Interviews occupy a conflicted place in the sociological imagination. As Benney and Hughes (1956, 137) wrote over half a century ago, “sociology has become the science of the interview.” Despite the surge of other methods in sociology, this still seems to be the case. As Lamont and Swidler (2014, 158) pithily put it, “Interviewing is sociology’s standard workhorse method.” Yet, of all sociological methods, interviewing—and especially in-depth interviewing—has received the least methodological attention. Whereas every other ethnographer seems to feel the urge to write methodological tracts about the epistemology and practice of fieldwork, there are fewer methodological writings on interview methods, and still fewer discussions of inference in interviewing—that is, what we can infer about the world from in-depth interview data.Footnote 1

What relatively little discussion there is of interviews and inference usually centers on a variation of two intertwined themes. First is the question of interpretivist versus realist understandings of interviews. That is, whether interviews should be primarily understood as an active interpretive construction that takes place in the bounded interview-situation, or whether interviews should be understood as a window into other contexts of action—and, if it is the latter case, what are the best ways to make such a window as transparent as possible. Second, there are discussions about the actual relationship between, on the one hand, what we say about how we see and act in the world, and on the other hand, what we actually do when a specific situation arises. In its most recent incarnation, Jerolmack and Khan (2014a) named their suspicion about the relationship between these moments “the attitudinal fallacy.”

In reaction to this second line of questioning, American sociology has recently experienced a rare surge in interest in the question of interviews and inference. As rejoinders to the “attitudinal fallacy” article have staked out, talk is important in itself since talk is a kind of action, and narratives are crucial for any understanding of the self and the social world (Lamont and Swidler 2014); there is still a relatively high correlation between what interviewees and survey respondents say and their actions later on, even if this is not a perfect one (Vaisey 2014); we have ways to get around the most acute problems of inference from interviewing (DiMaggio 2014; Cerulo 2014); interviewing contexts are more ethnographic than they are usually thought to be (Gerson and Damaske, forthcoming; Pugh 2013); ethnographers are as susceptible as interviewers to some of the same problems of inference between situations (Trouille and Tavory 2019); and, finally, can qualitative researchers please stop killing their own and instead try to think together about the things they care about?

This paper takes the debates about inference and situational variation in a different direction. Rather than assuming that there is one right answer to either methodological debate, it takes the question of inference as a practical problem facing sociologists as they try to make sense of their interview transcripts. The key metaphor I use in order to organize the analysis is that of a window between the interview situation and other contexts. Like a door (Peña-Alves 2019; Simmel 1994), a window can be open or closed, and importantly, it can also refract light in ways that allow the outside in indirectly, through a glass darkly. In most interviews—and especially in open-ended and semi-structured interviews—there are: (a) open contexts in which we can cautiously infer about other situations from the interview; (b) rare contexts that we should treat as hermetically closed, in the sense that the interpretation of interview data needs to stay squarely within the bounds of the interview situation; and, perhaps most saliently, (c) refracted contexts in which the relationship between the interview and other situations is less straightforward, although it is patterned in ways that we should better interrogate.

The article proceeds in two stages. First, it outlines and exemplifies moments in which we need to analyze interviews in each kind of interview context. It then focuses on two forms of refracted relations between interviews and other contexts of action, analyzing interviews as refracted images of both people’s landscapes of meaning and what I term talk’s promissory aspect. In doing so, the article makes two contributions. First, on a methodological level, it seeks to clarify how researchers think—and also how they should think—about the kinds of inferences they can make from in-depth interviews. Second, it is also meant as a contribution to our understanding of the relationship between different situations by highlighting how cultural elements used in one situation can act as anchoring devices for understanding how people make meaning elsewhere, as well as by emphasizing how talk constrains and enables future action because other people will organize their actions to align with what people say about themselves and their world.

Open Contexts: Processual Accounts and Symbolic Situations

To begin with, I outline moments in the interview process that we see as relatively open; that is, where researchers can cautiously argue that interview data represent a good-enough image of people’s lives beyond the interview situation. In other words, if the interviewer is careful and takes into account some common challenges, she can make claims about other contexts of action for particular kinds of interview questions. Thus, before we outline what we see as two key open contexts in interviewing, we pay tribute to some common pitfalls of open-context inference.

