The purpose of this article is to lay out a set of analytical tools for qualitative researchers who wish to fictionalize for ethical reasons without sacrificing rigorous attention to empirical data. I will address the epistemology and ethics of composite character narratives in sociological writing. In such a narrative, fictional characters represent aggregations of multiple real-world people who informed research. The author remixes data from multiple study participants to tell a coherent story of a character that could have, but did not, exist in the real world.

Viewed from the lens of postpositivist analytical sociology, when an author combines information from multiple real people into a single composite, they risk multiplying problems of data selection and misrepresentation inherent to social science generally. However, an author can also use the compositing process itself as an analytical exercise to systematically consider patterns and variations in data and their significance.

I will first explain how fictionalization can help protect research participants’ anonymity, which was the original reason I developed what I call the Sequence-Based Composites (SBC) method of sociological writing. I will then lay out ways that sociologists use fiction and the various interpretations and implications of claims that social science can be, should be, or is inherently, fictitious. This discussion will situate the method under discussion within a tradition of postpositivist analytical sociology, which occupies something of a middle ground between naturalism and postmodernism in disciplinary debates about the possibilities and goals of social research.

I then go on to detail a method of constructing composite narratives that prompts their author to abductively construct theory and interrogate the causal logics of participants’ experiences. I explain the SBC method with reference to four criteria for evaluating social scientific arguments from an analytical perspective–reliability, reactivity, representativeness, and replicability–taken from Katz’s (2015 [1983]) “Theory of Qualitative Methodology.” For examples of SBC character construction, I draw on my research on international news coverage of Turkey.

Fictionalization and Anonymization

From 2014 to 2016, I conducted ethnographic fieldwork in Turkey on the roles and strategies of “fixers”: local interpreters and guides who assist foreign reporters. This research provided the basis for my book Fixing Stories (Arjomand, 2022). Fixers introduce reporters to sources, do research and monitor media in local languages, and generally mediate between local societies and their clients’ cultural scripts and professional interests. The following research questions underlay my research: Who becomes a fixer, and how? How do fixers shape the news stories to which they contribute? How do fixers manage their relationships with client reporters and local sources? In answering these questions, I sought to identify and explain variation among study participants and over the course of participants’ careers.

Not all research poses enough risk to research participants to merit anonymizing them in the text (Jerolmack & Murphy, 2017). A growing number of ethnographers such as Murphy, Jerolmack, and Smith (2021) argue that identifying participants makes ethnographers more accountable to readers and makes their texts more useful to future researchers as data sources. I support this move toward transparency when it is safe for participants.

In my case, though, identifying real individuals with their statements to me and my observations of them at work could expose some of my interlocutors to professional, legal, and even physical harms. Some of my study participants behaved in ways that violated standards of journalistic ethics; some were highly critical of, and acted in ways antagonistic to, armed groups or Turkish or Syrian governments. Journalists have been jailed at alarming rates in Turkey and killed at appalling rates in Syria in recent years. My writing could have put them in further danger.

Ethnographic research methods are useful because of the wealth of situated social data they can provide about observed and interviewed people’s lives. Publishing some of that information might, however, harm study participants if readers are able to identify them individually. Simply changing names may not be enough to protect participants’ anonymity. When you change a name but keep a character and chronology of events intact, a third party need only know one identifying thing about that participant to deduce much other private information about them (Saunders et al. 2015: 617–621; Van den Hoonaard 2003).

In the case of my study, reporters and fixers produce publicly available work, track their colleagues’ and competitors’ reporting, and maintain large social networks. I recruited participants to my study in part using a “snowball” methodology–I asked my contacts to introduce me to their contacts–and so many of my study participants knew each other. These factors meant that if I were only to change names, individuals and news organizations could easily be unmasked based on descriptions of their persons, career trajectories, or the news stories they produced.Footnote 1 A reader could deductively attribute everything else I wrote about a character in the text to the real-world research participant they identified.

To prevent deductive unmasking, some ethnographic researchers scramble details and chronology, usually with the unelaborated claim that these changes are not significant (Jerolmack & Murphy, 2017: 9–11). Compositing takes such scrambling to its logical conclusion: remixing reality at each step of the story. A third-party reader may recognize an individual research participant in the ethnographic text, but that does not reveal anything new about the individual: the other actions attributed to that character could have been the deeds of others (Willis, 2019: 476–477).

Fiction and Social Science

But how should we classify the resulting story? It is a fiction insofar as its characters no longer correspond directly to real individuals with whom the researcher interacted. Can it still be a work of social science?

I would contend that sociological fiction is actually commonplace, albeit with great variation in form, motivation, and method. Ethnographic novels, parables that illustrate theoretical arguments, research-based theatre, interactive games and simulations, and auto-ethnographic poetry are among commonly recognized fictions that sociologists have deployed (e.g. Gamson 2009, Kontos and Naglie 2007: 800–802; Leavy 2013; Saldaña, 2003: 219–220). Quantitative research also produces fictional forms such as mathematical models, though these are less often labeled as such (Becker, 2007: 150–166).

However, there is no consensus as to the hows or whys of fictionalization as a tool of sociology (Atkinson and Delamont 2005: 821–824; Denzin 1997; Gubrium & Holstein, 1997). The range of views about the de facto and desirable relationship between fiction and sociology run the gamut from positivist to radical postmodernist schools of thought. It is worth presenting a brief overview of how these different sociological traditions have conceptualized fiction as a part of research before more deeply exploring the place and problems of fictionalization in the postpositivist tradition, a middle ground between realism and constructionism where the SBC method finds a home.


In the positivist mode of inquiry–also referred to with slightly different emphases as naturalist, realist, or empiricist–the sociologist’s job is to observe and record the social world as accurately as possible and then use that gathered data to generate theories about causal relationships among phenomena (Gubrium & Holstein, 1997: 19–37; Van Maanen 1988: 45–72). Only the most naïve of empiricists might claim, however, that a “just the facts” approach to sociology is straightforwardly possible.

Though positivists might bristle at the application of the “fiction” label to their work, when they move from collecting data on specific real-world phenomena to describing, classifying, and comparing social facts, they necessarily enter the realm of fiction in the broad sense of representing people and events produced from an author’s (sociological) imagination rather than as they are individually observed. Even positivists use fiction as a thinking tool, to compress and organize data, and to operationalize social facts that are not directly measurable. The very concept of the average member of a group is a fictional device that sociologists use as a tool to make social facts comprehensible and comparable. Statistical analysis creates fictional average members of groups invented by the sociologist to compare with one another, with precisely calculated differences between fictional averages not necessarily corresponding to any real individual’s experience in the world. Every sociological argument of causation employs–implicitly or explicitly–counterfactual “what if” fictions (Weber 1949). We imagine a parallel world identical to the real one except for one small difference and then imagine how things would have turned out differently, perhaps using data from a control group or comparative case as an approximate stand-in for the fictional alternative reality we imagine (Griffin, 1993: 1101).

These forms of fiction are necessary to compress complex social realities into manageable forms and to allow sociologists to make intelligible arguments, compare phenomena to other cases of the same type, or to create a formal theory that explains more than the case at hand (Becker, 2007: 2–25; Vaughan 2012). Fictionalization, in the broadest sense of abstraction from the concretely observable, allows the compression of data into a neatly expressible graph, table of statistics, or even composite story claimed to represent the typical experience of a group under study.

The reason for such exercises in fictionalization is, to positivists and especially to quantitative sociologists, fundamentally different from literary fiction lies in the methods for producing general propositions and comparisons among phenomena. Quantitative researchers follow standardized procedures for constraining their selection and rejection of data and for combining multiple observations into aggregated means, medians, modes, and standard deviations. They have developed statistical measures of significance that distinguish signal from noise and techniques to control for the influence of factors outside of the scope of their studies. Some qualitative researchers, including authors of composite character narratives discussed below, try to follow suit and imitate their quantitative colleagues by standardizing interview procedures, collecting randomized or purportedly representative data samples, and representing their findings in the form of charts and tables that aggregate data from multiple cases.Footnote 2

Radical Constructionism

On the other end of the spectrum, since the “crisis of representation” in the 1980s, some qualitative researchers have shifted to embracing a radical subjectivist doctrine according to which we live in a “hyperreal” world composed of discourses and images, none of which have greater validity than others in the sense of verifiably corresponding to some external reality (Baudrillard 1994 [1981]; Gubrium & Holstein 1997: 75–94; Atkinson and Delamont 2005). The critique of positivism present in some articulations of social constructionism, postmodernism, and poststructuralism “challenges even the possibility of empirical research” by rejecting scientific (or scientistic) aspirations to base theories on real-world observation, parse causality, or overcome subjective bias (Gubrium & Holstein, 1997: 97).

