# Lesson 3 — Sampling in Qualitative Research (v3 expanded)

*Companion-podcast transcript • Sarah & Kiffer*  
*~5,500 words • ~30 min audio*

---

**Sarah:** Welcome back to Office Hours. I'm Sarah.

**Kiffer:** And I'm Kiffer. Today is Lesson 3 of the qualitative methods arc. Sampling. And I want to flag upfront that this lesson does something that students often find disorienting. It asks them to unlearn the sample-size logic they spent the last three years internalizing.

**Sarah:** Right. Most of our listeners have a clean mental model of sampling already. There's a population, you draw a probability sample from it, you measure something, the sample mean is an unbiased estimator of the population mean, the confidence interval gets narrower as the sample gets larger. End of story.

**Kiffer:** And that model is correct. It's powerful. It's what you used in surveillance work, in screening evaluations, in cohort and case-control studies. The problem is that, as a description of how qualitative researchers actually sample, it is almost entirely wrong.

**Sarah:** Bernard, Wutich, and Ryan open chapter three of the textbook with a sharp claim. There are two kinds of samples in social-science research, and they were built to do different jobs.

**Kiffer:** Probability samples were built so you can estimate population-level magnitudes with calculable error. Nonprobability samples were built so you can characterize a phenomenon — identify its categories, understand its mechanisms, describe how people make sense of it. These are not different ways of doing the same job. They are different jobs.

**Sarah:** And the rest of the lesson is about taking that distinction seriously.

**Kiffer:** Right. Section by section. Two kinds of samples, the operational concept of saturation, the six nonprobability strategies, the difference between theoretical and purposive sampling, key-informant sampling, and what your methods section owes the reader when you defend your sample.

**Sarah:** Let's start with the operational difference. Probability sampling in one paragraph.

**Kiffer:** A probability sample is one in which every member of a defined population has a known, non-zero probability of being selected, and that probability is built into the design. Simple random sampling, stratified, cluster, multi-stage, probability-proportional-to-size. What makes them probability samples isn't that they involve a random number generator. It's that the selection probabilities are knowable. That knowability is what makes statistical inference to the population possible.

**Sarah:** And sample size for probability designs is determined by variance, precision, and effect size.

**Kiffer:** Right. The arithmetic is well-understood and lives next door to this course. The shape of the argument is what matters. More cases purchase more precision around the same quantity. Each additional case adds a diminishing amount of statistical information about the parameter you're estimating. Twenty is better than ten. Two hundred is better than twenty. Two thousand is better than two hundred. The only reason to stop is cost.

**Sarah:** Now nonprobability sampling.

**Kiffer:** A nonprobability sample is one in which the selection probabilities are not knowable and inference to a defined population is not the analytic goal. Convenience, purposive, quota, snowball, respondent-driven, theoretical, key-informant — all nonprobability designs. They differ from each other in important ways, but they share that defining feature. You cannot calculate the probability that any given member of a hypothetical population was selected for your study.

**Sarah:** And the temptation for someone trained on probability sampling is to read that as a deficiency.

**Kiffer:** Yeah, it's a big trap. The lesson is emphatic that nonprobability sampling was developed because some research questions are not population-magnitude questions. The questions qualitative researchers most commonly ask — what is loneliness, how does it show up, what configurations of loneliness exist, what are the mechanisms, what do people do about it — are not answered by knowing what proportion of British Columbians experience loneliness. They are answered by deliberately choosing cases that illuminate the phenomenon. The selection logic is curatorial rather than statistical.

**Sarah:** The textbook puts it as extensity versus intensity.

**Kiffer:** That's the phrasing. Probability samples are about extensity — the breadth of a phenomenon in a population. Nonprobability samples are about intensity — the depth, detail, and configuration of the phenomenon. The same loneliness can be studied through either lens. A probability sample of five thousand Canadians would tell you what percentage report frequent loneliness, how the prevalence varies by age and income, how the rate has changed over time. A nonprobability sample of twenty deliberately chosen interviews would tell you what the experience of loneliness is, what kinds of loneliness exist, what triggers it, how people interpret it, what they do about it. Each design answers a question the other cannot.

