# Lesson 10: Discourse Analysis: Conversation and Performance *Companion-podcast transcript • Sarah & Kiffer* *Office Hours episode to listen to after working through the lesson* --- **Sarah:** Here's a one-word experiment. Twenty family doctors in Los Angeles County and Pennsylvania were each given one of two questions to ask, once the patient had described the main problem. Is there anything else you want to address in the visit today? Or the same question with one word changed: is there something else you want to address in the visit today? **Kiffer:** And that one word mattered. In the study, led by the conversation analyst John Heritage and published in 2007, the something version cut patients' unmet concerns by about three quarters, compared with visits before the doctors used a scripted question. The anything version wasn't significantly better than those unscripted visits, and neither version made the visits noticeably longer. **Sarah:** So the question was already shaping the answer before the patient said a word. **Kiffer:** That's the starting point for discourse analysis. A question does work before any answer arrives, and that includes every question in a research interview. **Sarah:** Welcome back to Office Hours. I'm Sarah. **Kiffer:** And I'm Kiffer. What a question does to the answer that follows sits right at the centre of Lesson ten, Discourse Analysis: Conversation and Performance, so let's start there. **Sarah:** Question one. When we call an answer dispreferred, do we mean the speaker didn't want to give it? That's how the word sounds to me, and I suspect it sounds that way to most people the first time they meet it. **Kiffer:** It's the natural reading, and it's the one conversation analysts warn about. Preference describes the shape of a turn in relation to the turn before it. The doctor study shows this well. The patients wanted to raise their other concerns. They'd come in with them. What the wording changed was which answer fit the question. **Sarah:** Because anything is the word we use when we expect a no. Do you have any questions? Is there anything else? **Kiffer:** That's the explanation the study's authors gave. Any leans negative. It signals that a no is expected, so a no fits the question and can come quickly and plainly, with nothing attached. A yes to an anything question has to swim a little upstream. Something leans the other way, which makes raising a new concern the answer that fits. What changed between the two versions was which answer was easy to give. **Sarah:** Which is uncomfortable for qualitative interviewers, because a great many interview guides end with, is there anything else you'd like to add? **Kiffer:** They do, and Heritage and his colleagues suggested the effect may extend to questions like, do you have any questions? Their study was about clinical visits, so I'd hold the conclusion loosely for research interviews. Still, I think it's worth piloting a something version of the closing question and seeing whether participants raise more. Either way, the closing question belongs in your analysis, because it can shape what comes back. **Sarah:** It also shows why conversation analysis has found a home in health services research. When the finding is about the shape of a question, the fix can be as small as changing a sentence. **Kiffer:** That's a big part of its appeal in clinical work, and Heritage and Douglas Maynard built a large body of research on medical visits this way. Let's test the idea with something closer to the loneliness interviews. Here's a made-up exchange. A participant says, I know, I'm a bit pathetic, really. That's an assessment, and as a general rule, assessments prefer agreement. So what's the preferred next move for the interviewer? Take a few seconds, and pause the audio if you'd like more time. *(Pause)* **Sarah:** Disagreement. If the interviewer agrees, they're calling her pathetic. Anita Pomerantz described this in her work on assessments. After a self-deprecation, the usual preference flips. Agreement becomes the dispreferred response, and disagreement is the one that tends to come quickly and plainly, something like, oh, no, not at all. **Kiffer:** That changes how we can read Maya's line from the lesson, sorry, that's a weird answer. We read it as face-work that heads off the interviewer's negative evaluation. It's also a self-deprecation, so it makes reassurance the relevant next move. When an interviewer answers, no, that's a really interesting answer, the reassurance was called for by the sequence. It tells you about the sequence more than about the interviewer's private view of the answer. **Sarah:** And if the interviewer says nothing, because they've been trained to stay neutral? **Kiffer:** Then the silence is noticeable, because a disagreement was the relevant next move and it didn't come. The