Among postmodernists skeptical of any version of empiricism, the goal of social research shifts from striving for objectivity to embracing subjectivism, from description of the social world to transformation of it. Representations labeled “fiction” can serve the goal of emancipating readers from false consciousness, amplifying silenced voices, or building new forms of community at least as well as those labeled “non-fiction” (Denzin, 1997; Denzin and Lincoln 2005: 11–13 Leavy 2013: 22–25).Footnote 3 In this view, claims that some representations of society are “non-fiction” or “social scientific” and so fundamentally different that those labeled “fictional” or “artistic” are merely rhetorical and political attempts to defend hegemonic systems of knowledge, along with the status and symbolic capital of scientists, and delegitimize the knowledge of others (Rhodes and Brown 2005: 474–478).

While I am sympathetic to radical constructionist critiques of naïve objectivism and dogmatic scientism, I am concerned that the skeptical postmodernist turn to subjectivism and aestheticism may encourage researchers to double down on received wisdoms and repertoires of their sub-fields rather than cultivate an openness to inconvenient data. I share Atkinson and Delamont’s (2005: 834–835) worry that approaches that do not methodically test theories against empirical observations “create closure by creating a new basis for authorial privilege… Authors distort or obliterate the cultures they seek to account for if they translate everything into their own culture-bound aesthetics.”


Interpretivist, structuralist, ethnomethodological, and pragmatist approaches, which I will discuss here under the blanket term “postpositivism,” occupy a middle ground between naïve realism and radical constructionism. Broadly, postpositivists have argued that researchers can never enjoy access to the realities of the social world unmediated by the scientific paradigms, cultural worldviews, and social positionalities that they carry with them into the field. They can never interpret or report their findings unfiltered by pre-existing conventions, narratives, and theories that provide frameworks for intersubjectively understanding the world (Becker, 2007; Bourdieu and Wacquant 1992; Kuhn 1970 [1962]). Furthermore, sociologists are not flies on the wall of those they study but active participants in those people’s social worlds. Sociologists affect through, and are affected by, their acts of observation.

Ethnomethodologists and students of science, technology, and society view social science as a form of collective work to construct a shared sense of reality (Gubrium & Holstein, 1997: 38–56; Latour 2007). Social science is, in this view, always an exercise in storytelling, a literary genre as much as a measurement or representation of reality. In postpositivism as articulated for example by White (1980: 27), social scientific and historical representations of the world can only be coherently made through fiction: “value attached to narrativity in the representation of real events arises out of a desire to have real events display the coherence, integrity, fullness, and closure of an image of life that is and can only be imaginary.” Without imposing meaning, order, and closure on events to draw them into a fictional unity that fits within received storytelling conventions, the historian or sociologist is writing an incoherent transcript (White, 1980: 10–27; Kotišová 2019: 286, 295–296).

Where the postpositivists diverge from radical postmodernists is in their stance that there are key differences between literature and social science, not in the fact of fictionalization, but in the how and the why of it. Geertz (1973: 15) writes, “Anthropological writings are. . fictions, in the sense that they are ‘something made,‘ ‘something fashioned’ —the original meaning of fictiô —not that they are false, unfactual, or merely ‘as if’ thought experiments.“ (1973: 16) Geertz writes that the important difference between Gustave Flauber’s novel Madame Bovary and an ethnographic account of a Moroccan named Cohen “does not lie in the fact that [Bovary’s] story was created while Cohen’s was only noted. The conditions of their creation, and the point of it…differ. But one is as much a fictiô —‘a making’— as the other.”

Not all representations of society are equally valid in their claims or equally successful at explaining phenomena. Some stories and explanations maintain better fidelity than others to available data and stand up better when applied heuristically to other cases (Kuhn, 1978: 321–323). Empirical observations still play a central part in postpositivist knowledge production as bases for, and checks on, the reliability of researchers’ creative imaginings of the social world. Through successive “conjectures and [evidence-based] refutations” (Popper, 2002 [1963]) —or in pragmatist terms through a continual process of abductively searching for new data and methodological innovations that could challenge their tentative theories (Tavory and Timmermans 2014) —social scientists pursue the ultimately realist goal of analytical sociology: to enrich our understanding of social phenomena (Jackson, 1995: 163; Denzin and Lincoln 2005: 11; Hedström and Bearman 2011; Vaughan 2012).

Though sociologists can never work outside of culture to gather or transmit knowledge that transparently reflects external realities, they can cultivate objectivity in the sense of “a radical enlargement of the range of attention” to consider inconvenient data that force them to revise their theories and stereotypes (Lippmann, 1982: 157). A researcher’s critical self-awareness of and reflection on their own subjectivities and repertoires are important means to this end (Lippmann, 1922; Bourdieu and Wacquant 1992; Galison 2015). Unconventional methods of gathering or disseminating knowledge, including fictional forms such as ethno-theatre, allegorical writings, or roleplaying simulations, can serve the goal of enlarging researchers’ and their audiences’ ranges of attention. Advocates and practitioners have argued that such projects can better incorporate and more effectively communicate forms of data and perspectives that conventional academic writings and dominant cultural narratives leave out, such as emotions and subaltern modes of expression (Gulbrium and Holstein 1997: 71–74; Saldaña 2003; Gamson, 2009; Rhodes and Brown 2005: 472–473; Kontos and Naglie 2007: 801–802; Leavy 2013: 31–33; Kotišová 2019: 294–295). Adding fictional and other unconventional works to the scholarly corpus in turn allows secondary researchers to triangulate among richer and more variegated information sources to better interrogate their own emergent theories (Denzin and Lincoln 2005: 4–6).

It can also be an important ethical stance, from a postpositivist view, to embrace the “fiction” label as a way to remind readers of the creative and tentative nature of sociological contributions (Kotišová, 2019: 292–294). As Rhodes and Brown (2005: 470, 474) argue, whereas positivistic researchers tend to use facticity and objectivity claims to assert scientistic authority and avoid scrutiny of the cultural mediations inherent to their work, “the explicit use of fictional forms draws attention to the embedded fictionality in/of all research writing” and acknowledges that “researchers are responsible for their textual choices in selecting and emplotting narratives.”

Analytical License from a Postpositivist Perspective

The SBC method fits within the postpositivist school of thought, which acknowledges the limitations of empiricism and scholarly representation without abandoning the mission of systematically building and interrogating empirically grounded theories that help us make sense of the social world. Within the tradition of analytical sociology, though, the fiction of a composite character narrative seems detached enough from empirical data to raise questions about its validity as social science. The challenge from this perspective is to write fiction that is both sociological: connecting individual personal experience to larger social patterns and trends (Mills 2000 [1959]); and scientific: following a systematic logic based on differences that are observed in the world and theorized to be distinct in their causes and effects.

It is worth expanding upon Geertz’s (1973: 16) discussion of the difference between a novel and an ethnography. Literature and empirical social science are both “makings” but have different criteria for inclusion of content into a story. Literature selects or invents content based on dramatic rather than analytic license. “Drama,” Alfred Hitchcock told fellow filmmaker François Truffaut, is “life with the dull bits cut out.” (Truffaut & Scott, 1985: 103) Analytical sociology is life with the causally irrelevant bits cut out.

If Hitchcock had made a film about NASA’s 1986 Challenger space shuttle disaster, he would almost certainly have devoted more time to the sensory experience of the astronauts inside the space shuttle than does Vaughan (1997), whose sociological book on the subject ignores that dimension of reality because the astronauts’ feelings did not cause the erosion of O-rings in Challenger’s solid rocket booster. Vaughan (1997: 287–326) devotes more attention to the faxing of data charts than Hitchcock would have because the mode of transmission of information from Thiokol engineers to NASA headquarters, however dull, did have a causal effect on the launch decision that led to the disaster.

A postpositivist sociologist’s analytic license allows her to ignore some data and highlight others in order to create a world for the reader in which research subjects’ lives are simplified to the point of existing to provide evidence for the author’s causal argument. When realist and postpositivist social scientists discuss methodology, much of the debate concerns defining the boundaries of analytic license and questioning the assumptions that underlie authors’ simplifications.

Selections of content made according to the logics of dramatic vs. analytic license often coincide. In a naturalistic fictional story, it is in part relatability–that is, the ability for readers to put themselves in fictional characters’ shoes or think that given the situations that they wind up in (however foreign or fantastical those circumstances), those characters’ actions are comprehensible and plausible–that draws readers to keep turning the pages of a novel and to care what happens to the protagonists (Keen, 2006: 214–216, Becker 2007: 247–251).Footnote 4 Relatability, comprehensibility, and plausibility are all based on comparison to data that readers have unwittingly been gathering their whole lives about how society works.