**Sarah:** One thing I want to underline. When you talk about extensity versus intensity, the temptation is to read it as if probability is the rigorous one and nonprobability is the cheap one.

**Kiffer:** Yeah, and that's the framing the textbook is at pains to disrupt. Each design is rigorous on its own terms. A poorly defended convenience sample is sloppy. A well-defended maximum-variation sample of twenty is rigorous. A poorly designed probability survey with two thousand respondents is sloppy — even at large N. Rigor is not a property of sample size. It's a property of fit between sampling logic and research question. And one of the unlearnings of this course is dropping the heuristic that bigger is automatically better.

**Sarah:** And it cuts the other way too. A large convenience sample is not more rigorous than a small purposive one.

**Kiffer:** Exactly. The methods literature actually has a useful term for what large convenience samples produce — they amplify whatever bias the convenience introduces. Two thousand undergraduates recruited from one introductory psych course will give you a very precise estimate of a very particular kind of person. The precision is real. The generalizability is fake. And no amount of N saves you.

**Sarah:** And specifically, our capstone dataset is the second kind.

**Kiffer:** Right. Twenty transcripts, varying deliberately. Including, for example, P fourteen — Kenji — a sixty-year-old man who came out after a long heterosexual marriage. Kenji is not statistically representative of British Columbian men in their sixties. He almost certainly isn't. But his configuration — late-life sexual-identity disclosure interacting with the loss of long-standing social ties — is a kind of loneliness the literature under-describes. His transcript gives you analytic purchase on that configuration. A randomly drawn sample of twenty BC men in their sixties would almost certainly miss him. A probability sample of five thousand would catch him as nought point zero six percent of a frequency table and would not produce eight thousand words of his account.

**Sarah:** That's the curatorial logic.

**Kiffer:** You're not trying to be representative of a population. You're trying to be representative of a phenomenon's variation. When the variation is the explicit goal, the textbook calls it maximum-variation sampling. And what you owe your reader isn't a calculation of selection probabilities. It's a defense of the variation you chose to capture and an honest accounting of the variation you did not.

**Sarah:** Okay, section two. Saturation. This is the operational concept that replaces the power calculation.

**Kiffer:** Glaser and Strauss introduced saturation in nineteen sixty-seven as a feature of theoretical sampling in grounded theory. It has since been generalized far beyond that origin to almost every kind of nonprobability design. The working definition is simple. A sample is saturated when collecting additional cases stops producing new information relevant to the analytic question.

**Sarah:** And the operationalization is where it gets complicated.

**Kiffer:** Yes. Three flavours of saturation get distinguished in the methodological literature, and you have to keep them straight, because mixing them up is the most common source of methods-section confusion in published qualitative health papers.

**Sarah:** Theoretical saturation first.

**Kiffer:** The original Glaser-Strauss usage. A sample is theoretically saturated when additional cases stop producing new categories, properties, or relationships in your developing theory. It's tightly tied to theoretical sampling — you sample iteratively, in response to the emerging analysis, until the theoretical structure stops changing. This is the most demanding kind of saturation and the hardest to demonstrate.

**Sarah:** Code saturation.

**Kiffer:** More recent, more operational. A sample is code-saturated when additional cases stop producing new codes in your codebook. Code saturation tends to arrive relatively quickly. The major themes show up in the first several transcripts. And this is what most empirical saturation studies actually measure.

**Sarah:** And meaning saturation.

**Kiffer:** The point at which additional cases stop deepening your understanding of the codes you already have. This is harder and slower. You might have identified a code like "loneliness as spatial" after three interviews, but reach a real understanding of the variations within that code — the specific spatial metaphors people deploy, the sense in which space is doing analytic work — only after fifteen or twenty.

