Qualitative Data Collection
Indirect Observation, Direct Observation, Elicitation, Transcription & Field Notes
Learning objectives for this lesson:
- Distinguish indirect observation, direct observation, and elicitation as the three families of qualitative data collection.
- Identify the analytic uses (and limits) of behavior traces, archival records, and secondary qualitative datasets.
- Compare the three direct-observation methods (participant observation, continuous monitoring, and spot observation), locating yourself on the participant-observation spectrum from complete observer to complete participant and explaining what each position can and cannot see.
- Compare unstructured, semi-structured, and structured interviews on what they elicit, how they are designed and piloted, and what they cannot reach, including how interviewing for interpretative phenomenological analysis differs from the semi-structured format.
- Explain when focus groups outperform individual interviews and why their data are produced by group interaction.
- Describe the cultural-domain elicitation methods (free listing, pile sorts, triads, rankings) previewed here for a later module, and the place of open-ended survey items among elicitation methods.
- Explain why transcription is an analytic act, distinguishing verbatim conventions, Jefferson notation (preview), intelligent verbatim, and clean verbatim by what each choice forecloses, and identify the convention the course dataset uses.
- Write disciplined field notes using the Emerson, Fretz & Shaw progression: jottings → expanded notes → analytic memos.
- Explain why familiarisation with the data comes before coding, and how a familiarisation log records it.
This course was developed by Dr. Kiffer G. Card based on Bernard, H. R., Wutich, A., & Ryan, G. W. (2017). Analyzing Qualitative Data: Systematic Approaches (2nd ed.). SAGE Publications.
Indirect Observation: Behavior Traces, Archives & Secondary Analysis
Introduction and Overview
Most students arrive at a qualitative methods course assuming that “qualitative data collection” means interviews. It does, eventually, dominate the lesson. But we open with a deliberately broader frame. The data-collection landscape is organised around three families: indirect observation (where the researcher never interacts with the people whose behavior is being studied), direct observation (where the researcher is present and looks at what is happening), and elicitation (where the researcher actively asks people to produce data, through interviews, focus groups, free lists, or written responses). The three families are not a hierarchy. They are tools that answer different kinds of questions, that cost different amounts of money and time, and that come with different ethical obligations.
This section concerns the first family, indirect observation, because it is the one most often overlooked and the one with the most counterintuitive payoff. Indirect data are data that already exist in the world, left by past behavior, stored in institutional archives, or generated by other researchers for other purposes, and that the analyst harvests without ever asking a participant a question. Eugene Webb and colleagues coined the term unobtrusive measures for this family in their 1966 classic of the same name, and it remains one of the most fertile underused regions of qualitative methodology in public health.
Learning Objectives
- 1.1 Define indirect observation, distinguish it from direct observation and elicitation, identify its three sub-types (behavior traces, archival records, and secondary qualitative datasets), and explain why its data are nonreactive and longitudinal.
- 2.1 Describe behavior traces, distinguish accretion from erosion measures, and give a public-health example, including one that draws on the loneliness phenomenon studied in the course dataset.
- 2.2 Explain what archival records can and cannot tell an analyst, and why the selection bias built into every archive must be named explicitly.
- 2.3 Explain the promise and the hazards of secondary analysis of existing qualitative datasets, and articulate the ethical issues distinctive to it.
- 2.4 Weigh the analytic advantages of indirect data (no reactivity, longitudinal reach, low cost) against their limits (selection biases, missing context, the impossibility of probing) to judge when indirect methods are the right first choice.
The section first sets out the logic of indirect observation, then describes behavior traces, archival data and secondary qualitative datasets in turn, and closes with the conditions under which indirect methods are the right first choice.
1.1 The Logic of Indirect Observation
Indirect observation begins with a simple insight: people leave traces. They wear paths into grass between buildings, they leave fingerprints on museum glass, they fill landfills with packaging, they post on social media, they generate medical records, they file tax returns, they get arrested, they buy products, they vote. Each of those traces is data, and the discipline of harvesting them is, in principle, a qualitative discipline: the analyst is reading the world the way an interviewer reads a transcript. Eugene Webb, Donald Campbell, Richard Schwartz, and Lee Sechrest's Unobtrusive Measures: Nonreactive Research in the Social Sciences (1966; second edition 2000) is the canonical methodological statement. This lesson treats the Webb tradition as foundational for two reasons.
First, indirect data are nonreactive. When you interview someone about their loneliness, they know they are being studied; what they say is shaped by the relationship, by the setting, by the desire to appear coherent, by the willingness to disclose. When you analyse a transit log showing how often someone uses public transit alone versus in company, the data have no such reactivity, since the person was not performing for you. Reactivity is the qualitative-research equivalent of measurement bias, and indirect data are, in this respect, methodologically cleaner than elicited data on the same phenomenon.
Second, indirect data are typically longitudinal in ways that elicitation cannot match. An interview captures one moment, perhaps two if you do a follow-up. An archival record can stretch back decades. A behavior trace can accumulate over the entire life of the trace-generating activity. If you want to know how loneliness in a community has changed across thirty years, no interview design can give you that, but a content analysis of obituaries, of personal ads, of community-newspaper letters to the editor, of city-council minutes mentioning “isolation” can.
The cost of these advantages is a cost we will return to throughout the section: with indirect data you cannot ask why. You can see the pattern. You cannot ask the person who produced it what it meant to them. Combining indirect observation with elicitation, the two-source design, is often the strongest move.
The three sub-types of indirect observation share this limitation. The subsections that follow describe each in turn, beginning with behavior traces.
2.1 Behavior Traces
This subsection concerns behavior traces, the first of the three sub-types of indirect observation. The tabs below give a brief overview of all three sub-types and of the conditions under which indirect methods are preferred, and the paragraphs after them return to behavior traces in detail.
Physical evidence left by behaviour: worn carpet paths, garbage composition, library-book usage patterns, app analytics, swipe-card logs. Webb et al.’s 1966 Unobtrusive Measures is the founding text. Useful when self-report is unreliable, sensitive, or socially constrained.
Pre-existing records made for purposes other than research: clinical chart notes, government minutes, court transcripts, news media archives, parish registers, organizational policy documents. Almost every public health office sits on a treasure of underused archival data.
Re-analysis of qualitative datasets collected by other researchers. The UK Data Service held more than 1,000 qualitative and mixed-methods collections by December 2016; the Murray Research Archive at Harvard holds further qualitative collections, including audio and video. Important ethical considerations around consent and confidentiality.
Indirect methods are preferred when: (1) the behaviour is sensitive and self-report is suspect, (2) the population is hard to reach but leaves records, (3) historical reconstruction is needed, or (4) you want to triangulate against direct measures. Often combined with direct methods rather than replacing them.
Behavior traces are physical or digital records left by past action. Webb and colleagues divided them into two sub-types: accretion measures (things that have built up, such as graffiti, garbage, and fingerprints) and erosion measures (things that have worn down through use, such as the worn linoleum in front of a museum exhibit indicating how popular it was, the most-thumbed page of a Bible in a hotel room, the fading paint on a heavily-used playground). The distinction is more conceptual than practical; the analytic logic in both cases is the same.
In public-health research, behavior traces are useful for the questions that respondents cannot or will not answer accurately. Two examples make the point. Zuccato and colleagues (2005) measured cocaine and its main urinary metabolite in the River Po and in urban wastewater in Italy, and estimated community use at a level that greatly exceeded official national figures. Srigley and colleagues (2014) used an electronic system that logged every use of the hand-rub and soap dispensers on two transplant units of a Toronto hospital; dispensers within sight of a hand-hygiene auditor were used about three times as often as when no auditor was visible, so audits by observers overstate everyday practice.
The qualitative move with behavior traces is interpretive: what does this trace mean? The answer requires the same disciplined work as coding a transcript: specify what you are looking for, code consistently across cases, and be transparent about your interpretive leaps. A worn path between two buildings means someone walked there many times. It does not, by itself, tell you who they were, why they preferred that route, or whether they enjoyed it. The trace is the starting point of analysis, not its conclusion.
The box below applies this interpretive logic to loneliness, listing behavior traces that could be studied without interviewing anyone and explaining how they can complement the accounts in the course transcripts.
A loneliness example
If you wanted to study loneliness without interviewing anyone, behavior traces could give you a great deal. Library-borrowing records (which books are checked out by patrons who borrow alone vs. in groups), grocery-store self-checkout queue patterns at 9 p.m. on a Sunday (the time of week Maya, in the course dataset, names as the loneliest), the contents of the “food for one” freezer section across socioeconomically different neighbourhoods, the timestamps of who turns on their porch light first at dusk. None of these is conclusive on its own. Combined with the kinds of accounts in the course transcripts, they give the qualitative claim a quantitative shadow that makes it more defensible to a public-health policy audience.
Behavior traces record past action directly but say little on their own about its meaning. The next subsection turns to the second sub-type of indirect observation, the records that people, organisations and governments generate as a by-product of operating.
2.2 Archival Data
This subsection defines archival data, sets out what archival sources can and cannot reveal about loneliness, and considers the selection bias built into every archive. The key-insight box below states the central caution at the outset: every archive is a sample of what was recorded by people in a position to record it, and a defensible analysis names that selection explicitly.