Selecting data based on aesthetic appeal can meanwhile have analytic utility. Katz (2001, 2002) argues that the “neglected practical wisdom” embedded in gut feelings about what makes for a “luminous description” or poignant moments often reflects good causal reasoning. Drama may clue us in to relevance and provide useful windows for making causal claims. Katz (2001: 467) goes so far as to argue, “Ethnographic data should systematically over-dramatize the colorful character of people, their settings and their conduct, for reasons rooted in the logic of empirically grounding and testing causal explanation.”

To consider whether the composite character narratives drawn from my research in Turkey and their inclusion of dramatic elements fall within the bounds of analytic license, we can measure the conditions of their creation against criteria proposed by Katz (2015 [1983]: 131):

Qualitative studies may be guided and evaluated for [1] their power to warrant generalization, [2] their success in transforming researcher-subject interactions into resources for improving substantive explanations, [3] the constraint they place on bias and arbitrariness in data interpretation, and [4] their promotion of the re-testing of findings.

These criteria correspond to the “4 Rs” according to which positivist and postpositivist social scientists across numerous sub-fields assess the validity of research (Katz, 2015 [1983]):Footnote 5

  1. 1.

    Representativeness: Do the data represent more than just the particular cases and individuals under study such that general conclusions can be drawn?

  2. 2.

    Reactivity: How has the intervention of the researcher affected the very behaviors the researcher seeks to study?

  3. 3.

    Reliability: Does the study use research instruments that measure consistently and systematically?

  4. 4.

    Replicability: Can findings be confirmed by others?

In the rest of this article, I will explain the Sequence-Based Composites (SBC) method of fashioning ethnographic data into composite narratives, and in the process reflect on how well the method allows me to check off each item on the above list as compared to authors using other methods for representing the social world. I will switch the order of items to 3»2»1»4 for the sake of clarity in walking you through the SBC analysis and writing process.

In short, I optimize for reliability by making myself accountable to empirical data through a systematic method for compressing study participants into characters and remixing data into narratives (Criterion #3 above). I factor in reactivity (#2) and incorporate it into my analysis by considering myself as a participant in the world of fixers and reporters; I accordingly add myself to the pool of real people clustered into composites. I produce arguments that are as representative as possible (#1) by maximizing the variety of data I include and measuring composite characters’ behaviors against ideal types in order to generalize about the effects of both macro-structural and situational factors. My findings are replicable (#4) in that readers can and should test their utility as theoretical tools with which to interpret other cases, which the label “empirical fiction” encourages.

Reliability: Clustering Participants and Building Narratives

Does the study use research instruments that

measure consistently and systematically?

There can be legitimate reasons for social scientists to invent stories from whole cloth, discussed further in the section on replicability, but I did not do so in my narratives of news fixers. The study was grounded in real data, which served as a helpful constraint that forced me to continually rethink my theoretical arguments. Use of real data is not sufficient to build a reliable argument, though.

All social scientists face the problem that real data can be cherry-picked in order to “prove” the author’s pre-existing assumptions and biases (Fassin, 2014). Rich, real details can be deployed to create a rhetorical “reality effect” that convinces the reader that a story is true to life, even if the preponderance of evidence left out of the story points to a different interpretation of the events described (Barthes 1989; Dobler, 2019). In short, an untruthful argument can be factual (Frankfurt, 2005 [1986]). Not just reality but the method by which it is selected and interpreted is important for the reliability of a study.

Qualitative researchers have a couple other options to make their findings reliable through accountability to data drawn from the real world. The first option is external accountability: empowering the reader to check your work, to determine for themselves whether you cherry-pick or misinterpret data and so whether your theories derived from those data should be trusted.

Katz (2015) calls for qualitative sociologists to create a “triangular relationship [among] subject, reader and researcher” in which the reader can access a rich variety of data about the subject (cf. Denzin, 2009 [1970]). By providing the reader with a thick file of information to rummage through rather than just a refined and abstract argument (Becker, 2007), Katz argues, the sociologist holds their arguments accountable to disconfirmation:

By empowering the reader to detect contradictions, the web-like structure of data implications in an ethnographic monograph completes the triangular interaction. The researcher has effectively given up a monopoly on access to the subjects, at least for purposes of testing the author’s explanation. (Katz, 2015: 131)

When it comes to readers using authors’ own reports to hold the latter accountable, Katz is only correct to a limited extent: if the data fail to consistently match the theory, the reader can challenge the sociologist’s story. However, if the data do consistently match the theory, it is impossible for the reader to determine whether the sociologist is right or just a careful writer. The reader, after all, does not actually have independent access to the subject: every contact with the latter is mediated through the researcher’s writing; the researcher does not create an actual triangular relationship, just the appearance of one. What the reader is “testing,” then, is not contradiction between the theory and the actual events that the researcher observed, but rhetorical efficacy: the researcher’s skill in spinning a web that includes just enough and just the right data to create a reality effect that convinces the reader that the latter is directly accessing the subject and that the researcher’s arguments hold up.Footnote 6

For a truly triangular relationship, the reader needs to be able to directly and independently access research subjects without the author as a buffer between them. Journalists’ straightforward solution to this problem is to identify their sources whenever possible. In the age of social media and internet search engines, such transparency is a more effective reliability-enhancer than ever. Not only professional peers but also even lay readers can nowadays look up and communicate with the people whom journalists describe. A growing number of ethnographers are likewise calling for their colleagues to name names to allow such fact-checking in the interest of reliability and accountability (Jerolmack & Murphy, 2017; Reyes, 2018: 210–212).

Another, albeit weaker, form of transparency is to name places with relative exactitude. Ethnographers used to set their stories in “Eastern City” (Whyte 1993 [1943]; Anderson 1990) but now are more likely to name their research sites as Boston or Philadelphia, perhaps even as specific neighborhoods. Places are as much the subjects of social research as are humans, and creating a triangular relationship among author, reader, and place can to a limited extent empower the reader to interrogate whether an ethnographer has misrepresented or left out significant realities (Duneier, 2006; Reyes, 2018: 206–209).

In my own research writing, I could name places but not people. The whole point of creating composite characters was to protect my research subjects’ anonymity.

The second option available to qualitative researchers looking to produce reliable findings is to focus on the internal accountability of author to data. Internal accountability does not protect the reader against intentional malfeasance by the author, but it is a way for us authors to keep our own analytical laziness in check.

In “How Not to Lie With Ethnography,” Duneier (2011) suggests that once they have constructed an argument, ethnographers should conduct a thought experiment. They should put themselves on “ethnographic trial” and seek out an “inconvenient sample” by asking themselves,

Are there people or perspectives or observations outside the sample whose existence is likely to have implications for the argument I am making? Are there people or perspectives or phenomena within the sample that, when brought before the jury, would feel they were caricatures in the service of the ethnographer’s theory or line of argument? (Duneier 2011: 8)

The purpose of the ethnographic trial (which is equally application to other forms of qualitative research) is to encourage constant vigilance as to the selection of data and discourage authors from taking the shortcut of fitting data to explanation instead of the other way around. If the author is “committed to form a perfect relationship between data and explanation,” as Katz (2015 [1983]: 113) recommends, that means that evidence that contradicts or is left unexplained by their current explanation should—to the greatest extent possible—be scrutinized in order to improve that explanation.

When it comes to sociological writing that uses composite characters, the reliability problem expands. When data collected on multiple participants get pooled together to form a single composite character, it is even easier to cobble together a just-so narrative that conveniently conforms to the claims that the researcher would like to make. Writing in a conventional style of one-to-one participant-to-character correspondence at least constrains the author to a degree by insisting upon participant-character unity.

The conventional sociology writer cannot say Character Ayşe experienced X and then did Y, where X is a convenient explanation for Y, unless X then Y really occurred in a single participant’s life. X is not necessarily the cause of Y, but we know that an X-then-Y sequence of events is at least objectively possible, because it actually happened.

A composite character is created precisely by saying that Character Ayşe experienced X and then did Y, when really one person experienced X and a different person did Y. That combination of X and Y into the narrative of a single character may misleadingly create the appearance of causality. How do we know those two people are similar enough to be treated as interchangeable for the purpose of that narrative sequence?

It does not help that previous authors of composite narratives have often been vague about how they have gone about selecting and arranging data into stories and explanations. They have been far more articulate in explaining why they chose to fictionalize than the nuts and bolts of how they did so.