**Sarah:** And there's empirical work behind these numbers.

**Kiffer:** There is. Three pieces of literature you should know. The oldest is the Romney-Weller-Batchelder four-to-six rule. Their work on cultural consensus theory in cognitive anthropology showed that when cultural agreement within a group is high, as few as four to six knowledgeable respondents are enough to recover the shared cultural model with high confidence. That sounds implausibly small to a quantitatively trained reader. But it's mathematically grounded for the specific case of cultural-consensus tasks — free lists, pile sorts, that family of methods.

**Sarah:** The second is Guest, Bunce, and Johnson, the twenty-oh-six paper.

**Kiffer:** Most cited empirical study of saturation in interview research. They coded a corpus of sixty women's-health interviews from West African field sites and tracked when new codes were appearing. Seventy-four percent of all codes were identified after the first six interviews. Ninety-two percent after the first twelve. After twelve, the new-code-per-interview rate dropped to near zero. That gave the field the most-cited heuristic — if your sample is reasonably homogeneous and your analytic question is reasonably focused, saturation will arrive around twelve interviews.

**Sarah:** And the third is the Hennink-Kaiser systematic review.

**Kiffer:** Monique Hennink and Bonnie Kaiser, twenty twenty-two, in Social Science and Medicine. They identified twenty-three empirical studies that had measured saturation and synthesized the findings. Their conclusions are the contemporary state of the art. Saturation typically arrives between nine and seventeen interviews. Twelve is a reasonable central estimate but is neither a floor nor a ceiling. Saturation point varies systematically with study design — narrower questions and more homogeneous samples saturate faster. Different kinds of saturation arrive on different schedules — code saturation early, meaning saturation later, theoretical saturation latest. And most importantly, saturation should be operationalized and reported, not asserted.

**Sarah:** That last point feels like it does a lot of work.

**Kiffer:** It does. A methods section that simply says "saturation was reached" without indicating how it was defined and assessed has not done its work. The textbook standard is that you say which kind of saturation, you give the operational test, and you say what variation your sample didn't reach.

**Sarah:** And there's a useful concrete example for code versus meaning saturation worth flagging. Hennink, Kaiser, and Marconi did a study in twenty seventeen.

**Kiffer:** Yeah, that one. They showed that in a sample of twenty-five women's-health interviews, code saturation arrived at nine interviews — they'd identified the major coding categories. But meaning saturation required sixteen to twenty-four — the substantive understanding of what those codes were really about needed almost the full sample. So if you saw a paper say "saturation occurred at nine interviews," and you read that as "they had a full analytic understanding at nine," that's wrong. They had the inventory of codes. They didn't yet have the depth.

**Sarah:** And the bias in published reports.

**Kiffer:** A bias toward code saturation because it's faster, easier to measure, and easier to defend in a paper. But code saturation is the floor, not the ceiling. Most analyses worth reading have done meaning-saturation work whether or not the methods section uses that vocabulary.

**Sarah:** And there's a separate methodological strand that pushes back on saturation as a concept altogether.

**Kiffer:** Braun and Clarke, particularly in their reflexive thematic analysis work, have argued that saturation presupposes a realist epistemology that not every qualitative tradition shares. For interpretive work, the "new information" framing assumes meaning is waiting to be inventoried, when meaning is actually co-constructed. The course's pragmatic stance is that saturation is a useful default standard for applied health research and the empirical literature is helpful, but you should know there are traditions in which the concept is contested. If your capstone uses reflexive thematic analysis or constructivist grounded theory, you can defend a sample on grounds other than saturation.

**Sarah:** Before we go on to strategies, let me ask about something the lesson mentions. The "n equals twenty is underpowered" reflex.

**Kiffer:** It's a reflex that you will need to consciously override for the rest of this course. The right question for a qualitative study with twenty transcripts is not "was the power adequate?" It's "what configurations were captured? What variation does the sample cover? What claims can the analyst defensibly make on that basis?" Those questions are answered by the methods section, not by an arithmetic check.

