Introduction and Overview

Imagine sitting at a kitchen table in Burnaby with an 82-year-old man in long-term care who has just told you that his loneliness is “the empty space all those people used to fill.” A line like that is data. It has not been counted, scaled, or coded yet, but it is empirical (it comes from observing the real world), it is patterned, and it carries information about what loneliness is for the person living it. The problem of qualitative data analysis, the problem this course is about, is how to move from a stack of statements like that one to defensible knowledge claims a public-health audience will believe.

Much of the training in quantitative epidemiology works on data that arrive as numbers, or that can be coerced into numbers without much loss: case counts, surveillance series, survey scales, and the inputs to regression models. This course sits beside that training. It is the analytical companion that handles the evidence quantitative courses usually set aside: interview transcripts, field notes, archival documents, focus-group recordings, free-list responses, and the open-text comments at the back of every survey.

1.1 An Operational Definition

Bernard, Wutich, & Ryan (2017, p. 1) offer a working definition: qualitative data analysis is the search for patterns in non-numeric data and an explanation of why those patterns are there. Read that sentence carefully. There are three moving parts, and each one is doing work.

The search for patterns means the analyst is doing more than retelling what was said. Patterns are regularities: co-occurrences, sequences, contrasts, gradients, and absences. A theme that shows up in four out of twenty transcripts is a pattern. A code that always appears immediately after another code is a pattern. A topic that no one mentions, even though you asked about it, is also a pattern (a particularly informative one).

Non-numeric data is the easy part. The hard part is what counts as “non-numeric” in practice. A transcript is non-numeric. A photograph is non-numeric. A 30-second clip of a focus group laughing in unison is non-numeric. So is the layout of a clinic waiting room. We will be precise about the five kinds of qualitative data in a later section.

Explanation, the third part, is the move that separates analysis from description, what Geertz (1973) called the difference between thin and thick description (a thin account reports what was said; a thick account explains what it meant). A study that shows a pattern but does not explain it is, in Bernard, Wutich, and Ryan's terms, descriptive but not analytic. A loneliness study that finds the word “chair” mentioned in 11 of 20 transcripts has identified a pattern. It becomes analysis when the researcher proposes why, in this case perhaps because chairs are the most stable physical traces of absent people in domestic space.

1.2 Numbers and Words Are Not Different Substances

Introductory methods textbooks like to draw a sharp line between “quantitative” and “qualitative” research. In practice, the line is more porous than the textbook chapters that surround it would suggest. A few illustrations make the point.

The annual Canadian Community Health Survey contains thousands of pre-coded numeric items and a smaller set of open-text fields. Researchers routinely quantify the qualitative: they read the open-text comments, develop a coding scheme, count the resulting categories, and analyze the counts with chi-squared tests. They have just done qualitative analysis; they just did not stop there. Conversely, an interview-based grounded-theory study may begin with a frequency table of how many transcripts mention each emerging code. It has used a quantitative move (counting) inside a qualitative project.

The defensible distinction is not between numbers and words but between the type of question a study is answering and the kind of access the data gives the analyst to the phenomenon. Quantitative methods are typically the right choice when you want to estimate a population-level magnitude, test a pre-specified hypothesis, or measure an effect. Qualitative methods are typically the right choice when you want to discover what something is, how people make sense of it, or how a process unfolds. Both can use counting. Both can use words. The deeper choice is about whether you are measuring a known phenomenon or characterizing an under-described one.

1.3 The Textbook's Methodological Commitment

Bernard, Wutich, and Ryan are explicit about a stance that shapes the entire course: qualitative data analysis can and should be systematic, transparent, and replicable. The three words do specific work.

Systematic means that the analytic procedure is specifiable in advance (or at least in retrospect) and applied consistently across the dataset. If you decide to code every mention of the word “chair” in a transcript, you code every mention of the word “chair” in every transcript, and you do not code some while skipping others based on whether the mention is interesting. The systematicity is what makes the resulting pattern claim a real finding rather than an anecdote.