Sennett and Cobb explain the method used in their 1973 book The Hidden Injuries of Class as follows:

We have taken certain liberties beyond those necessary to protect anonymity: in various instances we have condensed remarks people made; when statements two people made on an issue were very similar, we have portrayed them as coming from one person. In a few instances we have put words in people’s mouths, words they were struggling for, we felt, but couldn’t find. Twice we have combined elements from several life histories into one. We hope that the people interviewed will forgive us for pushing the presentation of their lives so close to the boundaries of fiction. It is for clarity and art that we have done so, though we hope it guards their privacy all the more. (Sennett and Cobb 1972: 42–43)

Although the book is an otherwise realist sociological endeavor to describe working-class culture and social aspirations in America, Sennett and Cobb ( 1972: 43) argue that their work ought to be held to different standards of truthfulness than conventional sociology: “This raises a question for you, the reader: how believable is what you read? Art creates a different truth from the recitation and interpretation of facts.” Ending the discussion as they do with the appeal that their claims be assessed as would Art’s, we know nothing of the process by which they created composites. They do not explain their criteria for deciding which elements of which life histories to combine or for judging what words to put into which people’s mouths.

Other authors have adopted similarly opaque defenses of the reliability of their composite stories but donned the positivist armor of the Expert instead of the subjectivist cloak of the Artist. The researcher, these authors say, can responsibly shoulder the burden of complex data analysis and avoid misrepresentation because of their deep understanding of the subject matter earned from the ritual of long-term fieldwork (a classic line of argumentation whose application has not been limited to composite narratives). Take Willis’s (2019) explanation of how she combined 14 politician interviewees into four composite characters. Whereas she begins her methodological discussion by stating that she clustered interviewees into the composite pairings that “best convey[ed] the range of positions and view that the data reveal,” (Willis, 2019: 475) —a nod to a systematic logic of selection to “sample for range” (further discussed below) —Willis ultimately justifies her choices with a claim to social scientific expertise:

The researcher needs a level of understanding and familiarity with the context of the study, in order to judge what makes a meaningful composite. In my case, I had worked with politicians for many years, and understood [their] world… this meant that I had a depth of understanding which enables me to write up data in this way. (Willis 2019: 478)

One rare exception to the trend of mystification over explanation among authors of composite narratives is Goldman in her 1999 book Passionate Journeys, on women who joined a spiritual community in Oregon called Rajneeshpuram in the 1980s. Goldman is explicit about her methods for clustering 46 study participants into three composite characters based on age. At the end of the book she includes tables showing the number of participants comprising each composite and some demographic information about those composites (Goldman, 1999: 272–274). In a methodological piece reflecting on the book, she explains that she decided inductively on the demographic categories that would form the basis for her to cluster participants into composites:

I did not group the women into composites by some predetermined criteria, such as their ethnicity, age, social class, religion of origin, former occupation, or role at Rajneeshpuram. Instead, I read transcripts and listened to tapes in order to discover how they should be combined. My decision to organize the composites in terms of women’s ages reflected the strongest thread linking particular sets of women… I was surprised at how much women in each of the three groups forming the composites sounded like one another, using the same words and articulating similar concerns, even compared to women in the other two composites. (Goldman 2002: 155)

Goldman then presents a theoretical defense about American women’s worlds in general changing significantly in the decades between which her different composites came of age, differences significant to the reasons they chose to go to Rajneeshpuram and the way that they interpreted their experiences there, the central issue in which Goldman (1999: 2–3) was interested. This method of clustering participants makes sense insofar as Goldman first looked at similarities in what participants said and did as a basis for combination into the same character.

However, the advantage of clustering participants on the basis of a single variable (age cohort) that serves as a proxy measure for discovered similarity–as opposed to actually clustering them on the basis of that similarity–is unclear. What became of individuals who did not conform to their cohort? Were their nonconformities ignored, or was data from their lives included in composite narratives despite their nonconformity? Goldman (2002: 160) seems to have adopted the former method, as she states that she attributed major events to a composite’s life based on their occurrence in the lives of a majority of the participants clustered into that composite. In effect, this limits the range of data she selects and makes it impossible for her to address variation and inconvenient data within age cohorts to better “form a perfect relationship between data and explanation.” (Katz, 2015 [1983]: 113)

The challenge, then, is to design a method for creating composite narratives that pushes the author to notice and explain variation and inconvenient data, that is grounded in a systematic logic of data selection rather than an authoritative claim to artistry or expertise. I will now explain how I tried to keep my fiction accountable to the real world in the sense of allowing empirical data to shape and refute my emergent conjectures.

Step One: Compression

There are two steps to creating composite characters: (1) deciding how to compress or cluster individual research participants into characters, and (2) deciding which data to attribute to those characters. At Step One, my aim was to compress multiple participants into composite characters in a way that enriched rather than impoverished analysis.

The risk is that clustering participants based on unexamined assumptions about who seems similar to whom can effectively strip down the richness of data by squeezing participants into received stereotypes or categorizing them by demographic criteria that are not necessarily causally significant to the questions under study. The best way to cluster participants into characters may vary from project to project; the important thing is that they are compressed in a way that brings together individuals who are similar in a way that is relevant to the study’s guiding questions. Using an abductive strategy for clustering can help avoid tautologically producing composites who fit into categories suggested by initial research questions. Such composite characters would be inherently geared toward creating the just-so story that the researcher expected to find when entering the field (Tavory and Timmermans 2014).

Drawing on insights from Charles Ragin’s (2008) Qualitative Comparative Analysis,Footnote 7 Vladimir Propp’s (1970) methodology for typologizing folktale sequences,Footnote 8 and Andrew Abbott’s (1995) sequence analysis,Footnote 9 I developed a method for compressing cases (real people who participated in my study) into composite character narratives based on the degree to which the life sequences of those participants resemble one another.

I created what I call a sequence table, which consists of a series of entries arranged chronologically that summarize research participants’ careers in the news media and the backgrounds that brought them into those careers. The entries in the sequence table are drawn directly from the coding I apply to my data based on their hypothesized causal relevance to the variable outcomes that I am interested in: differences in the ways fixers enter into the field of journalism, do their jobs, shape the news, and use fixing to accomplish other goals.

An invented sequence table is below, which is shorter and more readable than the actual sequence tables I created for Fixing Stories:

Table 1 Participant Sequence (Open Coding)

Depending on how abstract the codes inputted are, it may be that initially no participant sequence much resembles another, as in Table1 above. The codes then need to be refined through a process of abstracting specifics into generalities to identify similarity. There must be some basic forms of similarity, unless I have made a mistake in bringing together these cases into a single study. The sequence table is not only a tool for determining similarity, but also a visual aide for transitioning from “open” coding that describes data to “selective” coding that sorts data into categories of core causal relevance (Strauss & Corbin, 1998); codes and table mutually refine each other.

Eventually, I have abstracted codes and left out the entries deemed–through logical inference (Small, 2009) —to be of lesser importance, to the point where I am left with several stripped-down basic sequences that multiple participants more or less fit into. In some cases, I add variables as I notice their importance across multiple participants’ careers. I put the table on “ethnographic trial” by thinking through what variables might be added to differentiate participants who appear similar on the table but whom I know to be very different in real life. For instance, I add “socialized with foreigners in Istanbul” as social ties to foreigners figured into both Participant A and B’s initial recruitment as fixers, in contrast to other participants.

It is around these sequences that I cluster participants into composite characters and build character narratives. In some cases, a young participant’s sequence looked much like the beginning of an older participant’s sequence, and so I made the inference that the younger participant was on a similar trajectory to their elder and clustered them together even though the shorter sequence would need numerous additions at the end to match the longer one. Note that participants do not always fit squarely into characters–Participant C, for instance, has much but not all in common with Character Can (e.g. Diyarbakir is in Eastern Turkey) —and so I end up with participants who bridge multiple characters.

Table 2 Character Sequence (Selective Coding)

The number of characters I created represents a balance between imperatives to show both patterns and variation. Were I to treat everyone who participated in my study as unique–and indeed, no two life stories are ever quite the same–then it would be impossible to generalize about patterns that were common to multiple people. Were I, by contrast, to lump everyone into a single archetypal character, The Fixer, in order to discover one overarching pattern, then I would leave none of the person-to-person variation that is essential to refining arguments and explaining causes and effects through comparison.

This table is meant as an aide to comparative analysis, not a substitute for it. When attributing data from Participant C to Character Can, I must acknowledge the incomplete match between the two. Being raised in the Kurdish-majority city of Diyarbakir, for instance, might be causally connected toward differences in Participant’s C’s real career that are not captured by the simplified table, for instance not just whether he works as a fixer but how he does the job. Where those differences are significant enough and the variation of Participant C from other participants composing Can noteworthy enough, I can create a minor side character with Participant C’s relevant background (e.g. raised in Diyarbakir) who does something important that I found Participant C did, with only minimal risk and potential harm of unmasking given the limited information about that side character that will appear in the final ethnographic text.