**Sarah:** And it's a reflex that journal reviewers sometimes still trigger.

**Kiffer:** They do. A reviewer trained in trials sometimes returns a review of a qualitative paper that says "the sample is too small." The polite response is a paragraph in the rebuttal explaining the sampling logic. The reviewers who know qualitative work don't make the comment. The reviewers who don't, learn. The author's job is to make the methods section so clear that the comment is hard to make.

**Sarah:** Okay, section three. The six nonprobability strategies.

**Kiffer:** Right. And they're not mutually exclusive — most real studies combine two or three. But each has its own logic, its own appropriate uses, and its own characteristic failure modes. Let me walk through them.

**Sarah:** Start with quota sampling.

**Kiffer:** The deliberate construction of a sample to match pre-specified target cells across one or more dimensions of variation. You decide in advance that you want, say, five women and five men, or three participants in each of four age quartiles. You recruit until each cell is filled, accepting whoever in that cell is available. The strength is that coverage of the dimensions of interest is guaranteed by construction. The weakness is that within each cell, selection is typically convenience-based, and the within-cell sample is therefore a convenience sample.

**Sarah:** And the loneliness dataset uses quota elements.

**Kiffer:** It does. The recruitment specified variation targets across four age quartiles, across gender, across living arrangement, across major life-stage transitions — recent immigration, recent loss, recent retirement, recent caregiving role, recent relationship dissolution. The sample was assembled to hit those targets.

**Sarah:** Second strategy. Purposive sampling.

**Kiffer:** Sometimes called judgment sampling or purposeful sampling. The deliberate selection of cases on the basis of theoretical or substantive criteria. The analyst chooses cases expected to illuminate the phenomenon, on the grounds that those cases are richer or more strategically located than randomly drawn ones would be. Selection criteria are explicit. The logic is curatorial. Patton catalogued more than a dozen sub-types. The most commonly invoked in health research are maximum-variation, homogeneous, extreme or deviant case, critical case, typical case, and confirming or disconfirming case sampling.

**Sarah:** And a concrete example for purposive sampling done well. The textbook leans on maximum-variation sampling because that's the modal capstone design. But the other sub-types matter.

**Kiffer:** Yeah. Worth giving a quick gloss on each. Homogeneous sampling is when you want within-group patterns to be visible — say, twenty interviews with first-year nursing students about clinical burnout, where you're not trying to vary across professions because the question is specifically about that group. Extreme or deviant case sampling is when unusual or boundary cases make analytically invisible features visible — interviewing the one person on the unit who hasn't experienced burnout is sometimes more informative than interviewing five who have. Critical case sampling is when you pick a case on the theory that if a phenomenon shows up here, it'll show up anywhere — or won't show up at all. Typical case sampling exemplifies the modal pattern. Confirming and disconfirming case sampling specifically tests or strains an emerging interpretation. Each one is a different analytic move.

**Sarah:** And the choice signals to the reader what kind of inference you're making.

**Kiffer:** Exactly. Naming the sub-type is part of being specific about what the sample can support.

**Sarah:** And the loneliness dataset is best characterized as.

**Kiffer:** Purposive with quota elements, leaning toward maximum variation. The twenty transcripts were not drawn to be statistically representative of British Columbia, and they were not assembled by simply filling demographic cells. They were chosen to capture variation the literature suggests matters for loneliness — age, gender, gender-identity history, immigration trajectory, caregiving role, life-stage transition, and identity configurations the literature under-describes.

**Sarah:** Third. Convenience sampling.

**Kiffer:** Recruitment of whoever is available, accessible, or willing, with no purposive criterion other than availability. The textbook is careful, not dismissive. There are defensible uses — piloting an interview guide, testing the recording equipment, training a new interviewer, building rapport in a community before formal recruitment begins. There are also indefensible uses, like a published study that recruited only the readily available and then claims findings that generalize beyond them. The line between the two is what the methods section says. If a convenience sample is reported transparently as such, with limitations honestly acknowledged, it is a defensible piece of empirical work. If it's dressed up as something more representative, it is not.