Desirability Bias

There have been mountains of work on the question of desirability bias. Suffice it to say that in many interviews, people want to put their best foot forward, and so are less likely to tell the interviewer things that they think would reflect badly on them. Thus, where questions touch upon stigmatized aspects of the self, we need to assume that some desirability bias is at play. To take one example, many political pollsters in the United States argue that survey responses should be corrected to account for “shy Republicans” or “gun-shy” respondents (Urbatsch 2019). In response, social scientists have thought long and hard about ways around this problem—perhaps people would be more comfortable reporting stigmatized behavior to a computer (Beauclair et al. 2013), perhaps in some situations strangers are better than local interviewers (Sana et al., 2016), or perhaps we should simply conduct long and probing interviews to mitigate the interviewee’s need to project a desirable self (e.g., Gerson and Damaske, forthcoming). Whatever the workaround, desirability bias is a problem that interviewers must grapple with.

Agency Bias

Far less developed in interview methodology is another form of bias, which may be termed agency bias. That is, the problem of the kind of narratives that are produced when people are asked directly about their lives and actions. In a nutshell, the challenge is that the situation synthetically pressures interviewees to present themselves as the protagonists of their own tales (see Gubrium and Holstein 2002, 6–8). Social life, however, is much more complex. While we are indeed key protagonists in our own lives, most of the action happens elsewhere.Footnote 2 As opposed to desirability bias, which is primarily fueled by the interviewee, agency bias is fueled by the interview situation itself, which structurally locates people as the focus of interaction. Moreover, as we return to below, this bias has its counterpart in an implicit assumption that sociologists commonly make about how the coherence (or incoherence) between interviews and other contexts of action emerges—through the agency of the individual.

Hindsight Bias

Lastly, especially in interviews that focus on processual questions, we need to take care not to assume that the processes narrated by interviewees had to happen the way they did. This form of hindsight bias, well-researched in cognitive psychology (Christensen-Szalanski and Willham 1991), is important in interview methodology. Interviewers too rarely ask about the paths not taken—the near-misses and the plausible turns that could have been—that our interlocutors are often well aware of.

Still, as a large literature points out (at least regarding desirability bias), these biases are not insurmountable challenges (e.g., Cerulo 2014; Gerson and Damaske, forthcoming; DiMaggio 2014). As long as interviewers are aware of such biases and take care to consider them in both their interview design and analyses, they can make arguments about the contexts of action described in interviews. But what are the kinds of moments in which such inferences are warranted?

Process-Oriented Interviews

Perhaps the most important open context in interviewing is what is sometimes termed the “ethnographic interview,” and can be more precisely termed a “process-oriented interview.” Rather than getting at people’s mental maps, schemas, narratives, or other cultural representations, it is an interview that centers on the unfolding of social processes. Practically, these are interviews that emphasize how, who, when, and where rather than why (see, e.g., Becker 1998; Katz 2001; Spradley 1979; Weiss 1995). That is, such interviews focus on processual details: how the phenomena unfolded, with whom, and “what happened next?” (Becker 1998).

The basic motivation for this line of interviewing is that while people may partly fudge their answers, keeping ourselves to “how” questions elicits relatively dependable processual answers. Interviewees more or less know through whom they got their first job or how they met their partner. By focusing on such factual, processual questions, the interviewer is constantly giving the interviewee a kind of mnemonic hook that they can use as they construct their narratives, while the interviewer assembles the building blocks for mechanism-based explanations.Footnote 3 And, moreover, in focusing on the unfolding of action, such process-oriented interviews are crucial if sociologists want to know about “roads not taken” by the people they study: jobs that didn’t pan out, neighborhoods they almost moved to, or partners they almost married. Especially with biographical narratives and organizational history, this is often the only way to get at such data, which is observationally invisible.

Furthermore, while the ethnographic interview is for the most part centered on interviewees’ and organizations’ biographical past, this is not always the case. The ethnographic interview is just as useful for getting at present routines. Interviewers often ask about typical days at the job, where people shop or eat, or who they usually meet. As long as the researcher is careful with their question design and does not assume that the answers they are given will necessarily hold as life-situations change, these are relatively open contexts.

Symbolic Situations

While process-oriented interviews focus on past and present, some interviews—more so surveys than in-depth interviews—attempt to predict later action. In that regard, we seem to know two things with a tolerable degree of confidence. First, that there is no direct transformation rule between what people say in in-depth interviews and situated action. As much as some people may lament the internal discord among the ranks of qualitative researchers, the reason that “talk and action” debates recur every few decades (LaPiere 1934; Deutscher 1966, 1973; Jerolmack and Khan 2014a) is that there is indeed a problem of inference at hand. Second, however, we also know that there is a relatively high correlation between what people say that they would do in interviews and situated action. Glasman and Albarracín (2006), who tried to estimate this relationship, suggest a mean correlation of 0.52; and, as Vaisey (2014) points out, this estimate is probably on the conservative side—a pretty high correlation as correlations go. To give a more readily available example, voting polls—a context of “talk” if there ever was one—are surprisingly accurate.