self-criticism is left standing. Interviewers can't step outside the sequence, so I'd treat their turns, and their silences, as part of the data. The participant's next turn will often show you how the silence or the reassurance was taken. **Sarah:** Question two, and it's a worry a lot of public health students have. If every interview answer is a front-stage performance, can we still believe what participants tell us? Robert says he drinks more on the off-weeks. Is that evidence about his drinking or about his self-presentation? **Kiffer:** It's evidence about both, and that was Goffman's point. In his sense, performance includes sincere talk. Sincerity is itself something people perform, and every audience gets a performance of some kind. The conversation Robert has with a friend afterward is back-stage relative to the interview, and it's still a performance for a different audience. So the analysis describes the work the talk is doing, for whichever audience it's addressed to. **Sarah:** Goffman also has a Canadian connection that surprises people. **Kiffer:** He was born in Mannville, Alberta, in 1922, grew up in Manitoba, and did his first degree at the University of Toronto. For loneliness, his most useful work may be Stigma, from 1963, subtitled Notes on the Management of Spoiled Identity. He separated the discredited, whose stigma is already visible or known, from the discreditable, whose stigma isn't immediately visible or known. For the discreditable, the everyday problem is managing information: whom to tell, how and when. **Sarah:** And loneliness mostly can't be seen. So a lonely participant is discreditable, and much of the interview involves decisions about how much to disclose. **Kiffer:** That's how I'd read the prefaces in the loneliness interviews. Lines like, this is going to sound dramatic, or, I'm not a feelings guy, are Goffman's information management, happening turn by turn. Reading them that way leaves Robert's report every bit as credible. The care he takes shows that he treats the disclosure as delicate, and you can point to the exact lines where he does it. **Sarah:** So where's the line? What can I claim from Robert's excerpt, and what can't I? **Kiffer:** You can claim what the talk shows him doing. He frames the feelings talk as out of character, he softens his evaluation of being caught drinking, and he closes each answer with a summary that hands the floor back. What the talk alone can't support is a hidden back-stage truth, for instance that he's more distressed than he lets on. Goffman's framework tells you a performance is under way. It gives you no access to a back-stage version the interviewer never heard. **Sarah:** That has a practical side for public health. If disclosure is managed this carefully, a screening question that asks, are you lonely, asks people to apply the stigmatized label to themselves in a single word. And survey designers seem to have reached a similar conclusion by another route. One widely used three-item scale asks how often you feel that you lack companionship, feel left out, or feel isolated from others, and it never uses the word lonely. **Kiffer:** Right, and survey guidance on measuring loneliness recommends asking those indirect items before any direct question, because the scale is meant to measure loneliness without naming it. The same guidance notes that some people, men in particular, may under-report loneliness when they're asked about it directly. It's a case where measurement design and discourse analysis point the same way. **Sarah:** Let me set one up on footing, because that's the part of Goffman I find hardest to apply. Here's an invented line from a participant. My doctor told me I should just get out more. Like I haven't thought of that. So, for everyone listening, in that first sentence, who is the animator, who is the author, and who is the principal of, get out more? Take a few seconds. *(Pause)* **Kiffer:** The participant is the animator, since she's the one producing the words. The doctor is presented as the author, and as the principal, because it's the doctor's position being reported. Then, like I haven't thought of that, is a footing shift. Now she's animator, author and principal at once, and the position is her own: the advice is obvious and of little use to her. **Sarah:** And the arrangement lets her reject the advice without arguing with it head on. She gives the doctor's words and then lets the irony do the disagreeing. **Kiffer:** There's face-work in it as well. The irony shows that she's already thought of getting out, so her loneliness can't be put down to a lack of effort. That's the kind of blame management discursive psychologists look for. **Sarah:** One caution, though. We don't know the doctor said just, or said any of it in those words. **Kiffer:** That's an important habit