerging theory needs. The whole procedure is held together by memo-writing, which we covered in Lesson 6 and which becomes the substantive text of the eventual theoretical write-up.

3.1 Open Coding

Open coding is the first analytic engagement with the data. Its purpose is to break the data open — to fragment the seamless flow of an interview into small analytic chunks that can be compared with each other and with chunks from other interviews. The fragmentation is deliberate. A transcript read straight through induces a kind of narrative trance: the analyst gets caught up in the participant's story and stops noticing comparisons. Open coding interrupts that trance by forcing the analyst to attend to each piece in turn.

Two main techniques are used:

1. Line-by-line coding. Originally proposed by Glaser and developed by Charmaz (2014), this is the most fragmenting form of open coding. The analyst attaches a code to every line, asking continuously: what is happening in this line? what is this an instance of? what would I call this if I had to name it for an audience that hadn't read the interview? The codes are typically gerunds (action-words ending in -ing: “naming the trigger,” “evading the shame,” “rationing the call home”) because gerunds make process visible. Line-by-line coding is laborious and tends to over-generate codes; it is most useful at the very start of a grounded-theory project.
2. Segment-by-segment coding. The analyst attaches codes to longer chunks — a paragraph, a turn at talk, a sequence of related lines. Segment coding is less fragmenting and generates fewer codes; it is what most working researchers actually do once they have a feel for the data.

Both produce a working list of open codes. The list at this stage is provisional and over-generated. A fifteen-line excerpt may produce eight to twelve codes; a forty-five-minute interview may produce sixty to a hundred. Many will turn out to be near-duplicates of each other and will be merged later. Many will turn out to be too narrow and will be subsumed under broader categories. The point at this stage is to be generous, not selective. The selection happens in axial coding.

3.1.1 An Open-Coding Worked Example

Take a passage from P01 Maya (lines 19–21 of her transcript):

“Honestly? Like, the SkyTrain at 9 p.m. on a Sunday. That's what comes to mind first. Which is weird because there's people on it. There's tons of people on it. But everyone's just on their phones, and nobody — nobody acknowledges that anybody else exists, and I just sit there and I feel like I could disappear and nobody would, like, notice. So that's loneliness for me, I think. It's being around people who don't see you.” — P01 Maya, lines 19–21

A line-by-line open coding pass might produce:

• naming a specific trigger (“SkyTrain at 9 p.m. on a Sunday”)
• locating loneliness temporally (Sunday evening as a marked time)
• locating loneliness in public-transit space
• marking the paradox (lonely because surrounded, not despite)
• attributing the cause to mediated attention (phones)
• articulating non-recognition (“nobody acknowledges”)
• imagining one's own invisibility (“I could disappear”)
• fearing non-noticing (“nobody would notice”)
• defining loneliness (“being around people who don't see you”)

Notice three features of this code list. First, every code is a gerund or noun phrase tied to an action or attribution. Second, the codes are pitched at the level of what the participant is doing in the talk, not what their inner life is. Third, several codes look near-duplicate (non-recognition, invisibility, non-noticing). The duplication is intentional at this stage; the codes will be compared and consolidated later.

3.1.2 Comparing Across Cases

Constant comparison enters immediately. Take a similar definitional passage from P06 Aarav (lines 19–21):

“Loneliness. Hmm. I think for me, loneliness is — okay, in Telugu we have a word, ekantam, but it means more like solitude, peacefully alone. It's a good word. And then we have another word, ekakitatvam, which is more like the feeling of being alone in a crowd. The second one is what I feel here. I am never physically alone — I have three housemates, I have a busy lab, I have classmates, I am on the bus with hundreds of people every day — but I am ekakitatvam. I am alone in the crowd.” — P06 Aarav, lines 19–21

Open codes from Aarav's passage might include:

• distinguishing solitude from loneliness (ekantam vs. ekakitatvam)
• using a non-English term for the felt distinction
• identifying with the “alone in a crowd” version
• cataloguing surrounding people (housemates, lab, classmates, bus)
• defining loneliness as crowd-paradox

Now constant comparison does its work. Maya and Aarav both define loneliness through the paradox of being surrounded yet alone. The code marking the paradox from Maya and the code defining loneliness as crowd-paradox from Aarav are versions of the same thing. We can merge them into a broader provisional category: loneliness-inside-crowdedness. That category now has two cases supporting it. We will keep looking for more.