One way to grapple with the tensions among these two insights is to focus on the structure of the interview situation. As the literature on interviewing makes abundantly clear, interviews are complex social situations. But what kinds of situations are they? What kind of action is engaging in a reflexive account of one’s life in front of someone you don’t know all that well? What family of situations looks like it? Although these are complex questions with far more than one answer—interviews can be sites of self-reflection, confession, self-presentation, or flirting, to take only a few possibilities—one useful answer is that interviews are symbolic situations. That is, interviews are situations in which: (a) the context of action asked about is not immediate; (b) the other actors and actions we often ask about are only present in imaginary and abstracted form; and (c) while the dynamics of the interviewer-interviewee relationship are immediately embodied, the situations we speak of are represented through the medium of talk (even if we come to partly re-live them in the telling [Pugh 2013]).

Thinking about interviews as such symbolic situations, however, does not necessarily mean that they are unique or even all that rare. As we return to below, such situations are becoming increasingly common. But even before the era of online-lives, symbolic situations were far from rare. Almost a century ago, Richard LaPiere (1934) pointed out that there are many contexts in our lives in which “situated action” looks a lot like an interview. While his paper was an early call for caution about inferring action from interviews, he was well aware that there were indeed situations where such inference was warranted.

As LaPiere (ibid, 235) noted, the case of voting, for example, is “exactly what the questionnaire can be justly held to measure; a verbal response to a symbolic situation.” When we ask someone about their political attitudes and who they would vote for, we take one symbolic situation to predict another symbolic situation. True, there is the gravitas of standing at the voting booth, the lines waiting for the vote, the excitement of the day, the smells and sounds, and even some problems of desirability bias for certain candidates and positions. But still, the act of voting is symbolic—e.g., we do not personally deport an immigrant when we vote for a candidate who promised to do so.

In other words, while open contexts usually point to social processes rather than attitudes and to the past and present rather than the future, that is not necessarily the case. The question of open, closed, and refractive contexts does not neatly map onto the temporality of past, present, and future. In many situations, asking about the past does not elicit anything resembling an open context (as the literature on collective memory attests to); in other situations, asking hypothetical questions can give us a decent sense of how action will unfold in the future.

Open contexts are a crucial aspect of the craft of interviewing. Much ink has been spilled on mitigating desirability bias and about taking care to word interviews in ways that elicit processual accounts—the difference between “why?” and “how?” questions. It is also the form of interviewing commonly employed by ethnographers looking to triangulate and complement their ethnographic data, as well as the mode of interviews employed by more process-driven sociologists.

Still, open-context interviews are only one kind of interviewing—and they have limits. First, there is no such thing as a purely open-context interview. You cannot ask how something happens without in part getting an answer about why it happened. More basically, if interviewers excise all but process and careful future-projection, they would have relatively little to work with. As we return to below, they would be left with very little of what makes interviews so compelling in the first place—the webs of meaning that people weave and within which they are suspended. The critical question is not whether interviewers should only ask open-context questions or if they should only use parts of their interviewees’ responses that could be analyzed in such an open way, but rather how to practice careful inference that avoids conflating inference from open and refracted contexts.

Closed Contexts: Ethnomethodology, Hypotheticals, and the Interview Society

Perhaps the most basic epistemological tension in interview methodology is that between realist and interpretivist modes of understanding the interview situation. Reacting to what they saw as an overly positivist image of the interview as a mirror of action, researchers such as Norman Denzin (2001) have called for a radical bracketing of the relationship between interviews and the situations interviewees speak of (see also Reed 2010). The interview is a space in which representations, fantasies, and categories are co-constructed—and is fascinating as such—but cannot be transposed beyond its boundaries. It is in this sense that we can think of it as a closed context.

Of course, as some were quick to note (e.g., Ho 2019), such a radical position both presents an impoverished sense of what interviews do and is seldom coherently held by researchers who claim it. If we truly believed that interviews are closed contexts, there would be little reason to be interested in this form of research—at least in our capacity as social researchers. Interviewers must assume that talk gets at something about the way people construct and think about their selves and the social worlds they inhabit. As we return to in the next section, most arguments by interpretive sociologists can thus be more productively understood as a call for thinking in refractive terms.

And yet, there are certain questions and types of interviews that do need to be considered closed contexts, or very nearly so. Moreover, there are cases where such closed-context inference is precisely what makes the interview interesting. These include two key approaches to interviews and kinds of interview questions.