The process of comparing rows challenged me to come up with novel theories to which my initial assumptions blinded me. For example, I initially divided my research subjects into the categories local fixer and foreign reporter, assuming the two labels corresponded to separate roles requiring different skills and so resulting from different trajectories. But when I plotted everyone onto the same huge sequence table, I found that the sequence table failed to neatly divide along these lines. Turks, Kurds, Syrians, Afghans, Europeans, and Americans did not conveniently sort themselves into national clusters or into “foreign reporter” and “local fixer” clusters. Some participants went from trajectories similar to others labeled “fixer” to trajectories similar to others labeled “reporter”; others vice versa. The sequence table forced me to think beyond the divisions received from my participants’ and my own initial categorization of the field of international journalism. The SBC method instead encouraged me to think about the fuzziness of those categories and about how and why individuals can move between being labeled as one or the other.

Good sociologists consider all of these issues of categorization, temporality, and relative importance of variables anyway. This coding and sequencing method is simply a thinking aide, a more formal way of organizing interrogation of the data and decisions about what constitutes relevant similarity and difference.

Stripping down lived experiences into such table entries inevitably means that certain entries look the same on paper but have very different meaning in actual participants’ lives. The shorthand of the sequence table fails to capture much of the richness of the ethnographic data collected. A central difference between quantitative methodologies–which convert participants into data series and then analyze only those data series–and this compositing method is the following: the sequence table is used only to systematically think through degrees of similarity and difference among participants, and not as a way of thinning down the data I ultimately include in character narratives. When it comes to writing narratives, I remix my data with the sequence table as a guide but include the same richness of description that I would in conventional ethnographic writing (Wertz et al., 2011; Willis, 2019).

Step Two: Remixing

Now that we have our cast of characters, how can data collected from interviews and observation of numerous individuals be mixed and matched into a fictional narrative in a way that is not arbitrary and that forces the author to question a priori assumptions rather than giving them the freedom to craft a tale around those assumptions?

I will explain my approach in comparison with Goldman’s (2002) aforementioned Passionate Journeys. Goldman (2002: 160) writes in defense of her composite character method as a legitimate form of sociological writing, “No major event such as marriage, divorce, or childbirth was recorded unless it was shared by the majority (emphasis added) of [participants] in a group.” The problem with Goldman’s approach is that it follows–without actually having a statistically significant or random sample–a logic of statistical fiction, selecting the characteristics and experiences most common to the group without attention to whether the selected characteristics and experiences coincide in particular members of the group or are logically compatible with one another. Goldman, at least according to her methodological appendix, ignores the logics of sequence and causality.Footnote 10

To illustrate why this is a problem, imagine that of the seven participants whom I cluster into the composite character Ayşe, six had two Turkish parents and one had an American mother. Four went to Turkish-language schools in Istanbul and three spent their childhoods in the US. Four of the participants (including the three who grew up in the US along with the one whose mother is American) spoke English with native fluency. The “statistical”Footnote 11 or majority-rules approach would here dictate that I include the theoretically-relevant events that are roughly similar across the greatest number of participants composing Ayşe and include those in the narrative, because they are most representative of what her component participants actually did. Thus, I would write that Ayşe has two Turkish parents (86% of participants), attended Turkish-language schools in Istanbul (57%), and speaks English with a native-sounding American accent (57%). This would be a problem for my narrative because logically, someone who grew up entirely in Turkey without a foreign parent or immersive language education would have difficulty attaining such bilingual fluency.

Instead of adopting such a “statistical” approach to data selection, I based judgment as to a datum’s inclusion not on merely of whether similar data exist in the lives of other participants who compose the same composite, but instead on causal inference (also referred to as logical inference in Mitchell 1983: 199–200) about how antecedent data limit the possibilities for subsequent data to be selected.Footnote 12 Narrative–rather than statistical–reasoning is a better option for a study that, after all, has not followed standards for statistical analysis.

Stone (1979: 3) defines narrative as “the organization of material in a chronologically sequential order and the focusing of the content into a single coherent story, albeit with sub-plots.“Footnote 13 A narrative explains what it includes later in a sequence by what happens earlier in that sequence; it justifies the inclusion of what happens early by its relevance to what happens later. Events cannot occur later in the sequence that events earlier in the sequence prevent or for which they do not create conditions of possibility. The narrative is inherently embedded with a causal logic (Griffin, 1993: 1098–1099).

Narrative sociology thus must deal with temporality and sequence. Events in a character’s life at Time 1 may be relevant to her actions at Time 2 (i.e. constrain possibilities at Time 2). Any social scientist using narrative must be careful in selecting from raw material for inclusion in a narrative and explicitly interrogate themselves about what is causally relevant (Griffin, 1993). For authors of a composite narrative, this problem expands beyond selecting from data of what actually did occur in a sequence of events in the real world to the more creative exercise of creating sequences that never really occurred.

The SBC solution is to consider the data I am combining from multiple sources not just as a collection of individual objects to be mixed and matched within the constraints of my table-assisted determination of similarity among participants, but as potential building blocks for a sequence that must follow a causal logic of objective possibility. The claim is that even if the sequences I create are unreal–they did not actually occur in the lives of real individuals–they could have occurred.

The concept of objective possibility is central to Weber’s (1949) argument for the utility of counterfactual reasoning in building causal arguments. Any argument of causal relevance includes, implicitly or explicitly, a counterfactual fiction that a significant event did not occur as in reality it did and, as a result, that things turned out differently. Creating a counterfactual–but objectively possible–fiction provides a way to isolate potential individual causes from the greater and more complex reality to understand their significance or lack thereof (Ringer, 2010: 81–84).

Causal inference constrained my selection of data to include in narratives. This causal inference was based on counterfactual comparison across cases about how antecedent events and characteristics limit the possibilities for subsequent events. For example, if Ayşe had not lived and studied abroad or had an anglophone parent, it is nigh on impossible that she would have achieved the same fluency speaking and joking around in English that set her foreign reporter clients at ease. I can make the inference that Participants A and B’s time abroad causally influenced their language abilities by comparing them with participants who did not spend time abroad. If time abroad is thus causally relevant, it is worthy of inclusion in the ethnographic text. I also determine that an observation I made of a fixer-reporter relationship breaking down because of language problems ought not be attributed to Ayşe, because I can infer that a fixer who had spent years living and studying abroad would not have had such linguistic difficulties.

The interrogation of the causal logic of character narratives is also a means of negotiating the aforementioned incomplete fit between participants and characters. When combining data from Participant C with that of other participants composing Can, I must ask myself, “If Participant C had grown up in western Turkey like Character Can, rather than in Diyarbakir, could he still have done X?” I compare Participant C side-by-side with other participants composing Can and determine whether they have done things similar to X. If not, then I may be on my way to building a theory about the significance of hailing from a minority province for fixing careers.

This facet of SBCs will not come as groundbreaking to anyone who has written a novel or short story or play. Of course the author should question whether their story’s characters make sense, and of course those characters make sense when their thoughts and behaviors in a scene are explained by the combination of the situation and the precedent developments that led to the scene.Footnote 14 But in a field like sociology, in which statistical reasoning is too often inappropriately applied to make a superficial claim of scientific reliability, it is worth taking the time to think through the merits of alternative non-arbitrary ways to select and arrange data.

Reactivity: The Ethnographer as Study Participant

How has the intervention of the researcher affected

the very behaviors the researcher seeks to study?

Positivist studies based on lab experiments, standardized surveys, or structured interviews tend to view reactivity as a problem. If research subjects’ behaviors and responses are affected by the presence of researchers–or worse yet by variations in researchers’ own behavior or appearance–then data become distorted and unreliable.

In contrast to such methods that seek to minimize or standardize the effects of the researcher’s own presence and actions, and in keeping with the postpositivist tradition, Katz (2015 [1983]: 138) argues that ethnography can generate “valuable substantive data out of the responses of members to the researcher’s methods.” According to this thinking, reactions to a researcher should be viewed not as noise interfering with the signal of people’s “natural” behaviors in the absence of that researcher, but as useful data in their own right. Katz (2015 [1983]: 138–139) notes,

In relations with researchers, members will take what is to them significant action by identifying researchers as significant others. To consider participant observers significant, a member must cast them into identities rich with indigenous meaning. In field studies of communities or organizations, researchers may be grilled as informants, sworn in as confidants, and debated as representatives of the views of various groups and leaders. In these relations, members reveal their concerns, not about the world of social science research as understood by the researcher, but about everyday aspects of their own social lives. If by their presence analytic field researchers change the scenes in which they participate, the data they take out are still about those communities and organizations… Rich data are available in members’ efforts to place a field researcher in role and at a distance useful for native purposes.

Qualitative sociologists should thus pay attention to the ways that participants make sense of and behave toward them in order to better understand those participants’ social worlds. In the case of my research, participants deployed the same cultural toolkit to interact with me that they had developed as professional intermediaries between local realities and international interests. The fixers I approached and asked to interview, shadow, or hire assessed me much the same way they assessed new clients, feeling out what I knew and what I was after, whether I could be trusted to not embarrass them or damage their relationships with people to whom they introduced me.