If we then read P05 Linda's definitional passage (lines 19–21), we find a different kind of loneliness:

“Oh. Bill's chair. That's what comes to mind. We had — there's a chair in the living room there, you can see it, the one with the green cushion. Bill sat in that chair every evening for thirty-some years. And it's still there. I haven't moved it. I haven't sat in it. I haven't given it away. It's just there. And every evening I look at it and it's empty. And that's loneliness, for me. It's a particular empty chair.” — P05 Linda, lines 19–21

This is not loneliness-inside-crowdedness. This is loneliness defined by the absence of a specific other. Constant comparison forces us to recognize that the data are pointing toward at least two distinct sub-types of loneliness: a presence-paradox loneliness (where the loneliness arises from undifferentiated co-presence with people who do not see you) and an absence-anchored loneliness (where the loneliness arises from the specific, recoverable memory of a particular person no longer present). This sub-typing is the kind of conceptual product open coding plus constant comparison generates.

3.2 Axial Coding

Axial coding is the second analytic stage. Where open coding fragmented the data, axial coding reassembles it — not in the original narrative order, but along analytic axes that link codes to one another. The name comes from the image of an axis: a category that other codes rotate around. The work of axial coding is identifying which categories are central and how the other codes relate to them.

Strauss and Corbin (1990, 1998) proposed a specific apparatus for axial coding, which they called the paradigm model. The paradigm model asks the analyst to organize codes related to each category along six dimensions:

1. Causal conditions: what gives rise to the phenomenon?
2. Phenomenon: the central category itself — what is happening?
3. Context: the specific properties of the setting in which the phenomenon occurs
4. Intervening conditions: broader background conditions that shape how the phenomenon plays out
5. Action/interactional strategies: what people do in response to the phenomenon
6. Consequences: what results from those actions

The paradigm model is the move that Glaser objected to as “forcing” the data. Charmaz (2014) retains the spirit of axial coding — identifying central categories and the codes that orbit them — but does not require the strict six-dimensional template. Her version is called “focused coding”: the analyst selects the most analytically productive open codes and uses them to re-read the data, building up categories and noting how they relate to one another.

3.2.1 Axial Coding on the Loneliness Data

Suppose open coding has produced (among many other codes) a cluster around what we will provisionally call “loneliness inside companionship.” This category captures the experience — visible across P01 Maya, P06 Aarav, and to some extent P11 Helen and P16 Elena — in which the participant is in regular contact with others but nonetheless feels lonely during those contacts. It is not the same as the loneliness of being alone. It is a specific failure of co-presence to translate into felt connection.

Axial coding around this category, using the Strauss & Corbin paradigm model:

The axial display has done analytic work the open codes alone could not. It has located the central phenomenon, identified its causal structure, separated context from intervening conditions, and given us a way to sort coping strategies into sub-types based on whether they match the underlying deficit. The same display also surfaces the variation within the category: refugee loneliness, immigrant loneliness, late-life widow loneliness, and late-life-coming-out loneliness (Elena) are versions of the same phenomenon with different causal-condition profiles.

3.3 Selective Coding and the Core Category

Selective coding is the third stage. By this point the analyst has many axial categories. Selective coding integrates them around a single core category that captures the central analytic insight of the study. The core category is the one that, when you tell it, the whole pattern of categories falls into place.

How the core category is chosen:

1. It must be central — it must appear in most of the cases.
2. It must have explanatory reach — the other categories must relate to it in specifiable ways.
3. It must be abstract enough to apply beyond the immediate data but specific enough to retain its content.
4. It must hold up under negative-case examination — cases that seem not to fit either are explained, or the core category is refined to accommodate them.