Ethnomethodological Inquiry and the Interview Society

One mode of analysis that is explicitly closed includes studies that use interview situations to look at their ongoing interactional production. Inspired by conversation analysis and ethnomethodology, such studies trace how particular modes of asking questions elicit certain answers or silences, how interviewees interrupt and co-construct the interview, or how interactional repairs to the interview are made (see Baker 2002). To take one prominent example of this stream of research, Doug Maynard and Nora Cate Schaeffer produced a series of writings on survey interviews (e.g., 1997; Schaeffer and Maynard 2002), employing tools from conversation analysis to show how interviewers and interviewees negotiate participation in survey interviews as well as how questions are resisted in survey interactions (see Maynard et al., 2010).

Maynard, Freese, and Schaeffer (ibid), for example, analyze how the initial request for a survey interview is co-produced. Based on transcripts of the initial moments of telephone recruitment attempts for the Wisconsin Longitudinal Study (WLS), they show that the request to participate in an interview—like requests more generally—are dis-preferred actions. In order to even get to ask the prospective respondent whether they agree to be interviewed, interviewers need to modulate the interactional strategies. Maynard and his colleagues show that they do so through multiple conversational moves, including tailoring their requests based on cues from prospective interviewees or avoiding prospective interviewees’ possible blocking moves by starting the interaction in ways that require the respondent to react (e.g., saying “Hello!” and waiting for a response).

Although it draws heavily on the conversation analytic literature on requests in general, as ethnomethodologists, such research purposefully brackets the relationship between the interview and other situations. Inspired by institutional conversation analysis (Drew and Heritage 1992), the authors assume that given the recurring pragmatics of interview contexts, their findings would be generalizable to this specific realm of interview requests—but to this specific realm alone.

Carefully delimiting their results to the interview-context does not, however, make this line of research less powerful. In an era of marketing calls, surveys, job interviews, and speed dating, varieties of the interview situation are increasingly commonplace. Thus, even if it is only generalized to other interview situations, understanding the interview as an interview is important in itself, given how important the interview context has become (see, Igo 2007; Riesman and Benney 1956). As Atkinson and Silverman (1997) put it, we live in “the interview society,” a society increasingly marked by the notion that the interview—sociological or otherwise—provides a window into our interlocutor’s soul, and that such baring of the soul provides us with an “authentic” pathway to understand our fellow man. This, moreover, is not simply something we should think about as qualitative methodologists, as “the techniques of contemporary mass media and the interests of social researchers converge in the cultural forms of the interview society” (ibid, 309). We both experience more explicit interview situations and see some of the features and dynamics of the interview percolate into other social situations.

Ad Hoc Hypotheticals

A very different mode of questioning that needs to be thought of as a closed context—even as it pretends to be an open one—is fortunately rare in sociological interviews. This is the class of hypothetical questions that we cannot assume that the interviewee has needed to think about much in her life. In other words, these are artifacts evoked in an interview context through the interactional pressure to produce a reply, even when no prior attitude existed in the first place. An extreme example of such mode of questioning, brought up by Jerolmack and Khan (2014a, 2014b), is economists’ use of “contingent valuation interviews” as a way to put a price tag on non-monetary goods (see, e.g., Carson et al., 2001).Footnote 4 Questions about what we would pay to protect a National Park or the dollar-value of neighborhood common spaces are almost always the first (and last) time in which people need to think about these questions in monetary terms. Any assumption that such answers tell us something about other situations is exceedingly suspect. Even from the point of view of studying culturally-shared fantasies or ideal modes of valuation, it is not clear what we can infer from answers to such questions.

Of course, the fact that the answers to such questions are artifacts does not mean that these answers cannot have very real effects in the world. Returning to the notion of the interview society, Fourcade (2011) shows that the striking difference between the governmental response to oil spills in Alaska and France had to do with precisely such weak hypothetical questions. Whereas in France, a panel of experts assessed the damage to the environment, their American counterparts assessed damages by employing contingent valuation techniques, asking a sample of American citizens how important the area affected was for them in monetary terms to calculate damages. Unsurprisingly, contingent valuation yielded far steeper damages than expert valuation. From a methodological standpoint, then, the only reason that interviewees’ responses to such closed-context questions matter is that these responses are sometimes real in their consequences.

In short, there are rare moments in interviews that should truly be defined as hermetically closed—where we should not imagine the interview as a window into other kinds of situations. Approached with care, however, this does not make these analyses “superfluous” or uninteresting. Living in an interview society, such closed-context analyses may be important. Still, it means that as interviewers, researchers would do well to go through their interview transcripts and highlight “closed” moments to ward off weak inference.