Transparent means that a reader of the eventual report can see what you did. This is the function of methods sections, audit trails, and codebooks (Lincoln & Guba, 1985). When a published qualitative paper says “themes emerged from the data,” Bernard, Wutich, and Ryan would consider that a methodological failure: themes do not emerge, analysts develop them through specifiable steps, and those steps should be in the paper (Braun & Clarke, 2006).

Replicable is the most contested of the three words and the most often misunderstood. Replicability in qualitative work does not mean that two analysts working on the same dataset would produce identical interpretations. Bernard, Wutich, and Ryan are clear that interpretation is partly perspectival. It means that two analysts following the same procedure would produce defensible interpretations and would identify similar patterns. The standard is not identity. The standard is “another competent researcher would arrive somewhere coherent with mine.”

1.4 Where Qualitative Work Sits in the Public-Health Evidence Landscape

One question that comes up early in every qualitative methods course is some version of: “Why would a public-health researcher do this kind of work?” The honest answer is that for many of the most important public-health questions, qualitative work is the only way in. Consider three examples.

Why do people refuse vaccines? You can count refusals with surveys (questionnaire design is taught in HSCI 341 Lesson 3, if you have taken it) and you can identify the predictors of refusal in a logistic regression (taught in HSCI 410 Lesson 3, if you have taken it). But if you want to understand the specific arguments people give themselves and each other for refusing, the narratives, the framings, the felt experience of distrust, you need interviews. The qualitative literature on vaccine hesitancy is what produced the interventions that the quantitative literature later tested.

What is it like to live with a chronic illness? The phenomenology (the felt, first-person experience) of, say, type 2 diabetes or long COVID is not legible in administrative data. Patient-reported outcome measures (scales of the kind examined in HSCI 410 Lesson 7, if you have taken it) give you scores; qualitative work gives you what the scores are measures of. A scale that says someone has a quality-of-life score of 0.62 tells you a number. The interviews behind such scales are what give the number its content.

Why did the program fail? Implementation science, the study of why evidence-based programs work in trials but stall in real-world rollout, relies heavily on qualitative methods. The numerical fact that a program failed is the starting point. The reasons are uncovered through interviews with implementers, observation of practice, document analysis of organizational policy, and the kinds of analytic moves you will learn in this course.

Introduction and Overview

Bernard, Wutich, and Ryan organize the entire enterprise of empirical research around four goals: exploration, description, comparison, and the testing of models. Every research study you have ever read or conducted can be located in one of these four (or, more commonly, in two or three of them at once). The goals are not a hierarchy. They are not phases of a single study. They are different jobs that empirical research can do, and each has its preferred methods. This section walks through the four goals, locates qualitative work within them, and uses the loneliness dataset as a running example.

2.1 Exploration

Exploration is what you do when you do not yet know enough about a phenomenon to make a hypothesis about it. The goal is to map the territory: to find out what is there, what the relevant categories are, what people consider important, and what the underlying mechanisms might be. Bernard, Wutich, and Ryan are explicit that qualitative work dominates exploration, because quantitative methods generally require that you have already decided what the variables of interest are, and exploration is the work of deciding that.

Most public-health questions begin in an exploratory phase, even if they later move into hypothesis-testing. When a new pathogen emerges, when a previously invisible population's experience comes onto the policy agenda, when a digital harm such as cyberbullying or AI-mediated relationships appears that no existing survey can ask about, the first scholars to study it are doing exploration. Their job is to give the rest of the field something to measure.

The loneliness dataset that anchors this course is, in part, exploratory. The transcripts contain accounts of loneliness from 20 people who differ in age, gender, life-stage, immigration status, caregiving role, and many other dimensions. A defensible analysis of these transcripts will help develop or refine the categories that future quantitative surveys of loneliness might use.

2.2 Description

Description is the careful characterization of a phenomenon: what is it, what does it look like, what are its dimensions, who experiences it, in what settings, with what consequences? Description is sometimes treated as the consolation prize of empirical research, the work you do when you cannot do anything “real.” Bernard, Wutich, and Ryan reject this framing emphatically. Many of the most influential studies in public health are descriptive: the Framingham Heart Study began as description, the BC Centre for Disease Control overdose mortality reports are description, the entire field of demography is description.

