Interviews

The interview is the most used and least understood method in the social sciences. It looks like a conversation, so people assume they can do it; it produces quotations, so people assume the quotations are evidence of what they appear to be evidence of.

Neither assumption survives contact with the method. An interview is a peculiar social occasion with its own rules, and what it produces is not a report from inside someone's head. It is an account, produced jointly, for a particular listener, at a particular moment.

Understanding that is not a caveat. It is what makes the method usable.

Everyone gave the same answer, fluently, and it was the wrong data.

A researcher is studying why experienced nurses are leaving a hospital trust. She arranges twenty interviews with people who have resigned in the past year.

The first six go beautifully. Everyone is articulate, everyone is generous with their time, and everyone says approximately the same thing. Burnout. Understaffing. Pay that has not kept up. The pandemic. Moral injury.

She has strong, quotable material and a clean, coherent story that matches everything in the press.

And she is troubled by how fluent it is. These answers arrive fully formed, in similar words, with the same emphases — as though each participant had already said it several times, to friends, to family, to an exit interview, to themselves in the car.

So she changes what she asks for.

She stops asking why did you leave and starts asking for events. Tell me about the last shift you worked. Walk me through the day you decided. What happened the week before that? Who did you tell first, and what did you say?

The material transforms.

One nurse describes a specific Tuesday: a request to swap a shift, refused by a manager who had granted the same request to someone else two weeks before, and a sentence — "we all have families" — delivered in front of three colleagues. She had been thinking about leaving for a year. She applied for another job that night.

Another describes a patient's relative complaining about her, an investigation that took four months and cleared her completely, and the fact that during those four months nobody senior asked how she was.

A third describes doing the arithmetic on childcare and realising that a bank shift paid more than her substantive post for the same work, and what it felt like to realise the institution had priced her loyalty at less than nothing.

Now look at the relationship between the two rounds of interviews.

The first round was not false. Every one of those nurses was burnt out, underpaid and understaffed. It was a summary — a socially available, publicly legitimate account of a decision, of the kind we all assemble to explain ourselves.

The second round is not "the truth behind the story" either. It is a different kind of material: specific, dated, checkable in parts, and containing things the summary had smoothed away — the arbitrariness, the humiliation, the moment.

And the difference between the two rounds was made entirely by the interviewer. Same people. Same subject. Different question form, different data. That is what "the interview is not a window" means in practice.

If the account is produced in the interaction, what is it evidence of?

There are two respectable answers, and a study should know which one it is giving.

The information-gathering view. The interview is a route to facts: what happened, when, in what order, who was present, what the person did. The account is treated as testimony, imperfect and improvable, to be probed for specificity, checked against other accounts, and corroborated where possible.

The meaning-making view. The interview is an occasion on which a person constructs a version of themselves and their situation, for this listener. Holstein and Gubrium call this the "active interview": the interviewer is not extracting a pre-existing answer but participating in producing one, and the analysis should treat the account as a construction — attending to how it is built, what it justifies, what it leaves out, whom it addresses.

Both are legitimate, and they are not compatible in the same sentence. A study cannot treat a quotation as a factual report of what happened and as an artefact of self-presentation, depending on which is convenient.

The common failure is silent oscillation: quoting an account as evidence of an event when it flatters the argument, and as evidence of "how participants construct their experience" when it does not. Declare which you are doing — this is 7.1.3's ontology question, arriving with a recorder on the table.

The varieties

Six forms, on a spectrum of who controls the shape.

Structured interviews — identical questions, identical order, fixed responses. This is a survey administered aloud (see 7.4.1) and belongs to that toolkit.

Semi-structured interviews — a topic guide of areas to cover, with wording and order adapted to the person. The workhorse of qualitative sociology, and the assumption is that comparability comes from covering the same ground, not from using the same words.

Unstructured or in-depth interviews — one opening, then following where the participant goes. Powerful with a skilled interviewer and shapeless with an unskilled one.

Life history and biographical interviews — the whole life, or a long span of it, usually across several sessions. The recall problem is severe, and the standard remedy is a life grid or life-history calendar: a chart of years against domains — housing, work, family, health — filled in together, so that anchoring events help date the others. Recall of when improves dramatically; recall of what improves less.

Elite and expert interviews — with people who are used to being interviewed, have prepared positions, and are practised at not answering. The power relation is inverted, and so is the technique: specific documentary knowledge, precise questions, and a willingness to say that isn't what the minutes record.

Oral history — closer to the archive than to sociology in its aims, concerned with recording testimony of periods and events. Its methodological literature on memory is the best there is, and sociologists use too little of it.

The craft

Eight techniques, roughly in order of how much difference they make.

One — ask for events, not opinions. "Tell me about the last time..." rather than "How do you feel about...". This is the single highest-yield move in interviewing, for the reason the story demonstrates: opinions come pre-packaged; events have to be reconstructed, and reconstruction produces detail the package had removed.