Refracted Contexts: Landscapes and Promises

Finally, most of the data that interviewers elicit in their research can be understood as neither inferentially open nor closed. Even interviewers who are interested in the processual aspects of social life often find that some of their most interesting, evocative, and illuminating data cannot be analyzed in such ways. After all, precisely because they elicit “talk,” interviews allow us to glimpse what is perhaps the most human of characteristics: our ability to tell narratives about who we are and what we do. Representations, fantasies, metaphors, and narratives are not “noise”—they are a crucial part of the social world that people co-construct.

How do we analyze such interview data if we approach it seriously as talk? First, we must assume that the representations and narrative constructions that we get at through interviews are refracted indications of how our interviewees represent their world in other contexts. That is, while aspects of the interview are transposable to other situations, we cannot assume that we simply get at “what people think.” Second, most interview researchers need to better account for how such representations structure the unfolding of social processes as collective acts (Blumer 1969). That is, how talk enables and constrains future action because other people are listening and calibrating their own lines of action in anticipation.

Landscapes of Meaning

We begin with the symbolic constructions and narratives that emerge in interviews. After all, as Lamont and Swidler (2014, 157) put it, in-depth interviews are primarily tools designed to get at people’s “representations, classification systems, boundary work, identity, imagined realities and cultural ideals” rather than to predict future action (see also, Rinaldo and Guhin 2019). For many in-depth interviewers, it is not the relationship between talk and non-discursive action that is directly at stake, but the relationship between symbolic structures elicited in the interview and how interviewees talk (to themselves and to others) about their world in their daily lives.

Thus, although researchers who focus their attention on cultural representations are generally sympathetic to seeing the interview as an active construction, they do not think about the interview as a closed context. Indeed, when Ho (2012; see also 2019) criticized Gubrium and Holstein’s work on the grounds that such radical constructionism makes it impossible to say anything about the world out there, the authors (2012) responded that they had never claimed there was no relationship between the interview and other contexts; while the interview may be an active construction (Holstein and Gubrium 1995), the different contexts were intertwined.

Yet, what exactly Gubrium and Holstein (and others) mean by such intertwining is less clear. On the one hand, while constructionist researchers may talk about the interview as an active construction, they seem to implicitly agree with Howard Becker (1996), who noted that researchers suffer from hubris if they truly think that people act and talk in radically different ways when we study them than they do when they are with others. Even in the context of an interview, people have better things to do than craft a completely new narrative for the interviewer’s benefit. And yet, looking again to Becker, we also know that people talk differently in different contexts (e.g., Becker and Geer 1957). Despite living in the interview society, an interview is a peculiar situation. Even if we don’t think that people are dishonest, it may still be that the interview situation elicits narratives that are different than the ones they would construct in other situations. How, then, do we reconcile these contending insights?

I suggest that a useful way of thinking about this tension is through the felicitous metaphor developed in Isaac Reed’s (2011) Interpretation and Social Knowledge—that of culture as a landscape of meaning. Insofar as in-depth interviews provide us with some sort of window into cultural sense-making, they can show us the general contours of the landscape upon which action takes place. These contours do not constrain only what elements we put in our landscape but also how we go about constructing it—that is, the kind of picture that we tend to construct, and how connections between elements of the landscape are organized. Even if we bracket the question of our landscape of representations’ causal power in shaping action, thinking about the relationship between interviews and cultural representations in terms of such a landscape is useful because it alerts us to different possible relationships between the representations that interviewers elicit and the narratives interviewees tell in other situations.

In other words, while few would argue that the narratives interviewers elicit have a simple translation into action, few would also argue that the narratives that interviewees construct during an interview do not tell us anything about how they construct narratives “in the wild.” After all, while interviewees’ narratives may be prompted in a general way (“tell me about X”), the drawing on the canvas is mostly their own. For example, as Michèle Lamont shows in cross-national studies of boundaries and respect (Lamont 1992; Lamont et al. 2016), people in different countries evoke different maps that locate both themselves and others in different positions in symbolic space. Moreover, these differences are patterned in ways that cannot be chalked up to the interview situation. That is, even if people in both France and the United States imagine moral life and boundaries differently in different situations, the symbolic boundaries (see Lamont and Molnár 2002) evoked in interviews provide useful points of reference for understanding some of the distinctions between the two national contexts. Here, the interview acts as an anchoring device in a methodological sense; it does not depict an exact representational landscape that we expect to replicate in any given case, but we expect that it does give us some notion of the kind of narratives that make sense in our interviewees’ world.