Key insight - The archival selection bias is invisible
Archival data are seductive: they already exist, they are often free, they cover periods you could not reach prospectively. But every archive is a sample of what was recorded by people empowered to record. Clinical charts overrepresent compliant patients seen by attentive clinicians; court records overrepresent successful prosecutions; newspaper archives overrepresent stories that fit editorial framings of the era. The selection bias is not a flaw to control for; it is a structural feature of the data. A defensible archival analysis names the selection explicitly.
Archival data are the institutional records people, organisations, and governments generate as a by-product of operating. The category is large. It includes:
- Institutional records: clinic charts, school-attendance registers, court transcripts, council minutes, child-welfare case files, ambulance dispatch logs, organ-donation registries.
- Historical documents: diaries, letters, newspapers, pamphlets, public-health reports from earlier eras, autopsy reports, missionary records, parish death books.
- Media corpora: television news transcripts, podcast episodes, social-media posts, advertising, government information campaigns.
- Personal documents: in the hands of researchers via consent or via the public domain: published memoirs, blogs, online support-group threads, GoFundMe campaign descriptions.
Archival work has historically been treated as a sub-discipline of history rather than as qualitative methodology, and the artificial boundary has cost the social sciences. The methods of qualitative analysis you will learn in later modules, including theme identification, content analysis, narrative analysis, and grounded theory, all apply to archival data. A clinic chart can be coded; a newspaper editorial can be theme-analysed; a corpus of council minutes can be content-analysed for keyness (which words become distinctively frequent) over time. The analytic discipline is exactly the same as for interview transcripts; the data just arrived without an interviewer in the picture.
Table 4.1 applies this reasoning to loneliness, listing five archival sources with what each can and cannot tell an analyst about the experience.
Table 4.1. Five archival sources relevant to loneliness research, with what each can and cannot tell an analyst about loneliness.
| Archival source | What it can tell you about loneliness | What it cannot tell you |
|---|---|---|
| Coroner reports of solitary deaths | Patterns in who dies alone, in what neighbourhoods, in what seasons, after how long undiscovered | Whether the deceased felt lonely, what their social life looked like before death |
| Newspaper personal ads (1980s–2010s) | Historical changes in how people advertise loneliness and seek connection; what language is and is not sayable | Whether the advertisers found what they were looking for |
| City-council minutes mentioning “social isolation” | When loneliness became politically nameable; what kinds of solutions were proposed and by whom | Whether the proposed solutions worked for the people the policy targeted |
| Online bereavement-support forum threads | How bereaved people describe loneliness to one another, what advice circulates, what stays unsaid | Whether non-posting bereaved people experience loneliness the same way (massive sampling concern) |
| Family-doctor chart notes (with consent) | How clinicians record patient loneliness; what gets coded vs. left in free text; medication histories | The patient's own framing; what they did not tell the doctor |
The selection-bias problem. The big methodological concern with archival data is that they were not produced for your study. They were produced for someone else's purpose, by someone else, with someone else's selection criteria, in someone else's institutional context. A coroner's report describes the deaths the coroner saw, written in the way the coroner was trained to write them. Newspaper personal ads were written by people who could afford to place them and read by people who read that newspaper. Online forum posts were written by people who used that forum, in English, with the literacy to compose a post, and who chose to post rather than to lurk. Each of these selection mechanisms is a population-defining filter, and your analytic claims have to be calibrated to what the data can actually support.
This course's recommendation is conservative: treat archival data as one source among others, identify what the source can and cannot represent, write the limitations into your methods section explicitly, and resist the temptation to generalise beyond the population whose traces actually entered the archive.
Archival records were produced by others for other purposes, and their value depends on naming what they can and cannot represent. The same concern applies to the third sub-type, qualitative datasets collected by other researchers, which the next subsection takes up.
2.3 Secondary Analysis of Existing Qualitative Datasets
Secondary qualitative analysis, using interview or focus-group data collected by other researchers for other studies, is the youngest of the three indirect-observation sub-types and the one growing fastest. The Qualitative Data Repository at Syracuse, the UK Data Service's Qualitative Bank, the SAGE Research Methods Cases collection, and a growing number of journal data-sharing requirements are making qualitative datasets available the way quantitative datasets have been available for decades.
The methodological promise is real. A graduate student who could never afford to do twenty new interviews on, say, vaccine hesitancy in rural Manitoba can re-analyse a deposited 2018 dataset on the same topic, asking new questions of the existing transcripts. A researcher building a meta-ethnography (a synthesis of multiple qualitative studies, the qualitative analogue of a meta-analysis) can use the deposited transcripts rather than only the published papers about them.
The methodological hazards are also real. Four stand out. First, the original interview guide constrained what got asked; if your research question requires a probe the original interviewer never offered, the data are silent on it. Second, the analyst was not in the room and cannot recover the relational and embodied context the original researcher had. Third, the participants consented to one study; using their words for a different study requires either explicit re-consent (often impossible) or a defensible ethics argument that the secondary use is sufficiently consonant with the original consent. Fourth, the analytic frame you bring may be inappropriate for the kind of relationship the original interviewer built.
The stance taken here is permissive but cautious: secondary qualitative analysis is legitimate and increasingly necessary, but it should be done with explicit acknowledgement of what was lost and what was not collected, and with humility about interpretations that go beyond what the original interviews can support.
The box below applies these cautions to the course's own loneliness dataset, whose analysis is in effect a form of secondary analysis.
A word about the course dataset
This course's loneliness dataset is itself a kind of pre-collected dataset, since the transcripts were generated as a teaching corpus and analyzing them is, in effect, secondary analysis. Familiarisation (Section 4, 9.1 Familiarisation Before Coding) means reading the transcripts deeply, in part because that deep reading is the qualitative analogue of getting to know a quantitative dataset before you model it. You cannot ask Maya what she meant when she said the SkyTrain at 9 p.m. on Sunday was the loneliest place in her life. The transcript is the data. Knowing the data well, before you start coding it, is the first move of disciplined secondary analysis.
With the three sub-types of indirect observation described, the final subsection considers when indirect methods are the right first choice and when they are the wrong one.
2.4 When to Use Indirect Methods
Indirect observation is the right first choice when one or more of the following is true: (a) the population is hard to reach by elicitation (e.g., the deceased, people in conflict zones, people who would not consent to interview); (b) the time horizon is long (decades or centuries); (c) reactivity is a serious concern (the topic is socially sensitive or self-presentation will distort responses); (d) the budget is small; (e) the research question is about an institutional or population-level pattern rather than about lived experience.
It is the wrong choice when you need to know what something felt like from the inside, when you need to probe a participant's understanding, or when the trace itself is missing the most interesting part of the phenomenon. For most loneliness research, indirect observation is a complement to elicitation, not a replacement. The chair Helen sits in (an accretion measure of where she spends her time) tells you something. What she says about the chair tells you something else. The disciplined qualitative researcher uses both.
This section has described indirect observation, its three sub-types (behavior traces, archival records and secondary qualitative datasets), and the conditions under which indirect methods suit a study. The key takeaways, reflection and knowledge check that follow consolidate these points before Section 2 turns to direct observation.
Key Takeaways
- 1.1 Indirect observation harvests data that already exist in the world, left by past behavior, stored in archives, or collected by other researchers, without the analyst ever interacting with the people studied; Webb and colleagues called these unobtrusive measures. It differs from direct observation, where the researcher is present and watches, and from elicitation, where the researcher asks people to produce data. Its two analytic advantages are that the data are nonreactive, because no one was performing for the researcher, and that they often reach back decades. Its central cost is that you cannot ask the person who left the trace what it meant.
- 2.1 Behavior traces are physical or digital records of past action, divided into accretion measures (things built up, such as graffiti, garbage, and fingerprints) and erosion measures (things worn down, such as worn linoleum or a dog-eared page). They are most useful for questions respondents cannot or will not answer accurately, as when cocaine residues in Italian river water and wastewater gave an estimate of community use far above official figures. Reading a trace is interpretive work that needs the same discipline as coding a transcript, and the trace is the starting point of analysis, not its conclusion. For loneliness, traces such as late-night self-checkout patterns can give a qualitative claim a quantitative shadow.
- 2.2 Archival data are the records that people, organisations, and governments generate as a by-product of operating: institutional records, historical documents, media corpora, and personal documents. Every method of qualitative analysis applies to them, but every archive is a sample of what was recorded by people in a position to record it, so selection bias is a structural feature of the data rather than a flaw to control. As Table 4.1 shows, a coroner's report can reveal who dies alone but not whether they felt lonely. A defensible archival analysis names the selection, writes the limits into the methods section, and resists generalising beyond the population whose traces entered the archive.
- 2.3 Secondary qualitative analysis re-uses interview or focus-group data collected by other researchers, now increasingly available through repositories and data-sharing requirements. It makes new questions affordable and supports meta-ethnography, but it carries four hazards: the original guide constrained what was asked, the analyst cannot recover the context of the room, the participants consented to a different study, and the analytic frame brought to the data may not fit the original interviews. The distinctive ethical issue is consent, which requires either re-consent or a defensible argument that the new use is consonant with the original one. Analysing the course's teaching corpus is itself a form of secondary analysis.
- 2.4 Indirect observation is the right first choice when the population is hard to reach by elicitation, the time horizon is long, reactivity is a serious concern, the budget is small, or the question concerns an institutional or population-level pattern. It is the wrong choice when you need to know what something felt like from the inside or to probe a participant's understanding. For most loneliness research, indirect data complement elicitation rather than replace it: the chair Helen sits in and what she says about the chair tell you different things.