For our running loneliness analysis, a candidate core category might be “the work of converting presence into recognition.” This is more abstract than “loneliness inside companionship” (which is just the phenomenon) and proposes an explanatory process: loneliness, in these data, is what happens when participants cannot convert physical co-presence into felt recognition. Their coping moves can then be sorted by whether they go after presence (more contact, more groups) or after recognition (specific others, shared frame, mother tongue). The shape of late-life loneliness (Linda, Frank, Helen) becomes a particular case of this work: the recognition-anchors are now memories of the dead, and the work is converting present-day life into something the deceased other would have recognized.

Negative cases sharpen the core category. P20 Frank's calm acceptance of “companionship with absent persons” could be read as not fitting — he is not doing the work; he has settled into it. But the closer read suggests that Frank has actually completed the work: he has built a stable relationship with the absent through ritual and routine, and his calm reflects the resolution of the work, not its absence. The core category survives the negative case and is sharpened by being asked to explain it.

3.4 Constant Comparison — The Engine

Constant comparison is the engine that drives every stage of the grounded theory procedure. It is the discipline of asking, with every new piece of data: how is this the same as what I've already coded? how is it different? does the category I have hold up, or does it need to be refined? does this datum suggest a new category? Constant comparison is what makes grounded theory more than freeform interpretation; it is the procedural discipline that holds the analyst accountable to the data.

Operationally, constant comparison happens at four levels:

1. Within-extract. Comparing parts of a single extract to one another, looking for internal contrast and movement.
2. Within-case. Comparing extracts within a single transcript, looking for the consistency or evolution of the participant's account.
3. Across-case. Comparing extracts and codes across transcripts, looking for cross-case patterns and contrasts.
4. Category-to-data. Comparing the developing analytic category back to specific instances in the data, asking whether the category does justice to what is actually there.

The discipline is not optional. Without continuous comparison, what looks like grounded theory devolves into thematic description with a fancier name. With continuous comparison, the categories the analyst builds carry the weight of their justification: every category has a paper trail of instances it explains and instances it had to be refined to handle.

3.5 Theoretical Sampling

Theoretical sampling is the practice of letting the developing analysis decide which next case or extract to read. Where conventional purposive sampling (Lesson 3) selects participants in advance to fill demographic or contextual quotas, theoretical sampling selects on the basis of what the emerging theory needs. The analyst might think: I have a category that captures loneliness inside companionship across young students and immigrants; I have not yet looked at how it manifests for someone who has chosen solitude as a deliberate practice (perhaps a monk or a long-distance hiker); let me recruit such a person to test whether the category extends or breaks.

In a study with live recruitment, theoretical sampling means deciding who to interview next. In a study with a fixed corpus — like your capstone — theoretical sampling means deciding which transcript or which extract to read next. The principle is the same: the choice is driven by analytic need, not by completeness.

For your capstone, theoretical sampling on a fixed dataset of 20 transcripts could look like this. You begin by coding three transcripts you have read fully (say, P01, P05, P15). An emerging category — loneliness-inside-companionship — suggests itself. You theoretically sample the corpus for transcripts that should test the boundaries of this category. Aarav's transcript (P06) is a near-confirming case: another instance of the phenomenon, in a different context. Frank's transcript (P20) is a candidate negative case: he describes companionship with the absent in ways that may or may not fit the category. Reading those two transcripts next is theoretical sampling. Your analytic memo should explain why those two were chosen and what they were chosen to test.

3.6 Theoretical Saturation

Theoretical saturation is the stopping rule. The classical formulation (Glaser & Strauss, 1967) is that you stop sampling when additional cases stop producing new conceptual content — when the categories are dense and stable and new data add nothing the existing categories cannot already accommodate. Strauss and Corbin (1998) extended this to three sub-criteria: no new properties of categories emerge; the relationships between categories are well-developed and validated; the categories cohere into a theoretical model.

Theoretical saturation is genuinely contested. Three critiques:

• The operational critique. “No new information” is in the eye of the analyst. Saturation has been used to justify ending data collection at sample sizes ranging from 6 to 60 in the published literature, with little consistency. Hennink and Kaiser (2022), in a systematic review, found that most published saturation claims could not be reproduced from the available methods information.
• The conceptual critique. Saturation assumes that data are samples from a finite conceptual space, which is an objectivist assumption Charmaz (2014) and others have argued is incompatible with constructivist epistemologies. If meaning is co-constructed, there is in principle always more to be constructed, and saturation is a research-design convenience, not a discovery.
• The practical critique. In fixed-dataset studies (like your capstone), saturation in the original sense is unreachable, because there are no further cases to sample. The best you can claim is “saturation of the analytic categories given the corpus.”