Data from researcher-participant interaction can also help ethnographers to adopt critical reflexivity and consider how their own social positionality may be affecting their interpretations. Just as study participants make sense of the researcher according the categories and boundaries indigenous to their social world, the researcher makes sense of participants according in part to the social position they adopt or are thrust into in that world.

To sort out how researchers affect and make sense of participants and vice versa requires a “sociology of social research” (Katz, 2015 [1983]: 138), an auto-ethnography of sorts that analyzes how the researcher fits into the world they study (Kotišová, 2019: 297–298). Many qualitative sociologists include reflections on this matter in ad hoc fashion, but the SBC method provides a systematic way of considering reactivity. A researcher need simply plot themselves on the sequence table alongside study participants.

This approach may not work for outsider researchers with little in common with their informants, but provides a useful way for “native” ethnographers or social scientists with cognatic backgrounds to those they study to measure their similarity to and difference from various participants and so map their own position in the world they study. In my case, as a sociologist, I was studying “sideways”: analyzing a cognate discipline of knowledge production with some semblance to my own (Hannerz, 2012: 3). My own bicultural background, educational and work experiences, and language skills were not so different from some of my study participants’. It made particular sense to chart myself on the sequence table and incorporate my own experiences into the composite narratives I wrote because my fieldwork was heavy on the participation side of participant-observation. I myself worked as both a reporter–hiring fixers to assist me–and as a fixer assisting reporters. Had I not ended my fieldwork, I would have continued along a career trajectory in the news media more similar to some study participants’ than to others’.

Consciously comparing and contrasting myself with participants and plotting the trajectory that brought me into and through the world of news production in Turkey enriched my analysis of data drawn from both participant-observations and interviews. The ways that client reporters treated me as a rookie fixer or that interviewee fixers presented themselves to me as a foreigner of a familiar type could be fruitfully compared with study participants’ experiences and self-presentation strategies toward other interlocutors. Where study participants’ reactions to me and mine to them fit into patterns I found among those participants clustered alongside me on the sequence table and logically fit, I incorporated them into SBC narratives alongside my other data.

Considering myself not apart from but rather as a participant in the world I was studying also helped me to critically examine my own subjective interpretations, assumptions, and loyalties. Was I biased in favor of some participants, thinking them more credible informants and better journalists, because they were more similar to me? Did I find it easier to meet and establish rapport with certain kinds of fixers and reporters for the same reason, leaving gaps in my knowledge of the field of journalism? The impeachment of myself as a partial witness became an element of the cross-examination in the “ethnographic trial” I held.

Folding myself into the composite characters I created had two further benefits. First, it made anonymization more robust. A number of media workers in Turkey knew who I was collaborating with–in some cases because I established working relationships through their referrals–as did sources we interviewed for news stories. Scrambling myself along with other participants prevents such insiders from deductively attributing other actions and opinions to my collaborators whom they recognize in the text. Second, as discussed further in the replicability section below, creating a composite character first-person narrator and periodically reminding readers of this fictionalization was a way of further self-consciously undermining my own ethnographic authority.

Representativeness: Testing Both Dispositional and Situational Hypotheses

Do the data represent more than just the particular cases and individuals

under study such that general conclusions can be drawn?

In quantitative research, the question is, did you collect a representative sample of the population of concern that would allow us to extrapolate your conclusions to that entire population? When it comes to qualitative research, the matter of representativeness shifts instead to the generalizability of claims to have identified causal linkages among elements under analysis (Mitchell, 1983: 199–200). Claims are generalizable in that they offer explanations of generic social processes like identity management, perspective acquisition, or intercultural brokering that find instantiation in other cases and contexts (Prus, 1987; Puddephatt and McLuhan 2019). Those claims might scale well to represent the average or predominant experience of a population, or at least provide tentative hypotheses to test against larger data sets and macro-level phenomena (Zerubavel, 2021: 35–36), but extrapolation to entire populations is not what qualitative research is designed to accomplish.

Representativeness, in my case, would not mean putting forward claims that most fixers do XYZ that hold true for all fixers everywhere, but identifying chains of cause and effect that we would expect to find in analogous cases as well. For example, I argue that if a fixer has a long-term working relationship with a reporter client, the latter will become more inclined to accept their suggestions for changes in stories’ editorial focus, both because the two have built trust through successive interactions and because the fixer has learned over time to frame their suggestions in ways that conform to that reporter’s formulae for news stories. This is a claim that can be tentatively extrapolated to all fixers and other kinds of brokers contributing to cultural production. But it is a different kind of general claim than, for example, extrapolating from the finding that “80% of [450] fixers [surveyed in a Global Reporting Centre study] report questioning or challenging the editorial focus of a client’s story,” to infer that approximately 80% of all fixers around the world do the same (Klein & Plaut, 2017).

The claims of qualitative social science to representativeness are always tentative because, in the pragmatist tradition, any new case should serve as potentially disconfirming data that might require an additional caveat added to the claim’s causal reasoning and generic applicability. For instance, I might discover a case in which a reporter has worked with the same fixer for a long time but never listens to their advice and realize that there is an additional significant variable I missed because of lack of variation in the cases I had previously studied. Katz describes this iterative process of revision of tentative theories–also known as abduction (Tavory and Timmermans 2014) —as fundamental to the progressive generalization of theory:

When encountering a ‘negative case’ —evidence contradicting the current explanation–the researcher must transform it into a confirming case by revising the definition of either the explaining or the explained phenomenon…The more differences discovered within the data, the greater the number of possible negative cases, and thus the more broadly valid the resulting theory… [T]he analytic method… actually promotes the discovery of internal variety and thus its logic for establishing external validity. (Katz 2015 [1983]: 133–136)

The inclusion of variety, then, is important for the development of theory that becomes more and more representative with every added causal caveat.

As described above, instead of adopting a “statistical” approach to data selection, I base judgment on a datum’s inclusion on causal inference about how antecedent data limit the possibilities for subsequent data to be selected. This method allows me to determine what data I should not include, but may leave me with multiple data (that conflict with one another but not with what happens at other points in the narrative) that I can include within bounds of objective possibility, analytic license, and causal reasoning.

Since, in order to build the most generally representative theories possible, I am interested in showing a variety of behaviors rather than in proving what sequence of behavior is most likely, I select the data that highlight differences among composite characters and so illuminate different causal mechanisms at work in the careers of different fixers in the sharpest relief (Katz, 2001: 467; Willis 2018: 475). “Sampling for range” —including and explaining the greatest possible variety of data–forces the author to refine arguments toward the ideal of a “perfect relationship between data and explanation.” (Katz, 2015 [1983]: 113; Weiss 1994: 22–24; Small 2009: 13; Zerubavel 2021: 24–29) In the example above, I might make Ayşe speak English fluently because no participants who compose other characters could do so. This would allow me to illustrate and explain variation in fixing practices caused by differences in linguistic abilities.

If I decided that it was essential to feature a character who speaks English fluently, but that behavior failed to fit logically into Ayşe’s character, then I could create a side character of limited importance and description who–like utility characters in novels who are introduced to perform a necessary action that a main character logically would not perform–would speak English fluently and then do the fixer behavior(s) that I have reason to believe only a fluent speaker could do and that merits mention in my account.

Two common ways that sociologists question one another’s claims to representativeness pit dispositional arguments against situational ones. Dispositionists like Bourdieu (1977) look for explanations of what most people do most of the time with reference to precedent life histories that endowed them with particular abilities, limitations, and ways of seeing the world, i.e. dispositions. Dispositionists tend to depict people as consistently behaving in ways determined by their positions in macro-level social structures like class or racial hierarchies. By contrast, interactionists like Goffman (1959, 1969) or Katz search for explanations within micro-level situations, examining how various moments of social interaction present problems and opportunities to social actors. These situationists tend to depict people as inconsistent and flexibly engaging in strategic action, albeit constrained by the cultural “tools” available to them (Swidler, 1986).

Dispositionists charge that situationists’ claims are not representative because, in limiting their scope to the analysis of situations, they ignore the unequal distribution of abilities to implement interactional strategies and so miss the forest for the trees. Situationists charge that dispositionists’ claims are not representative because they fall short of explaining exceptions to the overall trends or individuals’ inconsistencies over time and across different situations, and so miss the trees for the forest.