Reflection
You are doing a qualitative study of loneliness in long-term-care facilities. Interviews with residents are slow, expensive, and ethically delicate. What indirect data sources might supplement your interview corpus? Name two, and for each, say what it can and cannot tell you.
Minimum 20 characters required.
Question 1: The Webb et al. (1966) term unobtrusive measures refers to:
Question 2: An accretion measure in the Webb tradition is:
Question 3: Which of the following is NOT a methodological hazard of secondary qualitative analysis identified in this lesson?
Direct Observation: Participant Observation, Monitoring & Time Allocation
Introduction and Overview
Direct observation is the family of methods in which the researcher is physically (or virtually) present and watches behavior happen. It is the oldest qualitative methodology in cultural anthropology, since ethnographers have been doing it for more than a century (Hammersley & Atkinson, 2019), and it remains a workhorse of contemporary public-health research, especially in implementation science, clinical-encounter research, organisational ethnography, and Indigenous health. This section organises the discipline around three sub-types: participant observation (immersive engagement over time), continuous monitoring (structured watching of specific behavior streams), and spot observation / time-allocation studies (sampled snapshots of who is doing what, when).
The defining intellectual move in all three sub-types is what Spradley (1980) called “the discipline of seeing.” Ordinary perception is selective and habitual; it filters out the familiar and concentrates on what is novel or salient. Disciplined observation reverses the filter. The observer trains themselves to attend to what is so ordinary it has become invisible: the layout of chairs, the rhythm of who speaks and who waits, the timing of who arrives and who leaves, what is on the walls, what people wear, what they eat, when they touch and when they don't. The discipline is not natural. It is trained.
Learning Objectives
- 3.1 Distinguish the three sub-types of direct observation (participant observation, continuous monitoring, and spot observation), and locate yourself on Spradley's participant-observation spectrum from complete observer to complete participant, explaining what each position can and cannot see.
- 3.2 Describe the discipline of “explicit awareness” (Spradley) and the techniques used to build it.
- 3.3 Explain when continuous monitoring outperforms self-report (and vice versa).
- 3.4 Set up a basic time-allocation study.
- 3.5 Explain how the reactivity problem affects direct observation, and judge when direct methods are the right first choice.
The section begins with participant observation and Spradley's spectrum of researcher involvement, turns to the skill of explicit awareness, then describes continuous monitoring and spot observation, and closes with the conditions under which direct methods are the right first choice.
3.1 Participant Observation: The Spradley Spectrum
Participant observation is the first of the three sub-types of direct observation, the one built on immersive engagement over time. This subsection describes the method and the spectrum of researcher involvement that Spradley set out. The accordion below summarises four positions an observer can take in a setting, from complete observer to complete participant, together with the choice between continuous monitoring and spot observation.
The researcher observes without participating. Often invisible to participants. Used for natural-behaviour studies in public spaces. Tradeoff: minimal influence on the scene; minimal access to meanings, motivations, or backstage talk.
The researcher participates marginally but is identified as a researcher. Brief field visits, structured site walk-throughs, episodic observation. Tradeoff: easier to enter and leave; less embedded knowledge.
The researcher is a participant in the setting and openly identified as also doing research. The most common form in modern ethnographic and implementation research. Tradeoff: deep access; identity management work; complex consent.
The researcher is fully embedded as a member, with research role hidden. Now ethically constrained (most IRBs will not approve without strong justification). Historical examples: studies of religious cults, criminal organizations, healthcare-system ‘mystery shopper’ designs.
Continuous monitoring: observe one focal person or process across the full activity. Spot observation / time allocation: sample many people at random instants to estimate how time is spent across a population. The choice is between depth-on-one and breadth-across-many.
Participant observation (DeWalt & DeWalt, 2011) is the ethnographic core of qualitative methodology: prolonged engagement in a setting, during which the researcher both watches and (to varying degrees) participates. The classic anthropological model is Bronislaw Malinowski's Trobriand Islands fieldwork, with years of residence, daily participation, and slow accumulation of insight. The same logic applies to a public-health researcher who spends six months attending the weekly meetings of a peer-support group, eating in the staff lounge of a community clinic, or volunteering at a needle exchange. The method's power comes from time: prolonged engagement reveals patterns invisible to a one-time interview.
James Spradley's Participant Observation (1980) is the canonical methodological treatment. Spradley described degrees of participation running from nonparticipation to complete participation, and Gold (1958) had earlier named four field roles, from complete participant to complete observer. This lesson combines the two into a five-position scale, which it calls the Spradley spectrum, and Table 4.2 names the trade-offs at each position.
Table 4.2. Five positions of researcher involvement, combining Gold's four field roles with Spradley's nonparticipation, with what each looks like, what it sees well and what it cannot see.
| Position | What it looks like | What it sees well | What it cannot see |
|---|---|---|---|
| Complete observer | Researcher is fully external to the setting; people may not even know they are being observed (one-way mirror, public-space ethnography). | Behavior in its “natural” state, unaffected by the researcher. | Meaning. The observer cannot ask why; cannot probe. |
| Observer as participant | Researcher is known to be a researcher; spends time in the setting but does not take on participant roles. The default of most contemporary clinic-based ethnography. | Routine practice; staff-side dynamics; the ordinary working of the setting. | What it feels like to be a patient/client/insider; emic categories may stay opaque. |
| Participant as observer | Researcher takes a participant role (volunteer, intern, trainee) and observes from inside it. Known to be a researcher. | Insider perspective on the role taken; embodied knowledge of what the role requires. | Other roles in the setting; risk of “going native” and losing the analyst's distance. |
| Complete participant | Researcher's identity as researcher is hidden; they are seen as a member of the setting. Now ethically restricted in most public-health research. | The fullest insider view; what members say to each other when they don't think they are being studied. | Cannot probe analytically; serious ethical problems with informed consent. |
| Nonparticipation | Researcher observes only documentary or archival traces of the setting, the limit case bordering on indirect observation. | Patterns at scale; longitudinal change. | Anything that requires being there. |
The position you choose for a given study is a design decision with consequences. For most contemporary public-health work, the middle two positions, observer as participant or participant as observer, are recommended because they balance access against ethical clarity (Madden, 2017). The complete-participant position has a long history in mid-twentieth-century sociology (Festinger's When Prophecy Fails, Humphreys' Tearoom Trade) but is now restricted by institutional research-ethics boards in most jurisdictions.
The choice of position determines what an observer can see. Whatever the position, Spradley held that the central skill of participant observation is explicit awareness, the subject of the next subsection.
3.2 Building Explicit Awareness
Spradley argued that the central skill of participant observation is what he called explicit awareness: the trained capacity to notice things that ordinary perception filters out. The skill is the qualitative equivalent of learning to read an ECG: the lines on the paper are the same, but the trained eye sees what the untrained eye misses. Spradley described several specific techniques.
The everyday-object exercise. Spend ten minutes describing in writing the contents of a clinic waiting room you have been in dozens of times. Most students struggle for the first three minutes, then find a rhythm. The exercise reveals how little of the familiar setting was ever consciously registered. The training value is that subsequent visits to the setting will register more.
The naive-observer exercise. Pretend you have just arrived from a culture that has never seen a Canadian hospital before. Describe what is happening as if you do not know what a stethoscope is, what a hospital gown is for, why people are wearing masks. The discipline is to make the familiar strange. This is the canonical anthropological move and it is the engine of ethnographic insight.
The widening-attention exercise. Pick a single feature of a setting, say who speaks first at the start of each meeting, and track it across multiple sessions. Then add a second feature: who interrupts whom. Then a third: where people position themselves around the table. Layered attention builds capacity that single-feature attention cannot.
The body-knowledge exercise. Notice the bodily sensations the setting produces in you. The smell. The temperature. The acoustics. The pace. Your own discomfort or ease. Ethnographic insight is partly embodied; what your body notices is data, even if it never makes it into the final report verbatim.
The box below carries the discipline of explicit awareness over to the reading of interview transcripts, the form of data used in this course.
Why this matters for transcript-based work
Work with the course dataset involves no participant observation, because the data are pre-collected interview transcripts. But the discipline of explicit awareness applies to reading transcripts as well. The next time you read P01 (Maya), notice what she says about the SkyTrain at 9 p.m. on Sunday as an ethnographer would notice an everyday detail: what is the SkyTrain at that hour, what does it sound like, who is on it, what does the silence among passengers feel like? The trained ethnographic eye is the same eye you should be reading transcripts with. Without it, all you can see is the words; with it, you can see the world the words point to.
Participant observation and the training of explicit awareness concern the open-ended watching of a whole setting. The next subsection turns to continuous monitoring, which narrows attention to a specific behavior stream and records it systematically.
3.3 Continuous Monitoring
Continuous monitoring is the structured cousin of participant observation. The observer watches a specific behavior stream, not the whole setting, and records it systematically. Examples in public health include hand-hygiene observation in hospitals (the World Health Organization's “5 Moments” observation protocol), clinical-encounter observation (how much of a 15-minute primary-care visit is spent on what), classroom observation of school-meal consumption, and harm-reduction street-outreach observation of who uses what services in what order.