The defensible contemporary position is that saturation is a useful concept for guiding analytic effort but should not be claimed as if it were a binary state. Most thoughtful published grounded theory studies now report saturation alongside an explicit account of what would have shown saturation had not been reached — e.g., “by the eighteenth interview, no new properties of the four core categories emerged; analysis was stopped after the twentieth interview confirmed this.”

3.7 Memos as the Engine of Theoretical Development

Lesson 6 introduced memos as the analytic vehicle of qualitative work. Grounded theory amplifies this: memos are not optional in grounded theory, they are the substantive content of the analysis. The categories develop in memos; the relationships between categories are worked out in memos; the negative cases are tested in memos; the eventual core category is articulated in a memo before it becomes a finding.

Three kinds of memos do specific work in grounded theory:

1. Code memos: define a particular code, give exemplars, note its boundaries, and flag instances that nearly fit but do not.
2. Category memos: define an analytic category at a higher level, name its properties and dimensions, describe its relations to other categories, and explore its negative cases.
3. Theoretical memos: work out the overall structure of the developing theory, name the candidate core category, and show how the other categories relate to it.

The three kinds form a hierarchy of abstraction, and the analyst typically writes them in roughly that order, though all three are revised throughout. The final write-up of the substantive theory is, in effect, the integration of accumulated theoretical memos into a single coherent account.

Introduction and Overview

Sections 1 through 3 gave you the conceptual apparatus: matrix displays, the grounded theory variants, the open-axial-selective pipeline, constant comparison, theoretical sampling. This section is where the apparatus becomes practice. You will read the worked grounded-theory analysis of the loneliness corpus end-to-end, walk through the R and Taguette workflow for building a code-by-case matrix and for managing an open-coding pass, and complete the Week 7 capstone milestone. By the end of the section the choice between Option A (matrix) and Option B (grounded theory pass) should feel concrete, not abstract.

4.1 Worked Grounded-Theory Example: Loneliness Inside Companionship as a Candidate Core Category

This subsection narrates what a real grounded-theory pass on the loneliness corpus might look like, from initial open coding through to the articulation of a candidate core category. The narrative is more compressed than your own analysis will be, but its structure is faithful to the procedure.

Step 1 — Initial open coding pass on three transcripts

I read P01 (Maya), P05 (Linda), and P06 (Aarav) in full, then opened each transcript and coded line-by-line for the definitional passages and segment-by-segment for the rest. The first pass produced approximately 180 open codes across the three transcripts. A representative subset:

• naming a specific trigger; locating loneliness temporally; locating loneliness in transit space; marking the surrounded-yet-alone paradox; attributing to mediated attention; imagining invisibility; defining loneliness as non-recognition; distinguishing alone from lonely; describing post-hike alone-time as restorative; identifying nighttime ache; narrating phone-food-TV ritual; reporting failed connection attempts; reporting unexpected helper (cat & neighbour); criticizing the Vancouver no-talking rule; articulating shame at being 22 and lonely; using object as anchor (Bill's chair); refusing to clear the chair; noting bed-half-undisturbed; contrasting first-year acute grief with third-year quiet loneliness; identifying invisibility of long-term grief; using non-English term for crowd-loneliness (ekakitatvam); distinguishing solitude from loneliness through native language; cataloguing surrounding people who are not enough; locating loneliness in translation fatigue; describing the hour after the call home; reporting Diwali as the worst moment; reporting Sunday cooking rotation as small family; describing Canadian friendliness as kindness without closeness.