Two — go for specificity relentlessly. When was that? Who else was there? What did they actually say? What did you do next? Generalities are where interviews go to die. A participant saying "they never listen to us" has told you their evaluation; a participant describing a particular meeting has given you data.

Three — be careful with "why", especially early. Why invites a justification, and a justification is a performance of reasonableness. Ask what happened first, and let the reasons emerge from the sequence; ask why late, and treat the answer as an account rather than as a cause.

Four — use silence. The most common novice error is filling a pause. Three seconds of silence after an answer produces more material than almost any follow-up question, because the participant assumes more is expected.

Five — probe in identifiable ways. Elaboration ("tell me more about that"); clarification ("when you say 'managed', what did that involve?"); detail ("walk me through it"); contrast ("was that different from how it went before?"). Learning the vocabulary of the participant is the point of the second one, and it is where the interpretive work happens (see 7.1.1).

Six — do not lead, and notice when you have. "So that must have been frustrating?" has supplied the emotion, and the participant will usually accept it out of politeness. When you catch yourself doing it, the repair is to ask them to describe it in their own terms.

Seven — use vignettes for sensitive or abstract material. Present a short scenario and ask what the person in it should do, or what would happen next. It lets people speak about a situation without claiming it, and it is the interview equivalent of the indirect designs in 7.4.1.

Eight — end well. Is there anything I should have asked and didn't? is the most productive final question in the method, and a surprising amount of the best material in any study arrives after the recorder is switched off — which raises an ethical question you must resolve before it happens, not after.

Interviewer effects: who you are changes what you are told.

This is documented rather than speculative. Survey research has measured race-of-interviewer effects since the 1950s: respondents' answers to racially relevant questions differ systematically according to the perceived race of the person asking, with the effect largest on precisely the items researchers most want to measure.

Parallel effects exist for gender, age, apparent class, accent, dress, and any perceived institutional affiliation. A participant who thinks you might report back to management is answering a different question from one who does not, whatever you said at the start.

And "matching" interviewers to participants is a partial remedy with its own cost. Shared category membership may open some ground and close other ground — participants may withhold criticism of their own group from an insider, or, more subtly, skip the explanation: "you know how it is". The insider is told more and asked to understand more without being told.

The practical response is not to eliminate the effect — it cannot be eliminated, because someone must ask — but to record who you were in the room and reason about which way it would push. Which is 7.1.2's positionality, made specific and useful rather than confessional.

People cannot reliably tell you why they did things.

This is the finding most often ignored by people who use interviews, and it is not a claim that participants lie.

C. Wright Mills made the sociological version in 1940. Motives, he argued, are not inner springs that we report on. They are socially available vocabularies — the accounts a person offers, and is expected to offer, in a particular situation. Different settings supply different acceptable motives: what counts as an adequate reason for leaving a job differs from what counts as an adequate reason in a divorce, and both are learned. So when you ask why, you are collecting a vocabulary of motive, which is real and interesting data about the situation — and it is not a report of causation.

Psychology reached a converging conclusion by another route. Nisbett and Wilson's review assembled experiments in which people's behaviour was demonstrably influenced by something they were unaware of, and in which they nonetheless produced confident, fluent explanations of their choices — explanations that were, in the circumstances, provably not the cause. People do not generally have access to the processes producing their judgements; they have access to plausible reasons, and they report those.

Two practical consequences.

Treat self-attributed motives as data about accounts, not as evidence of causes. They tell you what is sayable in this world, which is genuinely valuable — it is exactly what an analysis of legitimation needs.

And get at causation the way the rest of this Part does: through sequence, through comparison, through what varied, through cases where the outcome differed. The nurse's Tuesday is worth more than the nurse's theory of herself — not because she is unreliable, but because nobody has that kind of access.

Practicalities that turn out to be analytic decisions.

Recording and transcription. Record if permitted; note-taking alone loses the words, and the words are the data.

And transcription is not clerical. Elinor Ochs's argument — that transcription is theory — is exactly right: every decision about what to include is an analytic commitment. Do you mark pauses? Overlaps? Laughter? False starts? "Um"? Do you tidy grammar into standard written form? Tidying makes participants sound more coherent and more educated, systematically and unequally, and a transcript that removes hesitation removes the evidence of difficulty. Conversation analysts transcribe in obsessive detail because their object is the interaction itself (see 7.5.3); a study of accounts may reasonably do less. What is not reasonable is doing it unreflectively.

Setting. Where an interview happens changes it. At home, at work, in a café, on the phone, on video. A workplace interview has the employer in the room whether or not anyone is present.

Number. The honest answer is that it depends on the question, on how varied the population is, and on how much each interview yields — with saturation the usual criterion (see 7.7.1) and Mario Small's argument the right frame: case-based logic, sequential selection, and no obligation to defend the number in the language of statistical representativeness (see 7.3.1).