to build. A quotation inside an interview is the speaker's construction. You can analyse what the quoting does in the interview. You can't treat it as a record of what happened in the clinic. **Sarah:** Question three. Can you really do conversation analysis on transcripts that were never made for it? The loneliness transcripts are verbatim, the pauses aren't timed, and there's no audio. And Jonathan Potter and Alexa Hepburn pressed a wider criticism of qualitative interviews in psychology in 2005. They argued that interview studies often leave the interviewer out of the analysis, let social science agendas and categories dominate the talk, and treat talk as a window onto inner mental states. **Kiffer:** It's a fair challenge. Their remedy was to analyse interviews as interactions and, for many questions, to use naturally occurring talk. Inferring Jefferson features from a verbatim transcript is a teaching exercise, and a real study would need the recordings. The exercise trains a habit of attention. It can also train a bad habit, if you're careless about where your evidence comes from. **Sarah:** Then let me try a re-transcription and see where it gets me. Here's an invented exchange. The interviewer asks, do you get to see your family? The participant says, yeah. Well, at Christmas and stuff. They're all busy, with the kids and everything. It's fine. And the interviewer says, okay, and what about friends? **Kiffer:** Go ahead and re-transcribe it in your head, then tell us what you'd mark and what you'd conclude from it. **Sarah:** I'd put a pause of about a second before yeah, because it sounds reluctant. Another half second before well. Stress on Christmas, and a falling pitch on fine. Then the reading: the delay before yeah shows the answer is dispreferred, the falling pitch on fine sounds resigned, and the whole answer tells me she's hiding hurt behind it's fine. **Kiffer:** Where did that one-second pause come from? **Sarah:** From me. I put it there because the answer sounded reluctant. Oh. And then I used the pause as evidence that it was reluctant. That's circular. **Kiffer:** That's the trap in inferred re-transcription. Every feature you infer comes from a reading you already hold, so it can't count as independent evidence for that reading. The falling pitch on fine has the same problem. And hiding hurt is a claim about a back-stage feeling, which we've just said the talk can't give you. So strip out everything you added. What in the answer still marks it as something other than a plain yes? Take a few seconds. *(Pause)* **Sarah:** Everything that's in the words. The yes is followed by well, and the verbatim keeps that. Then a restriction, at Christmas and stuff, which quietly takes back most of the yes. Then an account, they're all busy, with the kids and everything. Then a closing assessment, it's fine. A yes followed by a qualification and an account has the shape of a delicate answer, and none of it depends on my pauses. **Kiffer:** Right. And look at who the account protects. She explains the family's absence on their behalf, which heads off the inference that they've neglected her. That's blame management again. Then, it's fine, does the same job as Robert's, that's it. It's a closure that hands the floor back. And the interviewer takes it, by moving straight on to friends. **Sarah:** So the interviewer's okay, and what about friends, is data too. An interviewer who'd held back, the way Robert's interviewer did after, that's the whole answer, might have heard more. **Kiffer:** Yes, and that's the kind of thing Potter and Hepburn wanted reported. So a defensible write-up says the participant answers yes, restricts it to holidays, accounts for the family's absence on their behalf, and closes the topic with an assessment, and that the interviewer accepts the closure. Any pause or pitch you add, label as inferred, and keep it out of your evidence. **Sarah:** The same logic catches something about the dramatic prefaces. Maya's came right after a question about what loneliness feels like, and Robert's preface about not being a feelings guy came right after, how does it feel physically? **Kiffer:** That's a good catch. Both prefaces answer questions that ask for feelings talk, which these participants treat as delicate. So before you treat the frequency of prefaces as a finding about lonely people, check where they fall in relation to the interview guide. And ask whether people would preface any delicate disclosure to a stranger with a recorder. Without that comparison, I'd describe it as a finding about how people in these interviews handle delicate questions. **Sarah:** Question four, and this is where we disagree. Should the analyst bring in power and context that the participants never mention? **Kiffer:** This was argued out in the journal Discourse and Society in