Step 2 — First round of constant comparison

Reading the codes side by side, several near-duplicates emerged. Maya's marking the surrounded-yet-alone paradox and Aarav's defining loneliness as crowd-paradox (ekakitatvam) were the same phenomenon. Maya's imagining invisibility and her defining loneliness as non-recognition were two facets of the same code; I merged them. Linda's using object as anchor and noting bed-half-undisturbed belong to a different cluster: marking absence through preserved objects.

After consolidation the open codes resolved into approximately fifty distinct codes organized into eight provisional category clusters: trigger types, embodied features, coping moves, unexpected helpers, structural critique, identity stake, connection-attempts-that-failed, and one provisional cluster I called recognition failures.

Step 3 — Theoretical sampling on the corpus

The recognition failures cluster looked analytically productive. It contained Maya's non-recognition on the SkyTrain, Aarav's translation fatigue, Linda's people-who-don't-know-Bill, and a strong hint of something similar in passages I remembered from skimming P15 (Amira) and P16 (Elena). I theoretically sampled four more transcripts: P11 (Helen) to test the cluster among an older never-married participant; P15 (Amira) to test it under refugee conditions; P18 (Chen) to test it under achievement-bound isolation; and P20 (Frank) as a candidate negative case (a participant whose accounts of companionship-with-absent-others seemed to resolve rather than dramatize the recognition-failure).

Step 4 — Axial coding

I ran an axial coding pass around the central category, using a hybrid Strauss-Corbin / Charmazian approach: I named the phenomenon, mapped its conditions and consequences, and identified the action strategies participants used, but I did not force every code into the six-cell paradigm template. The phenomenon emerged as recognition failure inside available presence: the experience of being in regular contact with others who, for structural or relational reasons, cannot recognize what one needs to have recognized. The causal conditions clustered into four sub-types:

• Cultural-linguistic gap (Aarav, Amira): the surrounding others do not share the participant's mother tongue, frames of reference, or memory archive.
• Biographical-specific gap (Linda, Helen): the surrounding others did not know the absent specific person who is the anchor of the participant's identity.
• Life-stage gap (Maya, Chen): the surrounding others occupy a different life-stage or are mediated through devices in ways that defeat conversion.
• Identity-revealing gap (Elena, partly P02 James): the participant has experienced an identity shift (coming out late, divorce after long marriage) that the previously available others cannot accommodate.

Action strategies sorted into four sub-types: seeking matched others (calling David, settlement worker Fatima, Syrian women's group); building portable rituals (cooking from memory, daily phone home, daily walks); numbing (overwork, scrolling, TV, food); and building a new frame (climbing gym, women-in-STEM group, the unexpected cat-and-neighbour link).

Step 5 — Selective coding and the candidate core category

The candidate core category that integrates all the above is “the work of converting presence into recognition.” Each participant's account can be read as describing this work, the conditions under which it fails (the four gap sub-types), the strategies for doing it (the four action sub-types), and the consequences when it succeeds or fails. The category survives the candidate negative case (Frank), and his calm-with-absent-others becomes legible as the resolution of the work, not its absence. The category is abstract enough to extend beyond the corpus (it could organize a study of dementia caregiver loneliness or of expatriate loneliness) and specific enough to retain analytic content.

Step 6 — Memos and theoretical write-up

Below are three short worked memos that an analyst at this stage might write. These are illustrative of the kind of writing your Week 7 capstone milestone is asking for — not finished paper prose, but disciplined analytic prose that establishes a category, names its properties, and reaches a small finding.

4.2 R Workflow — Building a Code-by-Case Matrix from a Taguette Export

The R workflow below assumes you have done some Taguette coding (or will). The pattern is the same whether you are building a small worked matrix as in Section 1 or a full code-by-case matrix as part of your capstone analysis. Taguette exports coded extracts as a CSV. R turns the CSV into a matrix and visualizes it.

library(tidyverse)
library(readtext)

# Ingest all 20 transcripts
transcripts <- readtext("../term projects/HSCI_841/transcripts/*.txt") %>%
  as_tibble() %>%
  mutate(participant = str_extract(doc_id, "^P[0-9]+_[A-Za-z]+"))

glimpse(transcripts)