**Sarah:** Let me push on what "honestly acknowledged" looks like in practice. Because it's easy to say in the abstract.

**Kiffer:** Sure. A defensible convenience-sample acknowledgment in a methods section reads something like, "Participants were recruited through a community-clinic email list and through flyers posted in three Vancouver neighbourhoods. Recruitment was by convenience and did not target specific demographic or experiential variation. The sample is therefore likely to over-represent adults who use the clinic, who read the email list, and who responded to flyers, and to under-represent adults who are not engaged with formal health services. Findings should be interpreted as describing experiences within this recruitment frame and not as generalizing beyond it." That's a transparent acknowledgment. It tells the reader who you got and who you didn't.

**Sarah:** And it doesn't require apologizing.

**Kiffer:** Right. No apologies. Just specification. The sample is what it is. Your reader can decide what claims to weigh against it.

**Sarah:** And the dirty secret is that a lot of published qualitative health work is convenience sampling that hasn't admitted to itself that it is.

**Kiffer:** Yeah. A more honest field would acknowledge this and would more carefully describe what claims a convenience sample does and does not support. The four-screen discipline from last week applies here too — be honest about what your sample can support.

**Sarah:** Fourth. Network sampling. Two flavours.

**Kiffer:** Snowball and respondent-driven sampling, R D S. The distinction is methodologically important. Snowball sampling is the classical form. Initial participants — seeds — are asked to refer others, who refer others, and so on. Standard tool for reaching populations that are hidden, stigmatized, hard to identify from sampling frames, or organized around relationships rather than addresses. People who use drugs, sex workers, undocumented migrants, LGBTQ adults in regions where outness is risky. Strength is access. Weakness is that the sample is shaped by the social networks of the initial seeds, which are unlikely to be representative.

**Sarah:** And respondent-driven sampling adds machinery.

**Kiffer:** R D S was developed by Douglas Heckathorn in the late nineteen-nineties. Dual incentives — participants are paid for their own participation and for the participation of recruits they bring in. Limited coupons — each participant can recruit only a fixed number of others, typically three. Plus a tracking-and-weighting framework that allows population-level inference under specific assumptions about network structure.

**Sarah:** And R D S matters because it's the most serious attempt to turn network sampling into something with calculable inferential properties.

**Kiffer:** Under Heckathorn's assumptions — long recruitment chains, accurate self-reported network sizes, random recruitment within social networks — R D S estimates can be weighted to approximate population-level prevalence and association estimates with calculable error bounds. The assumptions are strong and have been criticized, but R D S remains the most defensible network-sampling design for many hidden populations of public-health interest.

**Sarah:** The loneliness dataset didn't use snowball or R D S.

**Kiffer:** It didn't. The purposive-with-quota design was assembled directly through recruiters. Worth recording in your methods section because snowball and R D S come with specific analytic obligations a non-network study cannot claim.

**Sarah:** Fifth strategy. Theoretical sampling. The Glaserian version.

**Kiffer:** This is the one most often confused with purposive sampling, and the textbook is clear that they are different things. Theoretical sampling is iterative, emergent, and analysis-driven. You collect and analyze some data, develop a preliminary theory, identify what additional data would test or extend the theory, sample those additional data deliberately, re-analyze, and continue until theoretical saturation.

**Sarah:** The sampling and the analysis are not separated.

**Kiffer:** That's the key. Each round of sampling is shaped by what the previous round revealed. Purposive sampling is a-priori — you decide before recruitment what variation you want. Theoretical sampling is emergent — you cannot say in advance what cases you will want, because the criteria are determined by the developing analysis. A methods section that calls a purposive sample a theoretically sampled one is making a category mistake, and it's a common one.