More specifically, by eliciting representations and narratives, researchers can identify structural aspects of interviewees’ landscapes of meaning. For example, Ann Swidler (2001) shows that people in the United States oscillated between two ways of talking about love over the trajectory of a relationship, and in relation to pragmatic pressures and affordances. The cultural repertoire of love meant that people experienced “Gestalt switches” between culturally-available alternatives: that of romantic ideals and that of more prosaic partnership. That love could be understood alternatively through the prism of cinematic romance or the mundanity of washing dishes allows researchers to study some of the structural elements of representation. While Swidler cannot know how people will act in specific situations, she identifies a repertoire of meanings, which interviewees have ready-to-hand “at their fingertips.” As Cerulo and Ruane (n.d.) show in their work on the cultural underpinnings of “American dreams,” while we may be skeptical of the precise content of the dreams that their interviewees constructed, they share key “grammatical” elements. Whether people wanted to be the president, an astronaut, a famous scientist, or start a home for the homeless, they stressed that it would come through hard work, and they were the key—often, only—protagonist of their dreams. There may be a multitude of American dreams and they may change situationally, but some elements of the cultural landscape anchor them in predictable ways.

In short, narrative and symbolic representations matter as a way for humans to construct their world together. Eliciting such a cultural landscape, interviews tell us something about how people make sense of their world well beyond the interview situation. Of course, interviewers need to know enough about the people they study to get a sense of whether the talk elicited in interviews resembles how the people they study talk in other situations in their daily life, as well as whether they even talk and think about these questions in such other contexts.Footnote 5 Yet, as long as we do not assume that this is the precise way that people talk in other situations—which is a fallacy to begin with, since it assumes there is one “true” way that people talk, rather than a myriad of slightly different constructions—we can learn about cultural landscapes in comparative perspective, about the kinds of narrative shifts that make sense to our interlocutors, and about the general grammar of representation.

From Individual Action to Collective Acts: The Promissory Structure of Talk

If the first mode of inference remains on a representational level, a second mode of thinking about refractive contexts takes us more explicitly back to the question of talk and action. As we noted above, there is an “agency bias” that haunts interviews—the implicit construction of interviewees as the key protagonists of their own lives. Yet this bias is perhaps just as prevalent in most theorizations of the relationship between talk and action. Both for those making and those questioning the connection between talk and action, it seems that the key mechanism they have in mind is a more-or-less fuzzy notion of individual motivational power that is then mediated by external situational pressures. That is, if we say that we will do something at point t1, we will tend to do it when moment t2 arises—unless the situation is so extreme that it trumps such motivations. To take a simple example, if the interviewee says that she is vegetarian at t1, she will probably not order the steak at t2 when presented with a menu at a restaurant. Since vegetarianism is important to her (or so we assume), she will enact her preferences when the situation arises.

This kind of mechanism is, of course, important. Denying these kinds of links between situations—and the extension of the self through time—would be falling into a radical situationism that few believe in.Footnote 6 And yet, thinking in these terms overlooks a point that sociologists routinely make elsewhere when they theorize social life: rather than the sum of individual choices, social life is constructed and patterned through a collective act. Thinking primarily in terms of the extension of self through time ignores how the things that we say and do at t1 shape the processes that lead to t2—how talk sets up the ways that different actors orient and construct the situation. Talk does not just leapfrog from one context to another, but may often influence the processes that lead to it.Footnote 7

But how do we translate these misgivings into a better fleshed-out theory? To do so, I propose we return to the classic study that set the tune for much of the debate about interviews and inference: LaPiere’s (1934) “Attitudes vs. Actions.” The study’s design bears repeating. In the context of post-WWI racism, white-owned businesses often avowed that they would not serve certain “racial” groups—not only African Americans, but also people of Chinese descent. LaPiere’s insight was that it was generally easier to be racist in the abstract than to practice open discrimination. Situated action is messy and refusing service may turn out to be harder in the flesh than it is in the abstract. LaPiere’s insight is thus similar to that of Pager’s (e.g., Pager and Quillian 2005), although its predictions are inverted. In a world where equality is expected, discrimination may be missed in interviews; in a world where racist attitudes are expected, people may discriminate less than interviews would lead us to believe.

To test this relation, LaPiere traveled across the US with a young Chinese couple he had hired, visiting 251 hotels, restaurants, and camping sites. Apart from one hotel, all the establishments welcomed the Chinese couple. Then, months later, LaPiere phoned the very same establishments asking if they would accept a “very important Chinese gentleman.” Over 90% said they would not, while the others said that it depended on particular circumstances. Only one said they would, citing LaPiere’s accomplices’ visit. If Pager found that people were more racist in their hiring practices than in interviews and questionnaires, LaPiere found them to be less racist.