# Ingest the Taguette export (your coded extracts CSV)
codes <- read_csv("../term projects/HSCI_841/exports/taguette_loneliness_export.csv") %>%
  rename(participant = document, code = tag, extract = content)

glimpse(codes)

focal_codes <- c("trigger", "embodied_feature", "coping_move",
                 "surprising_help", "structural_critique", "identity_stake")

focal_participants <- c("P01_Maya", "P05_Linda", "P06_Aarav",
                        "P11_Helen", "P15_Amira", "P16_Elena",
                        "P18_Chen", "P20_Frank")

matrix_long <- codes %>%
  filter(code %in% focal_codes, participant %in% focal_participants) %>%
  count(code, participant, name = "n_extracts")

matrix_wide <- matrix_long %>%
  pivot_wider(names_from = participant, values_from = n_extracts, values_fill = 0)

print(matrix_wide)

library(ggplot2)

ggplot(matrix_long, aes(x = participant, y = code, fill = n_extracts)) +
  geom_tile(color = "white", linewidth = 0.6) +
  geom_text(aes(label = n_extracts), color = "#03241F", size = 3.5) +
  scale_fill_gradient(low = "#E6F3F0", high = "#0B7B6B") +
  labs(
    title = "Loneliness corpus: 6 codes × 8 participants",
    subtitle = "Cell value = number of extracts coded; empty cells indicate informative absence or under-elicitation",
    x = NULL, y = NULL, fill = "# extracts"
  ) +
  theme_minimal(base_size = 11) +
  theme(axis.text.x = element_text(angle = 35, hjust = 1))

ggsave("capstone/wk07_matrix_loneliness.png", width = 9, height = 5, dpi = 150)

co_occur <- codes %>%
  filter(code %in% focal_codes) %>%
  distinct(participant, code) %>%
  mutate(present = 1L) %>%
  pivot_wider(names_from = code, values_from = present, values_fill = 0L) %>%
  select(-participant) %>%
  as.matrix()

co_occur_mat <- t(co_occur) %*% co_occur  # symmetric code × code matrix

print(co_occur_mat)

4.3 Taguette Workflow — Open Coding, Axial Relationships, Export

Taguette is the hand-coding companion to the R workflow above. Section 4.5 of Lesson 1 had you set up Taguette and upload one transcript. By Lesson 7 you should have most of the corpus loaded. The workflow below is specific to a grounded-theory-informed open-coding pass.

4.4 The Week 7 Capstone Milestone

The Week 7 milestone is where the matrix tradition and the grounded-theory tradition become a practical choice in your capstone. Both are defensible analytic strategies; both target the same underlying analytic move (comparison); the choice between them depends on whether your capstone's planned findings are best presented as a comparison-organized account (matrix) or as a substantive-theory account (grounded theory).

Bringing It All Together

Lesson 7 has given you two of the most powerful comparison engines in the qualitative analyst's toolkit. Matrix displays (Section 1) systematize comparison through visual arrangement, making cross-case and within-case patterns visible at the same time. Grounded theory (Sections 2 and 3) systematizes comparison through procedural iteration, building explanatory categories from data through constant comparison and theoretical sampling. The two traditions converge on the same underlying claim: qualitative analysis becomes credible analysis through disciplined comparison, whether the discipline is visual (matrix reading) or procedural (open-axial-selective coding).

What you take away from this lesson sets up Lesson 8 (Content Analysis), which turns the comparison engine outward to text corpora at scale and introduces the quantitative-leaning content-analytic tradition that sits beside the grounded-theory tradition. Lesson 9 (Schema and Narrative Analysis) returns to within-case interpretation. Lesson 10 (Discourse Analysis) attends to language as the object of analysis rather than the medium. Throughout, the comparison discipline you have built in Lesson 7 will be the methodological backbone.

The final reflection below asks you to step out of method-mode and name what you carry forward from Lesson 7 into the rest of the course. There is no single right answer; the goal is to leave the lesson with an articulated stance toward systematic comparison as the analytic posture that distinguishes credible qualitative work from skilled description.

This glossary collects the key concepts, people, and methodological stances introduced in Lesson 7. Use it as a reference while you work through the material, or as a review before the final assessment. Type in the search box to filter entries.