**Sarah:** And our dataset is not theoretically sampled.

**Kiffer:** It's not. The twenty transcripts were assembled in a single recruitment phase with variation targets specified in advance. That's purposive. Worth being clear about, because using the wrong term in your methods section signals to a careful reader that you may not understand the distinction.

**Sarah:** Quick aside on the Glaser-Strauss split, because it's something students bump into in the methods literature.

**Kiffer:** Worth a moment. Within grounded theory, the original co-authors split methodologically in the nineteen-eighties. Glaser kept theoretical sampling tightly tied to emergence from the data. Strauss, with Corbin, developed a more structured version that allowed more a-priori coding. Charmaz's constructivist grounded theory — which we'll meet later in the course — sits closer to the Glaserian original. For our purposes, when we say "theoretical sampling," we mean the Glaserian sense. Iterative, emergent, analysis-driven.

**Sarah:** And the methodological lesson is that words matter.

**Kiffer:** Yeah. A methods section that uses "theoretical sampling" loosely, to mean any deliberate sampling, is signaling that the writer didn't read carefully. The careful version specifies which tradition they're working in and uses the term in that tradition's sense.

**Sarah:** Sixth strategy. Key-informant sampling.

**Kiffer:** Key informants are individuals selected because they are unusually knowledgeable, articulate, or strategically located with respect to the phenomenon. In a study of clinic operations, the key informants might be the head nurse, the intake coordinator, the medical director — people whose roles give them an overview no individual patient could provide. In a study of a religious community, the key informants might be elders, clergy, longtime members. The selection criterion is access to information, not representativeness.

**Sarah:** And there are two uses.

**Kiffer:** Key informants can replace broader sampling when the research question is about a system or institution and the key informants are who knows it. A study of how a province's overdose-response protocol gets implemented might rely largely on interviews with the small number of people who actually implement it. Key informants can also supplement broader sampling when the analytic question requires both lay perspectives and expert ones. A loneliness study might interview twenty lay participants — like our dataset — and supplement with two or three key-informant interviews with clinicians, community-organisation directors, or older-adult-services coordinators.

**Sarah:** And the failure modes.

**Kiffer:** Key-informant interviews are typically deeper, longer, and more iterative than lay interviews. They are also more vulnerable to the informant's framings being adopted by the analyst — the "going native" problem familiar from anthropology — and to the political dynamics around who gets named a key informant in a community. Choosing a community's key informant is itself a political act, and you should think about who you're elevating.

**Sarah:** And the textbook flags something about the politics of being named a key informant.

**Kiffer:** It does. In community work, the people who are most willing to be key informants are often the most institutionally embedded — directors, longtime members, people with formal roles. They have the access and the legitimacy to speak for the community in formal interviews. But "speaking for the community" is itself a political act. Other community members may have different stories. A study that relies entirely on institutionally-embedded key informants is, in effect, taking a particular political position about whose voice counts. The honest move is to know that, name it, and supplement where you can.

**Sarah:** That's a piece of methodological reflexivity.

**Kiffer:** Which connects back to positionality. Who you choose as a key informant is partly a function of who you can reach, who responds to you, who feels comfortable with your institutional location. Your positionality shapes your sampling. The transparency move is to name it.

**Sarah:** Okay. Section four, briefly. What does your methods section owe the reader?

**Kiffer:** Four things, when you're defending a qualitative sample. First, the sampling logic. Name your strategy, or strategies — purposive with quota elements, plus key-informant supplement, say — and justify the choice with reference to the research question. Second, the variation you captured. List the dimensions you sampled across. Third, the variation you did not capture, honestly. Who isn't here, and what that means for the claims you can make. Fourth, the saturation account. Which kind of saturation are you claiming, on what evidence, and what's outside the scope of the claim.