Both qualitative and quantitative methods can do description, and they describe different aspects of the same phenomenon. A quantitative survey can describe what percentage of adults in BC report being lonely in the past year (this is description by counting). A qualitative study can describe what loneliness feels like from the inside, what triggers it, what people do about it, and how they make sense of it (this is description by characterization). Both are valid. Often, both are necessary.

An analysis of the loneliness dataset is heavily descriptive. It characterizes loneliness as experienced by the 20 participants whose transcripts it draws on: its dimensions, its triggers, its embodied features, the meanings participants assign to it. That descriptive characterization is the bulk of what a qualitative health study contributes.

2.3 Comparison

Comparison is what you do when you have characterized a phenomenon and now want to know how it varies across groups, settings, or conditions. The classical home of comparison in epidemiology is the case-control study (does the exposure differ between cases and non-cases?) or the cohort design (does the outcome differ between exposed and unexposed?). Comparison is more often associated with quantitative work because of the statistical machinery available for it, but qualitative comparison is a real and rigorous activity.

Bernard, Wutich, and Ryan dedicate substantial parts of this course to comparison-based qualitative methods. Grounded theory's constant-comparative method (Glaser & Strauss, 1967), qualitative comparative analysis (QCA), and matrix analysis are all systematic ways to compare across cases, with words instead of variables, and to draw defensible inferences about why groups differ.

The course dataset supports comparison across the 20 loneliness transcripts. An analyst might compare how older participants describe loneliness with how younger participants describe it, or compare immigrant participants' accounts with native-born participants' accounts. The comparison is qualitative when the units being compared are texts or interpretive cases rather than rows in a spreadsheet, and when the conclusions are about patterns of meaning rather than effect sizes.

2.4 Testing Models

The fourth goal, the testing of theoretical models, is what most introductory methods textbooks treat as the pinnacle of empirical research. You have a theory; you derive predictions from it; you collect data and see whether the predictions hold. This is the standard logic of confirmatory quantitative work.

Qualitative work can do this too, though it is rarer and more contested. Analytic induction (a later module) is a qualitative model-testing approach: you specify a hypothesis, examine your cases, find a case that does not fit, and revise the hypothesis until it fits all cases. Qualitative comparative analysis (also a later module) tests which combinations of conditions are together sufficient for an outcome, using simple and/or/not (Boolean) logic. These are real model-testing exercises, just conducted with qualitative cases.

An analysis of the course dataset could include a small model-testing element. For example, an analyst might hypothesize that participants who describe loneliness in existential terms (a feature of one's life-stage) cope through reframing, while participants who describe it in situational terms (a feature of current circumstances) cope through behavioral change, and then check the hypothesis against the 20 transcripts. That is qualitative model-testing.

Introduction and Overview

Bernard, Wutich, and Ryan organize qualitative data into five kinds: physical objects, still images, sounds, moving images, and texts. The categorization may feel pedantic until you realize that which kind of data you have shapes what analytic moves are available to you. A photograph and a transcript both look like “qualitative data” but they are coded differently, sampled differently, and reported differently. This section walks through the five kinds, with public-health examples for each, and then introduces the course's loneliness interview dataset.

3.1 Physical Objects

Anthropologists call this material culture: the artefacts people make, use, exchange, and discard. In public health, the relevant physical objects include medication packaging, syringe-exchange kit contents, vaccine cards, ad hoc harm-reduction supplies, mobility aids, the layout of a clinic waiting room, the contents of someone's medicine cabinet, and the materials available (or unavailable) in a school health office. Object-based analysis is not common in mainstream public-health research but it is increasingly important in implementation science, environmental health, and Indigenous health research where physical context carries meaning that words do not.

3.2 Still Images

Photographs, hand-drawn maps, satellite imagery, screenshots of social-media posts, anatomy diagrams in patient education materials, public-health campaign posters. Still-image analysis is a recognized sub-specialty (visual sociology, visual anthropology) with its own techniques: content analysis (a later module) is regularly applied to images, and the photovoice method, in which participants take their own photographs and discuss them, is a staple of community-based participatory research.