Continuous monitoring trades the breadth of participant observation for the precision of structured measurement. The observer typically uses a coding sheet, recording each event of interest with a timestamp. The data are dense and tabular. They can be analysed both qualitatively (what is the texture of the behavior? what coding categories needed to be added in the field?) and quantitatively (how often does each behavior occur per unit time? are there temporal patterns?).
Continuous monitoring outperforms self-report whenever the behavior is one people would mis-remember, under-report, or socially-desirably-report. Hand hygiene is the canonical example. On general medicine wards at two U.S. hospitals, nurses reported performing hand hygiene in 95 percent of the situations that call for it, while observers recorded it at 46 percent of room entries and exits; physicians' self-reports (91 percent) were closer to their observed rate (83 percent) (Kulkarni et al., 2026). The discrepancy is not because clinicians lie; it is because behavior at that frequency is invisible to self-report.
Continuous monitoring follows one stream of behavior in full. The next subsection describes spot observation, which samples many people at brief instants to estimate how a population spends its time.
3.4 Spot Observation and Time-Allocation Studies
Spot observation is the sampling-based cousin of continuous monitoring. Instead of watching a stream of behavior continuously, the observer makes brief observations at randomly sampled times, recording who is doing what in that instant. Aggregated across many such instants, the data yield a time-allocation profile: what proportion of a population's time, on average, is spent on each activity.
The classical use is in cultural-anthropology economic research, asking how subsistence-farming households allocate labour between fields, livestock, household, and leisure, but the method translates directly to public-health questions. Time-allocation studies have been used to characterise informal caregiving in aging populations, sedentary behavior in office workers, and child supervision in low-income households. The method's strength is that it does not depend on participants' ability to recall how they spent their time (which is notoriously inaccurate). Its weakness is that brief snapshots cannot capture the meaning of activities and may miss rare but important events.
Spot observation samples many people at brief instants rather than following one stream of behavior. With the three sub-types of direct observation described, the final subsection turns to the reactivity problem they share and to the conditions under which direct methods are the right first choice.
3.5 When to Use Direct Methods
All three sub-types of direct observation face a version of the measurement-bias problem. The box below describes this reactivity problem, the Hawthorne effect, and the steps recommended for addressing it.
The reactivity problem
Direct observation has its own version of the measurement-bias problem: the Hawthorne effect. When people know they are being watched, they often change their behavior, sometimes briefly (the first few minutes of an observation), sometimes for the duration. The recommendation is to acknowledge this explicitly, to extend observation periods so the effect attenuates, and to triangulate with indirect data (the dispenser log in Section 1.2 is a good case: a behavior trace revealed what observation had distorted).
Direct observation is the right first choice when you need to see behavior in context, when you suspect that what people do differs from what they say they do, when the setting itself is part of the object of study, when the behavior is too frequent or too embodied to be accurately recalled, or when you need to characterise an ordinary practice that participants take so for granted they cannot describe it. It is the wrong choice when the population is dispersed, when the behavior is private, when the budget for fieldwork is small, or when the question is fundamentally about meaning rather than action.
For loneliness research, direct observation would be a powerful complement to interviews if the question is, for example, “What does loneliness look like in residential settings?” You could spend weeks in a long-term-care facility watching the morning routine, mealtime seating patterns, who walks with whom on the corridor, when residents return to their rooms. None of this would replace interviews like Helen's, but it would let you see the loneliness Helen could only describe to you in fragments.
This section has described the three sub-types of direct observation, the training of explicit awareness, the reactivity problem, and the conditions under which direct methods suit a study. The key takeaways, reflection and knowledge check that follow consolidate these points before Section 3 turns to elicitation.
Key Takeaways
- 3.1 Direct observation has three sub-types: participant observation (prolonged, immersive engagement), continuous monitoring (structured watching of one behavior stream), and spot observation (sampled snapshots of who is doing what, when). The Spradley spectrum, this lesson's combination of Gold's four field roles and Spradley's degrees of participation, names the positions an observer can take, from the complete observer, who sees behavior unaffected but cannot ask why, to the complete participant, who gains the fullest insider view at a serious ethical cost, with nonparticipation as a limiting case. The middle positions, observer as participant and participant as observer, are recommended for most public-health work because they balance access against ethical clarity. The position is a design decision, and a defensible study names what its position gives up.
- 3.2 Spradley held that the central skill of participant observation is explicit awareness, the trained capacity to notice what ordinary perception filters out. It is built through exercises: describing a familiar setting in writing, observing as a naive outsider to make the familiar strange, layering attention on one feature at a time, and noticing what the setting does to your own body. The same trained eye applies to reading transcripts, where it lets you see the world the words point to, such as Maya's SkyTrain at 9 p.m. on a Sunday.
- 3.3 Continuous monitoring follows a specific behavior stream, such as hand hygiene or a primary-care visit, and records each event with a timestamp on a coding sheet, producing dense data that can be analysed qualitatively and quantitatively. It outperforms self-report when the behavior is one people would mis-remember, under-report, or report in a socially desirable way: in one study nurses reported hand hygiene in 95 percent of the situations that call for it, while observers recorded it at 46 percent. The gap reflects behavior too frequent to be visible to recall, not lying. Asking remains the better tool when the question is about meaning rather than action.
- 3.4 Spot observation makes brief observations at randomly sampled times, recording who is doing what at each instant, and aggregates them into a time-allocation profile of the share of time a population spends on each activity. A basic study therefore needs randomly sampled observation times, a record of activity at each one, and enough instants to aggregate. Its strength is independence from participants' notoriously inaccurate recall; its weakness is that snapshots cannot capture the meaning of activities and may miss rare but important events.
- 3.5 When people know they are being watched they often change their behavior, the Hawthorne effect, so observers acknowledge it, extend observation periods until it attenuates, and triangulate with indirect data. Direct observation is the right first choice when behavior must be seen in context, when doing and saying diverge, when the setting is part of the object of study, or when a practice is too frequent, embodied, or taken for granted to be recalled. It is the wrong choice when the population is dispersed, the behavior private, the budget small, or the question about meaning rather than action. For loneliness, weeks in a long-term-care facility would let you see what an interview like Helen's can describe only in fragments.
Reflection
If you were designing a participant-observation study of loneliness in a Vancouver senior-housing complex, which of Spradley's five positions would you take, and why? What would that position let you see, and what would you give up?
Minimum 20 characters required.
Question 1: Spradley's participant-observation spectrum runs from:
Question 2: Continuous monitoring of hand hygiene typically finds rates much lower than self-report. This is best characterized as:
Question 3: Time-allocation studies use which sampling strategy?
Elicitation: Interviews, Focus Groups & Cultural-Domain Methods
Introduction and Overview
The third and largest family is elicitation: methods in which the researcher actively asks participants to produce data. This family deserves close attention because it dominates contemporary qualitative health research and because the design decisions are non-obvious. The family contains several quite different methods: unstructured interviews, semi-structured interviews, structured interviews, focus groups, free lists, pile sorts, triads, ranking tasks, and open-ended survey items. Each has a different relationship between researcher control and participant freedom; each is good for different kinds of questions.
The course dataset was elicited using semi-structured interviewing, so this section dwells on that method longest. The other methods matter too, because they appear in the literature and because some of them (free listing in particular) reappear later in the course as the basis of cultural-domain analysis.
Learning Objectives
- 4.1 Explain the uses, strengths, and costs of the unstructured interview, and describe how interviewing for interpretative phenomenological analysis differs from the semi-structured format of the course dataset.
- 4.2 Identify the design features of a good semi-structured interview guide, using this course's loneliness guide as a worked example.
- 4.3 Describe the role of probes: what they are, when to use them, and what makes a probe work or fail.
- 4.4 Explain how an interview guide is piloted and how revisions are recorded.
- 4.5 Distinguish structured interviews from semi-structured and unstructured ones on the dimensions of researcher control, participant freedom, and analytic affordance, and explain when a structured phase is useful.
- 5.1 Explain when focus groups outperform individual interviews and why their data are produced by group interaction.
- 5.2 Identify when focus groups are the wrong choice, and weigh focus groups against individual interviews for a topic such as loneliness.
- 6.1 Describe the cultural-domain elicitation methods that a later module will return to, and what they can and cannot reveal.
- 6.2 Recognise the place of open-ended survey items in the elicitation landscape and how to design them deliberately.
The background box below defines the main interview formats and the focus group and points to the HSCI 207 lesson that teaches them, so that the subsections below can concentrate on what a graduate study adds.
Background: Interview types and focus groups
Interviews differ in how much the researcher structures them. In a structured interview, every participant answers the same questions in the same order, with little or no follow-up, so the interview works much like a spoken questionnaire. In a semi-structured interview, the interviewer works from a guide of open questions and probes but follows the participant's lead, changing the order and asking follow-up questions as the conversation develops. In an unstructured interview, the interviewer has a topic in mind but no fixed questions, and the participant largely sets the agenda. An in-depth interview is a long one-to-one interview, usually semi-structured or unstructured, that explores a participant's experience in detail. A focus group is a moderated discussion with a small group, usually five to ten participants (Krueger and Casey suggest five to eight), in which participants respond to one another as well as to the moderator; the moderator keeps the discussion on topic, draws in quieter members, contains dominant ones, and reminds the group that confidentiality among participants cannot be guaranteed. HSCI 207 Lesson 10, Sections 1.2 and 1.3 (Collecting Qualitative Data) teach these formats, and Section 3.3 of the same lesson covers moderating a focus group; both are optional reading. This section concentrates on what a graduate study adds: interviewing for interpretative phenomenological analysis, the semi-structured format of the course dataset, probing and piloting, the structured phase of a grounded-theory study, and the analytic case for and against focus groups.