the late 1990s. Emanuel Schegloff started it in 1997 with a paper called Whose Text? Whose Context? He argued that the relevant context is the one participants show to be relevant in their own talk, and that analysts who import their own preoccupations risk a kind of theoretical imperialism. Margaret Wetherell replied that this was too narrow for critical work, and Michael Billig argued that conversation analysis carries its own assumptions about the social world. **Sarah:** And I'm closer to Wetherell. Let me show why with an invented health promotion line. Picture a poster in a bus shelter. Feeling lonely? Reach out to a friend today. Connection starts with you. **Kiffer:** Before you read it, let's give everyone a turn. What does the line, reach out to a friend today, take for granted? Take a few seconds. *(Pause)* **Sarah:** That the reader has a friend to reach out to. That the barrier is initiative, so the fix is a decision you can make today. And that the friend will welcome the call. Then, in Fairclough's first dimension, connection starts with you makes the lonely person the agent of change. Landlords, employers, transit planners, whoever sets shift schedules: none of them appear in the sentence. **Kiffer:** That's a careful reading of the text, and I'd accept it as a reading of the poster. **Sarah:** Here's my worry, though. Take an invented participant who says, I know I should make more of an effort. If I have to wait for her to mention housing costs or shift work before I can bring them in, I'll wait forever, because the individual framing is so ordinary that nobody questions it. The ideas that have become common sense are exactly the ones people don't raise. **Kiffer:** That's the strongest form of the argument, and I take it seriously. Where I'd push back is on evidence. Her line shows her orienting to effort as the issue, and you can point to it. The participant from earlier who quoted her doctor met, get out more, with irony, so she's resisting the same framing, and you can show that too. A claim that public health messaging produced the framing needs different evidence: the messages themselves, read closely, and how they circulate. **Sarah:** And real texts give us plenty. The United States Surgeon General's 2023 advisory is titled Our Epidemic of Loneliness and Isolation. Epidemic brings the language of outbreaks with it: something spreading, to be counted and tracked, with public health in charge. Our builds a national we. I don't need a participant to say epidemic before I can notice what the word does. **Kiffer:** You don't, because there the document is your evidence, and I think that's sound critical discourse analysis. My concern is narrower. It's the jump from a reading of a public text to a claim about what's going on in participants' talk, when the talk doesn't show it. That jump is where analysts tend to find in the interview what they brought to it. **Sarah:** I'd still say an analysis that waits for participants to name power will miss what power makes hard to say. But I can live with a shared rule. **Kiffer:** Then here's one I think we both hold. Label each claim by its warrant. Some claims rest on what participants orient to in the talk, so show the lines. Others rest on your reading of a public text or a social context, so show that text and say what you're bringing to it. Either way, give enough detail that another analyst could disagree with you on specific points. **Sarah:** Agreed. The poster reading and the interview reading can sit in the same paper, as long as the reader can tell which kind of claim is which. **Kiffer:** Let's pull it together with three things to take away. **Sarah:** First, preference describes the shape of a turn in relation to the turn before it. A self-deprecation reverses the usual preference, and a single word like anything can make no the easy answer. **Kiffer:** Second, a participant can be performing and sincere at the same time. Analyse the face-work and footing the talk shows, and leave claims about a hidden back-stage self alone, because the interview never gave you access to one. **Sarah:** And third, when you re-transcribe from a verbatim source, nothing you infer can serve as evidence for the reading that produced it. Build claims on the words and on the interviewer's turns, and label the rest as inferred. **Kiffer:** If you'd like more practice, take the invented family excerpt from this episode and write it out yourself. Mark which features are in the verbatim words and which you'd be adding, then write two sentences of analysis that use only the first kind. **Sarah:** Next time, it's Lesson eleven, Analytic Induction, QCA and Decision Models, where we move to the model-testing side of qualitative work, with negative cases and truth tables. **Kiffer:** Take care, everyone. **Sarah:** See you in Lesson eleven.