3.3 Sounds

Recorded speech is the most common form, but other audio data is also analytically tractable: the sound of a clinic at peak hours, the music played in a hospice, the auditory environment of a school cafeteria. The vast majority of qualitative analysis in health research, however, begins with sound (an interview recording) and is converted to text (a transcript) before analysis. This conversion step, transcription, is itself an analytic act, and transcription conventions receive sustained attention in a later module.

3.4 Moving Images: Video

Video is sound plus image plus time. Clinical encounter recordings, simulated training scenarios, ethnographic field recordings, TikTok health-influencer content, and telehealth call recordings are all qualitative data. Video analysis is more time-consuming than audio or text but allows attention to non-verbal communication, embodied action, and spatial arrangement. Conversation analysis (a later module) has historically privileged video for exactly this reason.

3.5 Texts

Texts are, by far, the most common qualitative data in contemporary health research. They include:

• Interview transcripts: the most familiar form, and the form the course dataset takes.
• Focus-group transcripts: like interviews, but multi-party and with conversational dynamics.
• Field notes: the researcher's own written record of observation.
• Documents: policy papers, clinical guidelines, organizational reports, news articles, archived correspondence.
• Open-text survey responses: the qualitative tail of an otherwise quantitative instrument.
• Social-media corpora: posts, comments, threaded discussions, hashtag streams.
• Patient-generated text: diaries, symptom journals, illness blogs.
• Free-list responses: short open-ended elicitation data used in cultural domain analysis (a later module).

The reason texts dominate this course is partly practical (they are cheap to store, easy to share, computationally tractable) and partly principled: text is the form most amenable to the systematic, transparent, replicable analytic procedures Bernard, Wutich, and Ryan advocate. The methods you learn in this course will be applicable to images, video, and audio with adjustments, but the default unit of analysis is text.

3.6 The Course Dataset

The course dataset consists of 20 transcripts of semi-structured interviews about experiences of loneliness, conducted with adults in British Columbia. The dataset is paired with the interview guide used to elicit it. Both are stored in the course materials. The 20 participants vary deliberately across age (18 to 82), gender (women, men, non-binary), life-stage (students, parents, retirees, widows), and circumstance (immigration, caregiving, romantic dissolution, late-life coming out, occupational burnout, refugee resettlement, neurodivergent identity, long-term care).

Introduction and Overview

The first three sections gave you definitions, the four research goals, and the five kinds of qualitative data. This section turns operational. It unpacks the systematic-transparent-replicable manifesto in working terms, maps the commitments onto the trustworthiness criteria used in much of the qualitative literature, sets up Taguette, the coding software used in this course, and ends with the positionality memo as a worked application of the transparency commitment.

4.1 Systematic Analysis in Operational Terms

A systematic procedure has three features. First, it is specifiable: you can write it down and another researcher can follow it. Second, it is consistent: the same rule is applied to every case. Third, it is iterative when needed but documented when revised: if you change your codebook halfway through coding, you note when and why, and you re-code the earlier transcripts under the new scheme.

Unsystematic analysis, the kind Bernard, Wutich, and Ryan critique, reads as “we read the transcripts and noted what stood out.” That is not analysis; that is intuition. The systematic version of the same activity reads as “two analysts independently coded each transcript using a codebook developed from the first five transcripts; disagreements were resolved through discussion and the codebook was revised twice during the analysis (revisions documented in the audit trail).”

4.2 Transparency: What You Owe Your Reader

Transparency is what you owe your reader. Specifically, you owe them four things:

1. An explicit account of how you got the data: sampling logic, recruitment, interview procedure, and the instrument you used.
2. An explicit account of how you analyzed the data: coding procedure, the codebook (often in an appendix), how disagreements were handled, and what software you used.
3. An explicit account of your positionality: who you are, what you brought to the interpretation, and what you might have missed.
4. An explicit account of the limitations: what your design and dataset cannot tell you, even if you wish they could.

The convention in contemporary public-health qualitative research is that all four are in the methods section or an appendix, not in the body of the discussion as an afterthought.

4.3 Replicability: Coherent, Not Identical

Replicability in qualitative work is best operationalized (see Tracy, 2010) as the question: if a competent researcher with no prior knowledge of your study had access to your dataset, your codebook, and your methods, could they reach interpretations that are coherent with yours? The answer should be yes, even if not identical. Where the answer is no, the cause is usually one of: undocumented analytic decisions, idiosyncratic coding, or interpretations that go far beyond what the data support.