The subsections that follow begin with the unstructured interview, treat the semi-structured format of the course dataset together with its probes and piloting, then turn to the structured interview, focus groups, cultural-domain elicitation methods and open-ended survey items.
4.1 Unstructured Interviews
Unstructured interviewing lies at the participant-led end of the range of interview formats. This subsection describes its uses, strengths and costs, and then the interview style used for interpretative phenomenological analysis, which lies close to the unstructured end. The flip cards below summarise four of the formats discussed in this section: the unstructured interview, the semi-structured interview, the focus group and the structured interview.
The unstructured interview, defined in the background box above, is foundational in cultural anthropology (Spradley's The Ethnographic Interview, 1979, is the classic methodological treatment) and is used heavily in long-term fieldwork where the same researcher interviews the same participants many times over months or years.
The strength of unstructured interviewing is that the participant sets the agenda. They name the topics they think matter; they choose the words; they impose their categories before the researcher imposes theirs. This is invaluable when you are working in an unfamiliar cultural domain or with a population whose ways of talking about the phenomenon are themselves the object of study. A researcher trying to understand how Vietnamese-Canadian elders in East Vancouver talk about buon ba (the rough translation might be “loneliness”) will learn more from unstructured conversation than from a guide constructed in English-language categories. The cost is comparability: if every interview follows different threads, claims about patterns across cases rest on uncertain ground, so most public-health studies use unstructured interviewing only in an early scoping phase.
Interviewing for interpretative phenomenological analysis. Interpretative phenomenological analysis (IPA; Smith, Flowers, & Larkin, 2022) uses a distinctive interview style, close to the unstructured end of the range. The sample is small and homogeneous, often between three and ten people who share a particular experience, so that the analysis can attend to each case in detail before looking across cases. Each participant usually takes part in one long interview that is only minimally structured: a short schedule of open questions serves as a loose guide, and the interviewer follows the participant's account of a specific experience wherever it leads. Probes ask for meaning and concrete detail, such as “What was that like for you?” or “What did that mean to you at the time?”, and avoid steering the participant toward the researcher's categories. The contrast with the course dataset is instructive. The loneliness interviews (see 4.2 Semi-Structured Interviews: The Format of the Course Dataset) used a semi-structured guide across twenty deliberately varied participants, a design that suits comparison across cases; an IPA study of loneliness would interview a handful of people who share one situation, such as recent widowhood, about what that experience has been like for them.
Unstructured and IPA interviews give the participant wide control over the account. The next subsection turns to the semi-structured interview, the format used to produce the course dataset, which suits comparison across cases.
4.2 Semi-Structured Interviews: The Format of the Course Dataset
Semi-structured interviewing is the format used to produce the 20 transcripts of the course dataset, and this subsection treats it at the greatest length, covering the interview guide, the interviewer's positionality and the course's own guide, before the next two subsections turn to probing and piloting. Activity 4.1 sets the practical task for the subsection, sketching a 30-minute semi-structured guide for a topic the reader might study, and the worked-example box later in the subsection serves as its model answer.
Sketch a 30-minute semi-structured guide for a topic you might study. Include:
- An opening (warm-up, builds rapport, easy to answer): 3-5 min
- 2-4 core topic areas, each with a primary question plus 2-3 probe prompts ('Can you tell me more about that?', 'What was that like for you?'): 20 min total
- A closing ('Is there anything important I haven't asked you?'): 5 min
- Demographic items at the end (not the beginning, since you want them invested before you ask)
A good interview guide is short. Topics, not scripts. The questions you don’t ask are as important as the ones you do.
Model answer: the course’s loneliness guide, described in the worked-example box later in this section, shows the same design. It has open-ended main questions, two or three probes under each, and a deliberate order of domains. Compare your sketch with it.
Semi-structured interviewing is the workhorse of contemporary qualitative health research (Brinkmann, 2007), and it is the method used to produce the 20 transcripts of the course dataset. The method sits between unstructured and structured: it uses a written guide that covers the topics the researcher wants to address, but treats the guide as a checklist rather than a script. The interviewer follows the participant's lead within each topic, asks probes responsively, skips items the participant has already covered, and reorders items as needed.
The defining feature is the guide-but-not-script approach (Brinkmann & Kvale, 2018). The interviewer should be able to conduct the interview without looking at the page; the guide is internalised, not read. The reason is that reading questions aloud creates a school-test register that suppresses the discursive material the method is meant to elicit.
A semi-structured interview is also a relationship, and who the interviewer appears to be shapes what a participant will say. The interviewer's positionality (age, gender, perceived ethnicity, accent, and evident insider or outsider status) affects rapport, what feels safe to disclose, and which answers seem expected. This is why careful interviewers practise reflexivity, keeping track of how their own presence may be steering the account and noting it alongside the data rather than treating the interview as a neutral extraction of fact.
The box below works through the course's own loneliness guide, its six conceptual domains and three of its design features, as an example of guide design.
This course's loneliness guide as a worked example
Open the interview guide (Interview_Guide_Loneliness.md in HSCI_841_loneliness_data.zip, which also holds the 20 interview transcripts, participant tables and codebooks) and read along with this discussion. The guide is organised around six conceptual domains (defining and recognising loneliness; triggers and patterns; meaning and identity; responses and coping; social world; systems and policy). Each domain has a small number of main questions, each with two or three italicised probes. Notice three design features:
- Open-ended main questions. “When I say the word loneliness, what comes to mind for you?” (Q2) invites the participant to bring their own framing first.
- Targeted probes underneath. “Is loneliness the word you would use yourself, or would you use something different?” is the probe that produced Amira's response “In Arabic the word is wahda.” The probe opened a door the main question alone would have left closed.
- Domain order. The guide moves from definition to triggers to meaning to coping to social world to policy. The order is deliberate: definition first (before researcher framings contaminate it), policy last (when the participant is warmed up enough to step back from their own experience).
A guide sets the topics; what the interviewer does within each topic depends on probes, the subject of the next subsection.
4.3 The Art of the Probe
The probe is the interviewer's main means of following the participant's lead within each topic of the guide. The background box below defines the probe and its main types and explains why a leading probe is to be avoided.
Background: Probe types
A probe is a responsive follow-up that invites a participant to expand, clarify, or give an example; it responds to what was just said, while the guide's questions are planned in advance. The main types are the silent probe (waiting after an answer), the echo probe (repeating a fragment of what was said with a rising tone), the uh-huh probe (minimal acknowledgement that holds the space open), the tell-me-more probe (an explicit invitation to expand), and the long-question probe (a fuller question that gives a reticent participant more to respond to). A leading probe, which suggests the answer the interviewer expects, is to be avoided because it contaminates the data. HSCI 207 Lesson 10, Section 2.5 (Collecting Qualitative Data) teaches probes and prompts with examples and is optional reading.
Probes that worked, probes that didn't: lessons from the corpus. Read alongside the transcripts, you can see the probes working. With Maya, the silent probe after “the SkyTrain at 9 p.m. on a Sunday” produced the elaboration “everyone's just on their phones, and nobody acknowledges that anybody else exists.” With Helen, the echo probe “Fading...?” produced the elaboration about three days without speaking and the voice sounding strange. With Amira, the uh-huh probe held the space open during the difficult disclosure about the anniversary of when her home was destroyed.
You can also see, in places, probes that did not produce expansion. Amira's response to a probe about Canadian friends, “To explain everything is to live through it again,” signals that the probe touched material the participant did not want to elaborate on. The interviewer correctly did not push. The discipline of qualitative interviewing is partly the discipline of not probing: recognising when to step back, when silence is what the participant needs, and when the probe would extract material the participant would later regret giving.
The excerpt below reproduces Maya's answer to the guide's question on what the word loneliness brings to mind (Q2), and the paragraph after it reads the excerpt for the conventions of its transcription.
Notice in the excerpt that the bracketed [pause] is a transcription convention, an artefact of the analytic choice made during transcription, not of the spoken utterance itself. We return to that in a later section. Notice also that the participant's response is full of filled pauses (“like,” “um”) and false starts (“nobody, nobody”). The fact that these survived into the transcript means the transcription used a verbatim convention, which keeps every utterance; an intelligent-verbatim transcript would have removed them. The fact that they did not survive into a polished journal-article quote would mean the analyst then cleaned them for publication. Both are legitimate choices at different stages.
Probes are refined in practice, and the first practice usually comes before the main interviews, when the guide is piloted.
4.4 Piloting the Guide
Before the main interviews, a guide is usually piloted with one or two people who resemble the intended participants, or with colleagues when the population is small or hard to reach. The pilot shows which questions confuse, which produce short answers, which topics run long, and which probes open an account and which close it. The researcher then revises the wording, order, and probes, and keeps a dated record of each change and the reason for it, so that the methods section can report how the guide evolved and an ethics board can see whether a change altered the risk to participants. Whether pilot interviews are analyzed with the main data is decided in advance and reported. HSCI 207 Lesson 10, Section 2.7 (Collecting Qualitative Data) works through piloting an interview guide and is optional reading.