Four ways interview evidence gets over-read.

The quote mine. Collecting forty interviews and using six quotations that support the thesis. The reader has no way to know how many said something else — which is why serious qualitative reporting states how many participants expressed a view, and reports the ones who did not.

Quotes as frequency claims. "Participants felt that..." — how many? Two? Eighteen? Vague quantifiers in qualitative writing are doing quantitative work without accountability. Say most, several, two — and mean it.

The articulate participant. Some people give wonderful interviews. They are over-represented in every published qualitative study on earth, and they are not typical (see 7.5.1 on inconvenient cases).

Elite credulity. Powerful interviewees give confident accounts that are frequently self-serving, and their fluency reads as authority. An elite interview is a source to be checked, not a finding — the hierarchy of credibility from 7.1.2, arriving in a nice office.

Because interviews are excellent for a specific set of things and are constantly asked to do others.

What they are excellent for: the categories people use and what those categories distinguish; how people account for and legitimate their actions; chronologies and processes that no record captures; subjective experience; the meaning of an event to those in it; and reaching settings and populations that no other method can enter.

What they cannot do: establish prevalence (that needs a sample — see 7.3.1); measure behaviour (that needs observation or records); establish causal effect (that needs a counterfactual — see 7.4.2); and deliver an accurate account of why someone did something.

The reader's checklist is short. How many people, selected how? What were they actually asked — is the topic guide available? Are there quotations that complicate the argument? Does the author distinguish what participants reported happening from what participants believed about it? And do the numbers behind the words appear anywhere?

And the practitioner's single rule, if you remember nothing else: ask about events. Everything good in the method follows from that one habit, and most of what is thin in the method follows from asking for opinions instead.

An interview produces an account, generated in an interaction, for a particular listener. The nursing study makes it concrete: asking why did you leave produced a fluent, publicly available summary; asking for events produced a specific Tuesday, a refused shift swap and a sentence said in front of colleagues. Same people, different question form, different data.

Two stances, and a study must declare one. The information-gathering view treats the account as testimony to be probed and corroborated. The active-interview view treats it as jointly constructed and analyses how it is built. Silent oscillation between them is the common failure.

Six forms: structured (a spoken survey), semi-structured (the workhorse), unstructured, life history (with a life grid to anchor dating), elite and expert (where the power relation inverts), and oral history.

Eight craft moves, led by the two that matter most: ask for events, not opinions, and pursue specificity relentlessly. Then: be careful with why; use silence; probe in identifiable ways; do not lead; use vignettes for sensitive material; and always ask what you should have asked.

Interviewer effects are documented, not hypothetical — race-of-interviewer effects are largest on the items researchers most want to measure — and matching has its own cost, because insiders are told more and asked to understand more without being told.

People cannot reliably report why they acted. Mills: motives are socially available vocabularies, offered in situations. Nisbett and Wilson: people confidently explain choices whose real influences they demonstrably could not perceive. So treat stated motives as evidence about accounts, and get at causes through sequence and comparison.

Transcription is theory — tidying speech makes participants sound more educated, unequally, and removes the evidence of difficulty.

And the over-readings to refuse: the quote mine, vague quantifiers doing quantitative work, the over-represented articulate participant, and credulity towards fluent powerful sources.

Semi-structured interview / topic guide — areas to be covered, with wording and order adapted to the person.

Active interview — Holstein and Gubrium: the account as jointly produced rather than extracted.

Life grid / life-history calendar — a chart of years against life domains, used to anchor and improve dating.

Critical incident — a specific bounded event used as the unit of questioning.

Probe — elaboration, clarification, detail and contrast follow-ups.

Vignette — a short scenario allowing a participant to speak about a situation without claiming it.

Interviewer effects — systematic differences in answers by the perceived characteristics of the interviewer.

Vocabularies of motive — Mills: motives as socially available accounts offered in situations, not inner reports.

Transcription as theory — Ochs: every transcription decision is an analytic commitment.

Saturation — the point at which further interviews stop changing the categories.

Quote mining — selecting supporting quotations without reporting the distribution of views.

One — run the same interview twice. Ask someone why they made a recent decision. Then ask them to walk you through the week before it, hour by hour where they can. Compare the two accounts and note precisely what the second contains that the first did not.

Two — practise silence. In your next three conversations, count three seconds before responding to something substantial. Note what arrives in the gap.

Three — catch a leading question. Record yourself asking someone about their work for ten minutes. Listen back and mark every place you supplied a word, an emotion or a conclusion.

Four — collect a vocabulary of motive. Ask three people why they left a job. Note the shared structure of the answers, and what a socially unacceptable answer would sound like.

Five — audit a qualitative paper. Find one and count: how many participants, how many quoted, how many quotations complicate the argument, and whether any number at all is attached to a claim about what "participants felt".