One way to test replicability is the intercoder reliability check (Morse, Barrett, Mayan, Olson, & Spiers, 2002): two analysts independently apply the codebook to the same subset of transcripts and the agreement is calculated. You will do this later in the course. Intercoder reliability is not the only test of replicability, and for some interpretive methods it is the wrong test, but it is the most common operational standard in applied health research.

4.4 Trustworthiness: Lincoln and Guba's Four Criteria

Many qualitative researchers and journal reviewers describe rigour in a different vocabulary from the one this course uses. Lincoln and Guba (1985) proposed four criteria of trustworthiness. Credibility asks whether the findings represent the participants' accounts faithfully. Transferability asks whether a reader has enough description of the setting and sample to judge whether the findings apply elsewhere. Dependability asks whether the analytic process was consistent and is documented well enough to be followed. Confirmability asks whether the findings can be traced to the data rather than to the analyst's preferences. If you have taken HSCI 207, Lesson 11 Section 4 named these criteria; this section shows how they relate to the three commitments.

The two frameworks are compatible. Systematic, transparent, and replicable describe properties of the analytic procedure, while Lincoln and Guba's criteria describe what a reader should be able to conclude about the findings. Each criterion is met through specific techniques, and the table maps the criteria, the commitments, and the techniques onto one another. Either vocabulary can be used in a methods section, provided that the techniques it names were actually carried out.

Peer debriefing is the one technique in the table that no later lesson treats at length. It means reviewing emerging interpretations with a colleague who is not involved in the analysis and whose role is to ask how each claim is supported and what alternative readings were considered (Lincoln & Guba, 1985). It is close in spirit to investigator triangulation, but the colleague questions the analysis rather than coding the data independently, and the analyst records the questions raised and the changes made in an operational memo.

4.5 Taguette for Hand-Coding

Taguette is a free, open-source qualitative coding application. It does what NVivo and ATLAS.ti do (highlight passages, attach codes, build a codebook, export coded extracts) without the licence fee. It runs in your browser. There is nothing to install if you use the hosted version; you can also install it locally.

4.6 The Positionality Memo

The transparency commitment starts before any coding. A positionality memo is a short piece of writing, often about 500 words, in which an analyst records who they are in relation to the data before reading it closely: prior experience with the topic, training and disciplinary background, the assumptions they carry in, and what they might therefore be inclined to notice or to overlook. It is the qualitative equivalent of declaring covariates before fitting a model, because it puts on the record what the later systematic steps were built on.

A practical sequence is to read the interview guide, then read three transcripts side by side (P01 Maya, P11 Helen, and P15 Amira make a varied starting set from the course dataset), and then write the memo before any codes are created. The memo becomes the first entry in the audit trail and the raw material for the positionality statement in a methods section. Analysts return to it as coding proceeds and add dated entries when they notice their own experience shaping what they see. Because a reader can then see where the analyst stood, the memo supports the confirmability criterion described in Section 4.4.

Bringing It All Together

This lesson has set up the conceptual and operational vocabulary that the rest of this course depends on. The four-research-goal frame (an earlier section), the five-kinds-of-data frame (an earlier section), and the systematic-transparent-replicable manifesto (an earlier section) are the scaffolding the next eleven lessons will hang their specific techniques on. The positionality memo, written after reading three transcripts, shows the transparency commitment at its smallest scale, and the same habits of documentation carry through every later lesson.

What you take away from this lesson sets up the next lesson, Research Questions, Theory, and the Literature, which shows how a researcher formulates a specific research question. Later lessons build the design side, sampling and data collection, using the course dataset as a worked example; the design vocabulary is what a qualitative methods section is written in. Later lessons move into analysis proper.

The final reflection below asks you to step out of method-mode and name what you carry forward from this lesson into the rest of the course. There is no single right answer; the goal is to leave the lesson with an articulated stance.

Final Knowledge Assessment

This glossary collects the key concepts, people, and methodological stances introduced in this lesson. 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.