A piloted guide, used flexibly with responsive probes, is the core of semi-structured interviewing. The next subsection moves to the structured end of the range.
4.5 Structured Interviews
Structured interviewing, defined in the background box at the start of this section, blurs into survey research at the structured end and is most useful when cases must be compared on standardised items.
Structured interviewing is legitimate and underused in qualitative methods. A grounded-theory study that has matured to the point of needing to test specific propositions can benefit from a structured-interview phase, even if the early phases used semi-structured or unstructured methods. The constant-comparative method is well-served by structured comparison at a late stage.
The interview formats discussed so far collect data from individual participants. The next subsection turns to focus groups, in which the data are produced by the interaction among participants.
5.1 Focus Groups: When They Outperform Individual Interviews
Focus groups (or group interviews) are a qualitatively distinct elicitation method, well beyond “an interview with more people.” The defining feature is that the data are produced by the group interaction rather than by individuals serially (Kitzinger, 1994). Participants respond to each other; they agree, disagree, build on, qualify, and contest one another's framings. The data are dialogical. Richard Krueger's Focus Groups: A Practical Guide for Applied Research (5th edition, 2015) is the canonical methodological treatment; Kitzinger (1995) is the standard short-form introduction in the medical literature.
Focus groups (Krueger & Casey, 2015) are the right choice in five situations:
- When you want to surface the range of views in a community, where the group format brings out positions individuals might not articulate alone.
- When the topic is best discussed collectively: community priorities, policy preferences, evaluation of a programme, perceptions of a public-health campaign.
- When you want to observe how categories are negotiated in real time: what becomes consensus, what gets contested, what is unspeakable.
- When you have limited time and the question is breadth more than depth, six focus groups of eight people each produce a dataset of 48 voices in roughly the same time as eight individual interviews.
- When the population is reluctant to speak alone: some communities prefer collective speech; some topics are easier to discuss with peers present than with a researcher alone.
The same group dynamics that make focus groups productive also limit them, as the next subsection explains.
5.2 When Focus Groups Are the Wrong Choice
Focus groups are the wrong choice when the topic is too sensitive for collective disclosure (sexual behavior, intimate partner violence, suicidality, stigmatised illness), when the participants are in unequal power relationships with each other (boss and subordinate, parent and adolescent), when the population is too small to recruit groups, or when the research question requires the kind of sustained individual narrative that group dynamics interrupt.
For loneliness research specifically, the choice is non-obvious. Individual interviews allow the participant to disclose the intimate and stigmatising dimensions of loneliness (Maya's comment “loneliness feels like admitting I'm failing at being 22” would be hard to say in a group). Focus groups allow you to see how loneliness is talked about collectively: what the public vocabulary is, what is performable and what is not, what stays unsayable. A study using both methods would learn things neither alone could reveal.
Interviews and focus groups both produce extended talk. The next subsection previews a different family of elicitation methods, designed to surface how members of a community categorise, order and relate the concepts within a cultural domain.
6.1 Cultural-Domain Elicitation Methods (Preview)
This subsection introduces a family of elicitation methods specifically designed to surface the structure of a cultural domain: how members of a community categorise, order, and relate concepts within a topic area. These methods will be the focus of a later module, where you will analyse them with networking and clustering techniques. For now, learn the names and what they do.
- Free listing. The simplest cultural-domain method: ask a sample of people to list, say, “all the kinds of loneliness you can think of.” The frequency with which each item appears, and the average position in the list, together index its cultural salience.
- Pile sorts. Hand participants a set of cards (each with a concept or item) and ask them to sort the cards into piles of things that go together. Aggregated across participants, the sorts yield a similarity matrix that can be analysed with multidimensional scaling or hierarchical clustering (statistical techniques that turn those groupings into a map of what people treat as similar to what).
- Triad tests. Show participants three items at a time and ask which two are most similar (or which one is most different). Aggregated across triads, the data yield a similarity matrix similar to pile-sort output.
- Ranking tasks. Ask participants to order a set of items along a specified dimension (e.g., from least to most lonely, from most to least helpful).
These methods are powerful for studying the structure of a cultural domain, what its elements are and how they relate, in a way that open-ended interviewing cannot match. They are also fast: a free list can be elicited in two minutes per participant, allowing samples of several hundred people in a single study. The cost is depth: cultural-domain methods are structurally rich but phenomenologically shallow. You learn how loneliness is categorised, not what it feels like.
Cultural-domain methods gain structure and scale at the cost of depth. The final subsection turns to the last elicitation form, the open-ended survey item.
6.2 Open-Ended Survey Items
The final elicitation form is the humble open-ended survey item: the “please describe in your own words” box at the back of a questionnaire. These are the most underused qualitative-data source in public health, partly because the items are often included for completeness rather than analysed, partly because the data are typically thin (a sentence or two per response) and easy to dismiss.
The corrective is to design open-ended items deliberately, position them where respondents have time to answer (early in the survey, not buried at the end), provide enough space to invite expansion, and budget the analytic time the responses deserve. A 2,000-respondent survey with even a 40 percent response rate to one open-ended item gives you 800 qualitative responses, a respectable corpus that can be theme-analysed, content-analysed, or topic-modelled (a later module).
This section has described the elicitation methods, from unstructured, semi-structured and structured interviews to focus groups, cultural-domain methods and open-ended survey items, with the semi-structured format of the course dataset treated at greatest length. The key takeaways, reflection and knowledge check that follow consolidate these points before Section 4 turns to transcription and field notes.
Key Takeaways
- 4.1 The unstructured interview lets the participant set the agenda, naming the topics, words, and categories before the researcher imposes any, which is invaluable in an unfamiliar cultural domain such as Vietnamese-Canadian elders' talk about buon ba. Its cost is comparability across cases, so most public-health studies use it only in an early scoping phase. Interviewing for interpretative phenomenological analysis lies close to the unstructured end: a small, homogeneous sample, one long and minimally structured interview each, and probes that ask for meaning and concrete detail. The course dataset differs because its twenty varied participants and semi-structured guide suit comparison across cases.
- 4.2 The semi-structured interview, the format of the course dataset, uses a written guide as a checklist rather than a script: the interviewer internalises it, follows the participant's lead within each topic, skips what has been covered, and reorders as needed. The course's loneliness guide shows three design features: open-ended main questions that let the participant frame the topic first, targeted probes beneath each, and a deliberate order of domains running from definition to policy. Because the interview is also a relationship, the interviewer's positionality shapes what is said, and reflexive interviewers note its influence alongside the data.
- 4.3 A probe is a responsive follow-up that invites a participant to expand, clarify, or give an example; the main types are the silent, echo, uh-huh, tell-me-more, and long-question probes, while the leading probe is avoided because it contaminates the data. In the corpus, a silent probe drew out Maya's account of the SkyTrain, and the echo probe “Fading...?” drew out Helen's three days without speaking. When a probe touches material the participant does not want to elaborate on, as with Amira, the interviewer should not push, so part of the discipline is knowing when not to probe.
- 4.4 A guide is piloted with one or two people who resemble the intended participants, or with colleagues when the population is small or hard to reach, to find which questions confuse, which produce short answers, and which probes open or close an account. The researcher revises wording, order, and probes and keeps a dated record of each change and its reason, so that the methods section can report how the guide evolved and an ethics board can see whether the risk to participants changed. Whether pilot interviews join the main data is decided in advance and reported.
- 4.5 Interview formats differ in researcher control and participant freedom: the unstructured interview gives the participant the most freedom, the semi-structured interview balances a guide with responsive follow-up, and the structured interview fixes the same questions in the same order, so that it blurs into survey research. Standardisation is the structured interview's analytic affordance, since it supports comparison of cases on the same items. It is legitimate and underused, for example in the late phase of a grounded-theory study that needs to test specific propositions.
- 5.1 A focus group is not an interview with more people: its data are produced by the interaction among participants, who agree, disagree, build on, and contest one another's framings, so the data are dialogical. Focus groups outperform individual interviews when you want the range of views in a community, when the topic is best discussed collectively, when you want to watch categories being negotiated, when breadth matters more than depth and time is short, and when a population prefers collective speech.
- 5.2 Focus groups are the wrong choice for topics too sensitive for collective disclosure, such as sexual behavior, intimate partner violence, or suicidality, for participants in unequal power relationships, for populations too small to recruit groups, and for questions that need sustained individual narrative. For loneliness the choice is non-obvious: individual interviews allow intimate and stigmatising disclosures, while focus groups show how loneliness is talked about collectively. A study using both methods would learn things neither could reveal alone.
- 6.1 Free listing, pile sorts, triad tests, and ranking tasks surface the structure of a cultural domain, that is, how members of a community categorise, order, and relate its elements, and a later module analyses them with networking and clustering techniques. Free lists index cultural salience by how often and how early an item is named, while pile sorts and triads yield similarity matrices. These methods are fast and scale to hundreds of participants, but they are structurally rich and phenomenologically shallow: you learn how loneliness is categorised, not what it feels like.
- 6.2 The open-ended survey item is the most underused qualitative source in public health, often included for completeness and then left unanalysed because the responses are thin. Designed deliberately, placed early in the survey, given enough space, and budgeted for analysis, it can yield a sizeable corpus: a 2,000-respondent survey with a 40 percent response rate to one item gives 800 responses that can be theme-analysed, content-analysed, or topic-modelled.