This study—pithy, beautifully written, and insightful—is referenced by Deutscher (1966, 1973) as well as Jerolmack and Khan (2014a). However, reading it carefully raises an interesting problem. As mentioned above, it assumes that the relevant nexus we should pay attention to is the tension between stated individual attitudes and situated action, with its pressures and idiosyncrasies. The article’s moment of revelation is the inconsistency between them. Yet this ignores a simple problem: calling a hotel and asking about their policy does not simply elicit an “attitude.” Rather, it elicits a speech act.

As Austin (1962) and Searle (1969) have shown—and others have come to take for granted—words do things in the world. This is true not only for speech acts’ obvious “illocutionary” and “perlocutionary” aspects, meaning the changes in the world wrought by pronouncements such as “I now pronounce you husband and wife,” and the reactions such words trigger; it is also true for seemingly declarative, and even descriptive, enunciations. Any description of the world simultaneously constructs it (Berger 1967). Importantly, since the first tracts on speech act theory, promises (or “commissives”) were seen as a paradigmatic case of speech acts. That is, by saying that I would or would not do something, I make a certain commitment that has “sincerity conditions” attached to it (Searle and Vanderveken 1985). Whether or not I actually would do it later if placed in that situation is obviously important, but that I had explicitly or implicitly promised to do so is no less important.

This leads us back to the case of LaPiere. What do people do when they say that they will not serve someone who looks Chinese? In a nutshell, they make a promise (or, more darkly and accurately, a threat [see Salgueiro 2010]). They announce how they would act. And, at least for the moment, we can assume that they themselves believe what they say—that they fulfill the “condition of sincerity” that makes speech acts viable. In other words, as Goffman (1959, 1967) has pointed out, although people are not naïve, for the interaction order to be sustained, they need to generally assume that others are who they say they are and that the “line” they hold in interaction is genuine. Following this line of thinking, we can imagine a Chinese-American couple planning a romantic holiday but knowing that some places claim that they do not serve “their kind.” The question is not simply whether or not the worker in that establishment would or wouldn’t turn the couple away if they show up. The question is rather, given talk’s promissory structure, would the couple risk it? The answer, obviously, is that most people wouldn’t. Thus, the question of whether these establishments would or wouldn’t actually serve Chinese-Americans in contexts of situated action becomes moot.

Of course, this is not merely a hypothetical question. Between 1936 and 1966, a publication called The Negro Motorist Green Book was circulated in the United States, used by people of color to decide where they would stop for the night in the era of Jim Crow. The list in the Green Book was not based on trial and error, but on precisely the kinds of calls that LaPiere did. Interviews, in this case, were objectified into a guidebook. It may very well be that some motorists of color decided to risk it in establishments not listed in the Green Book and were admitted, but we can safely assume that many did not—evidenced by the fact that the Green Book, which started as a few-page brochure, quickly became a publishing success until the civil rights movement made it obsolete.

This complements Vaisey’s (2014, 229) argument that talk is part of a selection process into particular situations, yet it also turns it on its head. The connecting mechanism between talk and situated action is not the individual acting according to their beliefs and choosing their situations and networks accordingly, which then have a feedback effect on behavior (Vaisey and Lizardo 2010). Rather, talk sets up a collective act (Blumer 1969; see also Tavory and Timmermans 2013) that ends up validating the actor’s “line” and the implicit promises made in talk. By not going to the places that avow to deny service to Chinese couples, they confirm the establishments’ racist promise. To take the more banal example we used above, if someone says that they are vegetarian at point t1, we would tend not to serve them a steak to see if they “really meant it” when we invite them over to dinner.

Recognizing that talk is also connected to situated actions because it structures the collective act adds an interactional-level layer to an individual-level mechanism—but why does it matter? What would be the empirical correlates of such a causal nexus? I suggest that there are at least two distinct upshots of thinking in these terms.

First, even more than when researchers are focusing on landscapes of meaning, this perspective encourages careful consideration of whether and when the narratives that researchers elicit are narratives that they expect the interviewee would produce in other settings. Simply put, the more public the act of talk, the more the researcher could assume that it is likely to structure the collective act. This is a spectrum. On one end are artifacts such as The Green Book or even a sign on the door (e.g., the Nazi-era sign “no entry to Jews and dogs”); on the other end of the spectrum are things we tell nobody except perhaps the interviewer.