**Sarah:** That fourth one is the hardest for students who came from quantitative training.

**Kiffer:** Because it requires you to make a positive claim about an absence — "we did not reach the loneliness of unhoused adults" — that feels like an admission of failure. It isn't. It's a piece of methodological discipline. Every study has a finite scope. Saying explicitly what's outside your scope is what lets your reader trust what's inside it.

**Sarah:** And a quick callback to the four-screen discipline from last lesson. The ethics screen and the resources screen both bite on sampling.

**Kiffer:** They do. Ethics: who can you recruit and how do you do it without causing harm? Resources: what sample is twelve weeks of fieldwork going to actually produce? The four screens are not done after week two. They reappear at every step.

**Sarah:** Let me try the synthesis. First takeaway. Probability and nonprobability samples were built to do different jobs. Probability samples estimate population-level magnitudes. Nonprobability samples characterize phenomena. Neither is a degraded version of the other.

**Kiffer:** Second. The sample-size logic for nonprobability work is fundamentally different. Probability sampling adds precision around a parameter with each additional case. Nonprobability sampling is governed by informational redundancy — saturation. The shape of the argument is not the same.

**Sarah:** Third. Saturation has three flavours that need to be kept distinct. Theoretical saturation, code saturation, meaning saturation. They arrive on different schedules and require different operational evidence. A methods section that says "saturation was reached" without naming which kind has not done its work.

**Kiffer:** Fourth. The empirical literature on saturation — Romney-Weller-Batchelder four-to-six, Guest-Bunce-Johnson twelve, Hennink-Kaiser nine-to-seventeen — gives you defensible heuristics but not formulas. The numbers vary with the focus of your question, the homogeneity of your sample, and the kind of saturation you're claiming.

**Sarah:** Fifth. There are six nonprobability sampling strategies — quota, purposive, convenience, network including snowball and R D S, theoretical, and key-informant. Most real studies combine two or three. The strategies are not interchangeable. Theoretical sampling is not the same as purposive sampling, and using the terms interchangeably is a category mistake.

**Kiffer:** Sixth. Convenience sampling has defensible uses and indefensible uses. The line is what the methods section says. A transparent convenience sample is methodologically respectable. A convenience sample dressed up as something else is not.

**Sarah:** Seventh. Network sampling — and especially respondent-driven sampling — is the most sophisticated machinery the field has for reaching hidden populations with calculable inferential properties. The assumptions are strong, but it's the most defensible move for many public-health populations of interest.

**Kiffer:** Eighth. Your methods section owes the reader four things. The sampling logic. The variation captured. The variation not captured. The saturation account. Being explicit about what your sample doesn't cover is a piece of methodological discipline, not an admission of failure.

**Sarah:** And the callback to earlier in the course.

**Kiffer:** The three commitments from Lesson 1 — systematic, transparent, replicable — are doing real work in this lesson. Your sampling needs to be specifiable, your methods section needs to make the moves visible, and another competent researcher needs to be able to read your sampling defense and reach coherent conclusions about what your study could and could not support. Sampling is one of the most visible pieces of the transparency commitment.

**Sarah:** Anything you want listeners to carry into next lesson?

**Kiffer:** One thing. The capstone dataset was assembled for you, so you don't get to make the sampling choices. What you do get to do — what your methods section has to do — is describe the sampling logic in a way that's honest about what twenty curated transcripts can and cannot support. The work this week is documentation, not decision. But documentation is real analytic work.

**Sarah:** Next lesson, we move from sampling to data collection. Indirect observation, direct observation, elicitation. Interviews and focus groups in detail. And the analytic status of transcription, which it turns out is itself an analytic act, not a technical chore.

**Kiffer:** Before then, draft the sampling memo and the sampling matrix for the loneliness dataset. Be explicit about the variation captured and not captured. And start thinking about what kind of saturation you'll claim.

**Sarah:** Thanks for listening. We'll see you in Lesson 4.

**Kiffer:** Take care of yourselves. See you in class.