Reflection
You are designing the next phase of a loneliness study, building on the existing 20 semi-structured interviews. Pick ONE elicitation method you have NOT yet used (focus groups, free listing, pile sorts, structured interviews, or an open-ended survey item) and explain what specific question it would let you answer that the existing interview corpus cannot.
Minimum 20 characters required.
Question 1: This course's loneliness interview guide is best characterised as:
Question 2: An echo probe is:
Question 3: Focus groups are typically the WRONG method choice when:
Transcription, Field Notes & the HSCI 841 Dataset
Introduction and Overview
You have now seen the three families of qualitative data collection: indirect, direct, and elicitation. The last section of this lesson turns to what happens after the data are collected: the work of transforming spoken speech into transcripts and field experience into field notes. Both are treated here as analytic acts in their own right, not as neutral preparation. The decisions you make at this stage shape what later analysis can see, and, for a pre-collected dataset such as the course's loneliness interviews, the decisions someone else made before the analyst arrived have already shaped what is available.
Learning Objectives
- 7.1 Recognise transcription as an analytic act, not a neutral technical step, and match the level of transcription to the analytic intent.
- 7.2 Distinguish four transcription conventions, namely Jefferson notation (preview for a later module), verbatim, intelligent verbatim, and clean verbatim, by what each keeps and what each suits.
- 7.3 Identify which convention the course dataset uses, and which analyses its transcripts can and cannot support.
- 8.1 Write disciplined field notes using the Emerson, Fretz & Shaw progression: jottings → expanded notes → analytic memos.
- 8.2 Keep an analytic-memo file that separates description from analysis when working with pre-collected transcripts.
- 9.1 Explain why familiarisation with the data comes before coding, and how a familiarisation log records it.
The section first treats transcription as an analytic act, compares four transcription conventions and places the course dataset among them, then describes the three stages of field-note writing and the use of analytic memos with transcripts, and closes with familiarisation before coding.
7.1 Transcription as Analytic Act
Transcription is the conversion of recorded speech into written text. The conversion seems, on first encounter, mechanical: type what you hear. In practice, transcription requires hundreds of small interpretive decisions per transcript. Should “um” be transcribed? What about “you know” and “like” when they are filler? Should false starts be preserved or smoothed? Should the pause between words be timed? Should laughter be marked, and if so, how? Should overlapping speech be shown with brackets? Should non-verbal sounds (a sigh, a cough, a chair scraping) be annotated?
Each of these decisions shapes what later analysis can see. A transcript that omits filled pauses and false starts cannot support conversation analysis, because the analytic categories conversation analysts work with (a later module) are built on exactly those features. A transcript that omits laughter cannot support narrative-affective analysis, because the laugh-marker is often the most analytically interesting moment in a turn. A transcript that strips out vocalised pauses turns the speech of an articulate participant and the speech of a halting one into the same kind of text, masking a phenomenon that may matter for what you are studying.
This calls for a methodological commitment: the level of transcription should be matched to the analytic intent. If you plan content analysis or thematic analysis, light transcription is sufficient. If you plan discourse or conversation analysis, you need much heavier transcription. If you do not yet know which method you will use, err on the side of more detail: you can simplify a detailed transcript later, but you cannot recover what was never written down.
Matching transcription to analytic intent requires a clear set of transcription levels. The next subsection sets out four conventions, from the most detailed to the most edited.
7.2 Four Transcription Conventions
Transcription conventions differ in how much of the spoken record they preserve. Table 4.3 compares four of them (Jefferson notation, verbatim, intelligent verbatim and clean verbatim) on what each keeps, what each removes or edits, and the kinds of analysis each suits.
Table 4.3. Four transcription conventions, with what each keeps, what each removes or edits, and the kinds of analysis each suits.
| Convention | What is kept | What is removed or edited | Suits |
|---|---|---|---|
| Jefferson notation (preview, a later module) |
The most detailed form of verbatim: every word and every audible feature, including pause durations to the tenth of a second, overlaps marked with brackets, in-breaths and out-breaths, intonation contours, stress, latching, and laughter inside words. | Nothing; the convention's whole point is to preserve everything. | Conversation analysis; discourse analysis; the close study of interactional sequencing. |
| Verbatim | Every utterance: the participant's words with all filled pauses (“um,” “like”), false starts, repetitions, and laughter (marked as [laughs] or [laughter]). Long pauses are noted ([pause]) without timing. | Nothing that was said. Pause length, overlap, and intonation are not notated. | Thematic, content, narrative, and grounded-theory analysis, and any study whose analytic method is not yet settled. |
| Intelligent verbatim | The speaker's wording and meaning, including dialect and grammar as spoken. | Fillers, false starts, and repetitions, removed without changing the wording or meaning. | Thematic and content analysis in which hesitation and self-repair are not part of the analytic question. |
| Clean verbatim | The substantive content of what was said. | Fillers, false starts, and repetitions; grammar is also edited into readable prose. | Studies whose aim concerns what was said more than how it was said, such as some applied and policy research, oral history, and journalism. |
No convention is right for every study. The level follows the analytic aim: a study of what participants said can work from intelligent or clean verbatim, while a study of how they said it needs verbatim or Jefferson detail. Whatever the level, the research team writes down which sounds it keeps (fillers, laughter, pauses, overlaps) and applies that rule to every transcript, so that the transcripts are comparable and the methods section can report the convention.
The next subsection applies these conventions to the transcripts of the course dataset.
7.3 What Level Is This Course's Dataset?
The course dataset is transcribed at the verbatim level, without Jefferson detail. Read any transcript and you can see the conventions: filled pauses (“um,” “like”) are kept, false starts and self-corrections are visible (Maya's “nobody, nobody acknowledges” and “hollow isn't the right word, it's like an ache”), laughter is annotated [laughs], long pauses are marked [pause] or [long pause] without timing, and dashes mark mid-utterance self-interruptions. Because fillers and false starts are kept, the transcripts are verbatim; an intelligent-verbatim version would have removed them.
What the dataset does not have is Jefferson-level detail: pause durations are not numerically timed, overlapping speech is not bracketed in CA notation, intonation contours and stress are not marked, in-breaths are not transcribed. A later module, which works on discourse and conversation analysis, will return to this. For most analytic work in this course, the verbatim level is what the transcripts provide.
The box below sets out which analyses the verbatim transcripts support without modification and which would require re-transcription at Jefferson level.
What you can and cannot do with the existing transcripts
The verbatim level of the transcripts supports thematic analysis (a later module), grounded theory (a later module), content analysis (a later module), narrative analysis (a later module), analytic induction (a later module), and computational text analysis (a later module) without modification. It does not support fine-grained conversation analysis. If you wanted to do a CA-style examination of, say, how Helen's interview turns are constructed sequentially with the interviewer's, you would need to re-listen to the original recording (which the dataset does not provide) and re-transcribe at Jefferson level. We will work around this later in the course by transcribing a single short excerpt by hand.
Transcription converts recorded speech into written text. The next subsection turns to the parallel task for observational work, the writing of field notes.
8.1 Field Notes: Emerson, Fretz & Shaw
Robert Emerson, Rachel Fretz, and Linda Shaw's Writing Ethnographic Fieldnotes (Emerson, Fretz, & Shaw, 2011) is the standard methodological treatment of field-note writing in contemporary qualitative methodology. The Emerson/Fretz/Shaw progression organises field-note writing into three stages, each with a distinct purpose and discipline.
Stage 1: Jottings. Brief, in-the-moment notes scribbled in a small notebook (or, increasingly, in a phone notes app) during or immediately after observation. Jottings are not for the reader; they are for the writer's later self. They typically include a few key words per moment, enough to trigger memory when the writer sits down to expand them. The discipline is to capture the specific concrete particular: who said what, what they were doing, what time it was, what was on the wall. The temptation is to write analytic interpretations on the spot; the discipline is to resist this and write only what could be photographed if a camera had been present.
Stage 2: Expanded notes. Written as soon as possible after the observation (ideally the same day), expanded notes turn the jottings into full descriptive prose. The discipline is to write at length and in the present tense, recreating the scene so a reader could enter it. Expanded notes are typically several pages per hour of observation. They include direct quotations where possible, descriptive detail of setting, and observations of body language, sequence, and timing. They do not yet include analytic interpretation: the discipline is to keep description and analysis separated, because analysis dressed as observation gets remembered as observation and cannot be challenged later.
Stage 3: Analytic memos. Written separately from the expanded notes (often in a different document or notebook section), analytic memos record the researcher's evolving interpretation: what the field experience seems to mean, what patterns are emerging, what theoretical connections are forming, what hypotheses are taking shape. Memos are dated, indexed, and treated as part of the data corpus, and you cite your own memos in the eventual methods section because the audit trail of analytic development is part of what transparency owes the reader.
The third stage, the analytic memo, carries over directly to work with transcripts, as the next subsection explains.
8.2 Analytic Memos in Transcript-Based Work
Work with a pre-collected dataset involves no fieldwork in the traditional sense, but the discipline of separating description from analysis applies to reading transcripts. Analysts keep an analytic-memo file (a single .md or .docx document, with dated entries that are never deleted) throughout a project. Every analytic insight about the data, every “what if” thought, every emerging connection, every “Maya and Helen both...” observation, goes into the memo file. Lesson 6 returns to memos; by the time codebook development begins, the accumulated memos are its raw material.