To best contend with the arguments of both schools, a study should both consider life histories and allow for inconsistencies at the situational level. In my study, this means that it can be illuminating to show seemingly different characters engaging in similar behaviors. For example, I might have Ayşe, seemingly despite her elite pedigree and cosmopolitanism, joining news sources in talking suspiciously about her journalist client, implying that all foreigners are spies. Such behavior, which let’s say I have evidence of only one of the participants composing Ayşe engaging in, might be more frequently noted among participants composing the different character Can. Nonetheless, I might include the behavior in the narratives of both Ayşe and Can as a way of disconfirming a possible hypothesis that social status or cosmopolitan upbringing has a strict determining power over fixers’ behaviors. SBC characters, then, are not meant to be mutually exclusive in their characteristics and actions or to be internally consistent, as long as the inconsistency is deemed objectively possible through logical inference and comparison.

The incorporation of inconsistency and consideration of objective possibility is what differentiates composite characters from ideal types. Like SBC composites, ideal types do not exist in the real world but are fictional thought experiments created by an author. An ideal type is an exaggeration of a tendency observed in the world but pushed to its most extreme conclusion. The function of an ideal type is to serve as pole of comparison against which the researcher measures real world phenomena. Max Weber, as quoted by Ringer (2010: 102), writes,

Whatever content the ideal type is given… its only value… for empirical investigations lies in its purpose: to ‘compare’ empirical reality with it, so as to ascertain… the distance or degree of approximation between [reality and the type], and thus to be able to describe and causally to explain [reality] in terms of clearly understandable concepts.

Ideal types do not represent the average or collective characteristics of an empirically observed group of humans or phenomena, and their creator makes no claim that ideal types could plausibly exist in the world. Indeed, part of the sociologist’s job is to explain why an ideal type, “deliberately constructed to project a hypothetical ‘progression’ of external behaviors that could be fully explained in terms of understandable motives and beliefs about means of action” (Ringer, 2010: 101–102), is actually never realized in the world.

SBCs and ideal types share a kinship: both are built from rough sensitizing concepts (Blumer, 1954: 7–8; Zerubavel 2021: 3–7) like social mobility or professionalization that help guide a researcher’s attention toward the social patterns that best explain differences and similarities in how people behave and what opportunities they are afforded. Ideal types then offer the bases for comparing multiple research participants’ experiences in relation to those concepts during the compressing/clustering stage of the SBC method.

SBC characters are thus derived using ideal types as analytical tools. However, the characters that ultimately appear in the research text are not simply ideal types dressed up in people clothes. Instead, they are used in combination with ideal types to build theory to satisfy both dispositionist and situationist criteria for representativeness. The ideal types that appear in my text are figures like the outsider and patterns of behavior like pro-reporter bias against which I compare my characters, asking for example who is closest to the outsider pole of the insider-outsider spectrum and under what circumstances.

Constructing and then comparing fictions of (1) objectively possible composite characters and (2) ideal types affords me the advantage of allowing the identification of different kinds of individuals who work as fixers (satisfying dispositionism) without ignoring the internal inconsistency that may be evident in a person’s behavior as mere noise interfering with the signal of the typical behavior of a given character (satisfying situationism). It may be significant if characters are inconsistent, pointing to complex strategic behavior and/or to the causal importance of situations in which multiple characters find themselves (as opposed to those characters’ life histories and embodied dispositions) in determining behaviors. I might write that when reporting on the Istanbul political elite, Ayşe is biased in favor of news sources, but in other cases, she exhibits pro-reporter bias. The distance between Ayşe’s behavior and an ideal-typical behavior is measured as it potentially varies over the course of her career and in different situations.

The SBC method allows me to evaluate the relative importance of situations and life histories. If composite characters and typical patterns of actions align, i.e. each character consistently follows one pattern of action different from the other characters, it points to the importance of research participants’ life histories and dispositions (either in determining their reactions to situations that other characters also experience, or in determining what kinds of situations they encounter in contrast to other characters). If ideal type and composite character do not correlate, a situational analysis is invited.

It might be that I find no correlation between composites and ideal types in relation to any dependent variable: in a similar situation, no character is more likely than any other to conform to a particular pattern of behavior. If this is the case, I can safely jettison the sequence table and composite freely or, for that matter, introduce characters who exist only in particular situations and do not need their life histories detailed. The important thing is that I reach the conclusion that it is appropriate to focus on situations rather than on careers through examination of the data and not as an a priori assumption that then prevents me from being surprised by variation correlated to life histories.

In most cases, the answer I have presented in Fixing Stories is that a combination of situational and dispositional factors shape behaviors. In illustrating how fixers manage conflict that arises between their clients and local sources, for example, I discuss how fixers sometimes act in ways grounded in their life histories and positions in social structure that do not represent simply a situation-based compromise between reporter and source expectations. Yet life histories alone cannot explain all fixer behaviors. Thus to enhance the representativeness of my findings, the fit between data and theory, I introduce, for example, the concept of rhythmic ambivalence between professional and local loyalties that arises on a moment-by-moment basis in the course of particular interactions. When thrust into a situation of rhythmic ambivalence, I find, fixers with very different backgrounds and dispositions tend to react in similar ways.

Replicability: Fiction-Checking Through Analogy

Can findings be confirmed by others?

Replicability, like reliability, is a form of accountability, a constraint on a researcher’s freedom to depart from the real world beyond the bounds of analytic license. Replicating a study is a way to verify findings without privileged access to the people or private data that formed its basis. Statistical and experimental studies standardize their procedures not just to enhance internal reliability, but also to allow external audits in search of the same results.

Ethnographic research is different because data is situated in highly contingent real-world interactions and procedures cannot be standardized or duplicated. When participants are compressed and data remixed into composite narratives, the impossibility of redoing a study in the matter of a lab experiment becomes yet more obvious. I would argue that this is a good thing and encourages readers to look elsewhere for replicability.

By forthrightly labeling my work as empirical fiction and highlighting the creative hand of the author in the construction of narrative, I intentionally undermine the scientistic authority that might convince a reader that replication is unnecessary or that replicability can be taken for granted because of assurances of random sampling, statistical significance, experimental controls, and the like.Footnote 15 By creating a composite narrator and so depriving the reader even of the comfort of knowing which evidence is based on interview data and which on observations, what is hearsay and what I experienced myself, I intentionally undermine the naïve empiricist authority of the eyewitness or objective-by-default researcher (Rhodes and Brown 2005).

In so doing, I aim to encourage readers to evaluate my substantive arguments by retesting them in the same way they are developed: through the search for disconfirming data in analogous cases. We naturally find ourselves “replicating” studies in our minds whenever we think through whether the author’s arguments hold up in other cases we know (Katz, 2015 [1983]: 142–143).

This method of replication ought to be applied to qualitative research generally, but the weak claim to factuality of the fiction label helpfully guides users to comparatively replicate claims because it makes other forms of verification more transparently impossible, even absurd (Markham, 2012: 334–335). Nobody would try to re-run my study by somehow duplicating my fieldwork and seeing if they end up with the same fictional characters.

Theoretically oriented writings have long used another type of fiction–illustrative anecdotes and hypothetical thought experiments–to explain arguments and encourage replication through comparison. Erving Goffman, one of the most influential sociological theorists of the 20th century, cobbled together much of his “evidence” from a mixture of probably-invented anecdotes and snippets of novels, memoirs, and other studies. To take an example from Goffman’s Strategic Interaction:

Harry, the native spearsman [sic], having strayed from the territory populated by his tribesmen, comes into a small clearing to find that another spearsman from a hostile tribe is facing him from what would otherwise be the safe side. Since each finds himself backed by the other’s territory, retreat is cut off. Only by incapacitating the other can either safely cross the clearing and escape into his own part of the forest. (Goffman 1969: 93)

Goffman probably never observed any real-life standoff between tribal spearmen, but nonetheless confidently describes Harry’s subsequent actions and even thoughts. The persuasiveness of Harry’s story as evidence is not its factuality but that it prompts readers to consider the example with reference to analogous cases with which they are independently familiar. Harry’s specific fictional dilemma stands in for a generic form of social dilemma that readers can identify in its real-world instantiations (Zerubavel, 2021: 37–58). What strategic options were available to readers when they found themselves in an analogous standoff from which they could not retreat? How have countries conducted their diplomacy when facing rival countries whose interests conflicted with theirs?

As Katz (2015: 124) notes, the very shakiness of these vignettes and scenarios as evidence can be a strength insofar as it encourages readers to look beyond the text for verification of its claims and to not simply believe the author because of the facts the latter presents:

Erving Goffman’s oeuvre is extraordinarily weak methodologically in that he rarely identified where, of whom, and how he made his observations, but also extraordinarily strong in discovering forms of social life that readers had already come to know…Goffman enfranchised his readers to make up their minds about the empirical validity of his analyses in the private voting booths of their own intimate experiences…The very lack of detail in his texts about how he did his research is a methodological strength. Readers can ‘test’ his analysis on their own experiences….