Most talk that takes place in interviews probably falls somewhere in between these extremes—neither objectified into publicly-available artifacts, with the scaffolding of action such artifacts imply (see Jerolmack and Tavory 2014), nor secrets the interviewee tells no one else. Still, asking where such talk may emerge in actors’ everyday conversations is crucial if we are to understand the construction of the collective act. Once the “promise” is known, people will help the individual set up interactions that do not challenge the context of talk, and thus defend their line of action (Goffman 1959, 1967). Talk leaves its traces, objectified both as generalizations about the future (Ricœur 1973) and as promises about future action in the sense elaborated above. The relationship between talk and action in other situations is often relegated to others.

Second, this means that in order to productively think about the relationship between what people say they would do and what takes place later on, researchers need to take into account the interpretations that others are likely to give to such talk. Individualistic causation assumes that it is the actor who protects her cherished preferences and acts upon them when the time comes. But if researchers take the collective act as a key causal structure, the questions they ask must shift. Rather than asking how salient a narrative or preference is for the individual, the issue becomes whether others will be more likely to set up the situation to defend the actors’ line if they infer that such talk is salient for the person enunciating it. Thus, the way the person enunciates their talk matters. An off-hand comment made as part of the flow of interaction will likely trigger less recalibration of situations from other actors; promises that are made in ritual contexts will matter far more. It is not primarily talk’s salience for the actor that matters, but the assumptions about that talk made by others.

Lastly, yet again, we return to the notion of the interview society (Atkinson and Silverman 1997; Gubrium and Holstein 2002), but an interview society that is mediated by the structure of new forms of communication. It is not simply that more and more situations in our lives are mediated by symbolic contexts—we buy our books from Amazon, we chat with friends and find romantic partners through online platforms. Rather, it is that people spend more time in symbolic media where our representations and soliloquys are there for all to see. As our implicit promises become increasingly public, we can expect others to modulate their actions to fit the image of self that we broadcast. In a world where people are constantly announcing the personal, we may be moving towards a better correlation between interviews and situated action.

Making Sense of Interviews

Rather than postulating the one right way to make inferences from interview data, this article made the case that it would be more accurate to think about the interview as a complex situation that calls for different modes of inference at different moments. The reason we cannot tell an interviewer how to analyze their data in advance is that different parts of their transcript will warrant different modes of inference. Some moments in the interview are cautiously open, as when we make inferences about symbolic situations or when we try to get at processual questions—the how, with whom, and when of social life. Some rare contexts are hermetically closed. And, perhaps most commonly, researchers need to approach much of the data they generate in interviews as a refracted representation or a speech act distributed in a web of collective activity. The craft of inference from interviews (and, indeed, part of the craft of interviewing more generally) is to know which part of the interview calls for what kind of inference.

This, of course, is partly a problem of research design. Some questions are more likely to elicit certain kinds of answers (e.g., questions that start with “why do you think that…” almost invariably elicit refracted contexts). Still, even the most careful interview design cannot avert the need to read our transcripts carefully and figure out the context of inference, sometimes line by line. Starting from that methodological dilemma, this paper is meant to be used as a tool for coding and analyzing interview transcripts. Ideally, when interviewers read through their interview, and especially when they decide they want to use an excerpt in the emerging analysis, they could use the typology provided in this article as an orienting inferential device.

On a second level, this article is meant as a broader contribution to the way we treat the relationship between situated actions and interviews. The argument here is inspired by cultural sociology, interactionist sociology, and speech act theory. Drawing from cultural sociology, I argue that while representations and narratives will fluctuate among situations, attending to the structure of the landscapes of meaning that people produce provides a useful anchoring device for studying both the kind of things they may say in other situations, and the structure of the narratives they may produce. While people say different things in different situations, interviewers can expect the relationship between elements in interviewees’ repertoire and common aspects of these constructions to be similar to those elicited in the interview. Drawing from interactionism and speech act theory, I argue that analysis of the relationship between interviews and situated action has been stymied by an unexamined individualism. Whether they argued for or against inferences about situated action from interviews, researchers have asked far too much about the person talking rather than about how talk generates a collective act by propelling other people to “save” the line that the actor presented at an earlier time. Interviews are individualizing by their very nature, but talk seldom is. If we are to tackle the question of coherence among contexts of action, we need to think about the way in which others act to construct a world that does not challenge talk.

Thus, although rooted in a methodological puzzle, this paper is also meant as a contribution to our thinking about the way in which situations are strung together. As sociologists, much of our data comes from a limited amount and kinds of contexts. Even ethnographers, who are somewhat more agile in their research, never move freely through people’s lives (Trouille and Tavory 2019). With the exception of the most micro-oriented situationists, sociologists must constantly make assumptions about how one situation is tied to others and how closely they hew together. For both methodological and theoretical reasons, it is thus important to pay more careful analytic attention to the relationship between contexts, along with their moments of coherence and inconsistency. This paper is a step in that direction.