Field notes and analytic memos keep description and interpretation apart. The final subsection turns to familiarisation, the close reading of the data that precedes coding.
9.1 Familiarisation Before Coding
Familiarisation is the first phase of most approaches to qualitative analysis, including thematic analysis (Lesson 5, Section 1.4). Most experienced qualitative researchers are explicit that familiarity with the data is the precondition of analysis. Transcripts that have only been skimmed cannot be coded well, and a theme is hard to recognise in a transcript read for the first time inside coding software.
A practical tool is a dataset familiarisation log: a structured table with one row per transcript, recording who the participant is, the main topics of the account, striking phrases and metaphors, moments where the transcription conventions matter, and first analytic hunches, followed by a short reflection on what the reading has taught the analyst about the data. For the course dataset, six to eight transcripts chosen to span age, gender, and life stage give a good first sense of the corpus. The transcripts are in HSCI_841_loneliness_data.zip (the 20 interview transcripts, participant tables and codebooks).
This section has treated transcription and field-note writing as analytic acts and has described familiarisation as the work that precedes coding. The key takeaways, reflection and knowledge check that follow consolidate these points.
Key Takeaways
- 7.1 Transcription looks mechanical but involves hundreds of interpretive decisions per transcript: whether to keep fillers, false starts, pauses, laughter, overlaps, and non-verbal sounds. Each decision shapes what later analysis can see; a transcript without filled pauses and false starts cannot support conversation analysis, and one without laughter loses moments that narrative-affective analysis depends on. The level of transcription should therefore match the analytic intent, and when the method is not yet settled, err toward more detail, since you can simplify later but cannot recover what was never written down.
- 7.2 The four conventions run from most to least detailed. Jefferson notation keeps every audible feature, including timed pauses, overlaps, and intonation, for conversation and discourse analysis; verbatim keeps every utterance with fillers, false starts, and laughter but does not time pauses; intelligent verbatim removes fillers, false starts, and repetitions without changing wording or meaning; and clean verbatim also edits grammar into readable prose, for studies of what was said rather than how. Whatever the level, the team writes down which sounds it keeps and applies that rule to every transcript.
- 7.3 The course dataset is transcribed at the verbatim level without Jefferson detail: fillers, false starts, laughter, and long pauses are kept, but pauses are not timed, overlaps are not bracketed in CA notation, and intonation is not marked. That level supports thematic analysis, grounded theory, content, narrative, and computational analysis, and analytic induction without modification, but not fine-grained conversation analysis, which would require re-transcribing the original recordings at Jefferson level.
- 8.1 Emerson, Fretz, and Shaw organise field-note writing into three stages. Jottings are brief notes made during or just after observation that capture concrete particulars, only what a camera could have recorded; expanded notes, written the same day, turn them into full descriptive prose without interpretation; and analytic memos, kept separately, record the researcher's evolving interpretation and are dated, indexed, and treated as part of the data corpus. The central discipline is to keep description and analysis apart, because analysis dressed as observation is later remembered as observation and cannot be challenged.
- 8.2 Work with pre-collected transcripts involves no fieldwork in the traditional sense, but the same separation of description from analysis applies. Analysts keep a single analytic-memo file of dated entries that are never deleted, recording every insight, “what if” thought, and emerging connection across participants. By the time codebook development begins, the accumulated memos are its raw material.
- 9.1 Familiarisation is the first phase of most approaches to qualitative analysis, because transcripts that have only been skimmed cannot be coded well. A dataset familiarisation log records it: a table with one row per transcript noting the participant, the main topics, striking phrases and metaphors, moments where the transcription conventions matter, and first hunches, followed by a short reflection on what the reading has taught. For the course dataset, six to eight transcripts chosen to span age, gender, and life stage give a good first sense of the corpus.
Reflection
Pick one transcript you have read so far and identify ONE feature of its transcription that shapes what later analysis can see. For example: a marked [laughs] annotation that opens a particular analytic possibility, a [pause] that did the work of unspoken disclosure, a false start preserved that revealed something the participant talked away from. What would have been lost if a clean-verbatim convention had been used?
Minimum 20 characters required.
Question 1: The course dataset uses which transcription convention?
Question 2: The Emerson/Fretz/Shaw field-note progression is:
Final Assessment
Bringing It All Together
This lesson has mapped the three families of qualitative data collection (indirect, direct, elicitation), located the course dataset within the semi-structured-interview tradition, treated transcription and field-note writing as analytic acts in their own right, and described familiarisation with the data before coding. With this lesson complete, you have the full design vocabulary the rest of the course draws on: Lesson 5 turns to coding, Lesson 6 to conceptual models, and later lessons to specialised analytic techniques. Familiarisation is the step that connects data collection to that analysis.
Key Takeaways from this lesson
- Three families of data collection: indirect observation (Webb's unobtrusive measures: traces, archives, secondary datasets), direct observation (participant observation, monitoring, time allocation), and elicitation (interviews, focus groups, cultural-domain methods).
- Indirect observation is non-reactive and longitudinal but cannot ask why. Behavior traces, archival records, and secondary qualitative analysis are powerful complements to elicitation.
- Direct observation (participant observation, continuous monitoring, and spot observation) requires the discipline of explicit awareness (Spradley). The five-position spectrum from complete observer to complete participant names what each position can and cannot see, and continuous monitoring and time-allocation studies outperform self-report for behavior too frequent or embodied to recall.
- Unstructured, semi-structured, and structured interviews trade participant freedom against comparability. Semi-structured interviewing is the workhorse of public-health qualitative research and the format used to produce the course dataset. The guide-but-not-script approach, the responsive use of probes, and the discipline of letting the participant set the agenda are the central skills. Interviewing for interpretative phenomenological analysis uses a small homogeneous sample and long, minimally structured interviews, and every interview guide benefits from piloting with a dated record of revisions.
- Focus groups produce qualitatively different data from individual interviews, dialogical rather than narrative, and demand trained moderator skills. They are wrong for sensitive topics and right for community priorities.
- Cultural-domain elicitation methods (free listing, pile sorts, triads, ranking tasks) surface the categorical structure of a topic area at scale. Preview for a later module. Open-ended survey items, designed deliberately and budgeted for analysis, are the most underused qualitative source in public health.
- Transcription is an analytic act. Jefferson notation, verbatim, intelligent verbatim, and clean verbatim each foreclose different later analyses, and the level follows the analytic aim. The course dataset is verbatim without Jefferson detail.
- Field notes follow the Emerson/Fretz/Shaw progression: jottings to expanded notes to analytic memos, with description and interpretation rigorously separated.
- Familiarisation comes before coding: a familiarisation log records 6–8 varied transcripts and what reading them has taught the analyst.
Core Concepts Reviewed
An earlier section: The three families of data collection; Webb's unobtrusive measures; accretion vs. erosion traces; archival data and the selection-bias problem; secondary qualitative analysis and its four hazards (constrained guide, lost context, consent, frame mismatch).
An earlier section: Spradley's five-position participant-observation spectrum; explicit awareness as a trained skill; continuous monitoring vs. time-allocation studies; the Hawthorne / reactivity problem in direct observation.
An earlier section: Unstructured, semi-structured, and structured interviewing; this course's loneliness guide as a worked example of semi-structured design; six probe types (silent, echo, uh-huh, tell-me-more, long-question, leading); when focus groups outperform individual interviews; the cultural-domain elicitation family (free listing, pile sorts, triads, ranking); open-ended survey items; interviewing for IPA; piloting an interview guide.
An earlier section: Transcription as analytic act; the four conventions (Jefferson notation, verbatim, intelligent verbatim, clean verbatim); the convention used in the course dataset; the Emerson/Fretz/Shaw field-note progression; familiarisation and the familiarisation log.
The final reflection asks you to integrate what you have learned about data collection with what you know about the course dataset, before the next lesson turns to coding.
Reflection
The course dataset arrived in a particular form, 20 verbatim semi-structured interview transcripts plus a guide, and that form shapes what an analysis of it can and cannot do. In one paragraph, name TWO research questions about loneliness that this dataset can defensibly answer, and ONE that it cannot. For the one it cannot, briefly describe the data-collection design that would be needed instead.
Minimum 30 characters required.
Final Knowledge Assessment
Question 1: The three families of qualitative data collection identified in this lesson are:
Question 2: Webb et al.'s 1966 term unobtrusive measures refers to:
Question 3: An erosion measure in the Webb tradition is:
Question 4: Spradley's five-position participant-observation spectrum runs from:
Question 5: Continuous monitoring of hand hygiene typically finds rates far below self-report. This is best read as:
Question 6: The course dataset was elicited with:
Question 7: A probe that repeats a short fragment of what the participant just said, with slight rising intonation, to invite expansion is called:
Question 8: Focus groups are typically the WRONG method choice for:
Question 9: Free listing, pile sorts, triad tests, and ranking tasks are examples of:
Question 10: Transcription is best characterised as:
Question 11: The transcription convention used in the course dataset is best described as:
Question 12: The Emerson, Fretz & Shaw progression for field-note writing is:
Question 13: A dataset familiarisation log is best described as:
Question 14: A defensible research question for the existing 20-transcript loneliness dataset is:
Glossary: Key Terms, People & Methodological Stances
📚 Reference page, available throughout the lesson
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.