Thus, Katz concludes (in a different article),

In a fundamental way, the allegation of fictive data is less meaningful when applied to qualitative field studies. In fact, many of the best interweave observational and interview data with excerpts from novels… But given the relation between author and subsequent researcher, this is a very constraining freedom. (Katz, 2015 [1983]: 143)

The reader is also a subsequent researcher, whether they conduct a formal comparative analysis to retest a study’s arguments or just check their validity as they read by assessing whether those arguments hold up to the comparative data of other cases the reader knows. The replicability of empirical fiction is thus inseparable from its transcontextual representativeness and analytic utility. Theories are tools for making sense of the world (Blumer, 1954). Sociological explanations are tools to analyze patterns of relations that transcend specific contexts and situations. A tool is “tested” through application to users’ problems, i.e. whether it offers a useful response (which is not to say a universal iron law) to more than just one specific context-bound question. A good tool should be applicable to a generic class of questions (Puddephatt and McLuhan 2019: 146–148; Zerubavel 2021). This “testing” is also in a sense the ultimate goal of public sociology: to encourage readers to analyze their own worlds in light of sociological explanations and see if things make more sense.

Any reader for whom my research is of any use will be able to find analogies in their own experience or in other stories against which to test my claims. (If they cannot, it means my research is irrelevant to their lives and it matters little whether my claims are true or not). Cases of the processes that Fixing Stories addresses–of mediation of social interaction, of indirect communication through a broker, of the reformulation of information as it passes between institutions, of moral ambivalence when engaging with interlocutors with divergent expectations–are commonplace. Cases for comparison need not be similar overall to that of fixers in Turkey, in terms of substantive historical context, individuals involved, or level of analysis (Zerubavel, 2021). Cases for replicability testing need only be comparable to my case on a particular axis of interest, much as prisons, boarding schools, and monasteries could all be fruitfully compared by Goffman to the mental hospital in which he conducted fieldwork because, despite vast differences among them, those institutions all socialized members in a similar way relevant to the theory that Goffman was developing (Vaughan, 2014: 62).

I went through this comparative process myself in the course of my research, drawing on analogous cases of information brokerage as far-flung as real estate agents (Bresbis, 2016), party organizers (Mears, 2015), auxiliary security forces (Cherney and Chui 2011), Black elites mediating between their communities and state bureaucracy (Pattillo, 2007), local representatives of a multinational aid organizations (Swidler and Watkins 2017), native clerks in 19th -century colonial India (Raman, 2012), and mixed-race elites in 16th -century Oaxaca (Yannakakis, 2008) in search of external data that could challenge and so help refine my thinking about fixers in Turkey. This final step of external comparison with other organizational forms and historical moments was essential for integrating my case study into a broader understanding of brokerage in knowledge production (Vaughan, 1992, 2014).

I provide readers with references to such analogous cases in the footnotes of Fixing Stories, and those sources provide one point in my research that does bear external audit. I can control a reader’s access to the data I collected and to my research participants, but I cannot control their ability to access those other studies or to test my arguments against other comparative cases they find on their own.

The Value of Non-Fiction and Tradeoffs of Compositing

If representation of society is always a creative process, why not simply make up evidence that best illustrates my argument–the most surefire way to cut out the causally irrelevant bits of reality? After all, doesn’t verification of a theory ultimately depend on the theory’s utility for understanding and reproduction across other cases rather than on its factual basis?

One answer, provided in preceding sections, is that one is unlikely to be surprised by data invented by oneself. A methodology that seeks out surprises by comparing and looking for non-obvious logical connections and incompatibilities among real data is more likely to transcend initial assumptions and biases in order to build novel, reliable, and representative theory.

Another answer is that people do not read research writings only for the theory. Thus far, the goal advocated for analytical sociology has been to build theories that explain more the case at hand. But from a postpositivist and pragmatist perspective, research is never truly complete. Every study also serves as a data source on which others build their own theories or as a rough draft of a theoretical argument to be refined by future social scientists (Murphy, Jerolmack, and Smith 2021).

Generations of debates over how to write ethnography and social science more generally have pitted chroniclers against theorists. Whereas the ideal chronicler records simply everything without prejudice to its importance–i.e. without reference to theory (Danto, 1962: 152–155) —the ideal theorist includes only what references to the world support their theoretical argument. From the theorist’s camp, Evans-Pritchard attacked his predecessor Malinowski’s “haphazard” documentation of a jumble of data, asserting that “facts can only be selected and arranged in light of theory.” (quoted in Clifford 1983: 126) From the chronicler’s camp, historians attacked Hegelian “philosophy of history” for creating just-so stories that coerce readers into their limited and ideologically driven interpretation. The writing of Hegel and his intellectual descendants, they charged,

…consists of nothing but plot; its story elements exist only as manifestations, epiphenomena, of the plot structure, in the service of which its discourse is disposed. Here reality wears a face of such regularity, order and coherence...presenting an aspect of such wholeness and completeness that it intimidates rather than invites to imaginative identification. (White 1980: 24)

I have thus far situated SBC methodology in the camp of the theorist, but it is worth considering the concerns of the chronicler and how a bit of the chronicle might be productively included even in a narrative of fictional characters. In counterpoint to Evans-Pritchard, Clifford (1983: 136) points out a possible value of a chronicle of reality:

[Malinowski] published much data that frankly he did not understand. The result was an open text subject to multiple reinterpretations…In the modern, authoritative monograph there are, in effect, no strong voices present except that of the writer. But, in Argonauts and Coral Gardens we read page after page of magical spells, none in any essential sense the ethnographer’s words.

Thus the very messiness of Malinowski’s writings makes them a resource for readers to create their own theories. Including rich data that extends beyond a theorist’s particular arguments can make it more useful to readers with different agendas. Ethnography as a data source can outlive the ethnographer’s own theories.

What is critical to readers’ abilities to read against the grain or orthogonally is the provision of excess, a term I borrow from cinema studies that means a sampling of reality that does nothing to provide evidence for the presented theory or narrative. Excess can be thought of as noise that disrupts the smooth signal of a theory.Footnote 16 Nichols (1991: 142) writes, with reference to documentary film,

Narrative is like a black hole, drawing everything that comes within its ambit inward, organizing everything from decor and clothing to dialogue and action to serve a story…Excess is that which escapes the grasp of narrative and exposition. It stands outside the web of significance spun to capture it.

Data are excessive from the perspective of the theorist: they do nothing to strengthen their argument. If the author does not think a subject’s gender is causally relevant, to mention it–perhaps to liven up a scene or flesh out a character–is to introduce excess. From the perspective of the reader, however, this excess can be at least as interesting as the author’s argument for a re-analysis that considers the case outside of the web of significance that the author has spun. The reader can use this offhanded mention to question whether the author has made a mistake by ignoring gender as a variable in their analysis, or can use it as evidence for an argument unrelated to the author’s (example drawn from Jerolmack & Murphy 2017:10–11). Jerolmack and Murphy note,

What [one] ethnographer considers ‘distractions’ [read: excess] might be considered by others to be part of the ongoing social scientific dialogue in which the reader can consider alternative explanations by independently exploring additional points of access into the setting or group– points that may not have even been considered by the ethnographer, given his or her particular theoretical interests or social position. (2017: 10)

In other words, one man’s noise is another man’s signal (Serres, 2007), and sociology is not a one-off exercise in theory building but a collective and ongoing effort in which scholars reexamine and reinterpret one another’s cases and claims (Tavory and Timmermans 2014: 111–120; Clifford 1983: 141; Vaughan 2014: 68–83). Research, according to this argument, should be disseminated in a way that includes excess in order to make it most valuable to the scholarly community. This may be particularly valuable in studies in which researchers manage to collect data on difficult-to-access subjects and situations, providing comparative data to others who cannot study the same people or places themselves.Footnote 17

Some forms of sociological writing are more prone to excess than others. Directly quoted speech is more likely to contain data that does not advance the particular argument being penned than are indirect paraphrasing or description (Clifford, 1983: 136–141). Direct speech more tightly bonds excess to evidence than does paraphrasing; prose description more so than classification in a chart; descriptions of real situations more so than fictional situations sifted through the filter of the author’s theorization and imagination.Footnote 18 Sennett and Cobb’s Hidden Injuries of Class, for instance, is not a very useful source of excess that can be turned into evidence for a different argument, because they put words in characters’ mouths and make it unclear where reality ends and fictionalization begins.

Compositing does not in itself prevent the inclusion of excess. Remixing real situations and quotations rather than fictionalizing them outright offers a compromise in which situational excess is allowed into writing, even as the compositing process by its nature excises biographical excess. Compositing makes it impossible to track data on individual research participants over time; this is precisely how the method prevents the deductive unmasking of those participants.

This reduction of potentially valuable biographical excess must be acknowledged as a trade-off to readers. As Nichols (1991: 144–145) notes,

The attempt to probe an individual both to understand that person and to use that person to reveal larger patterns or socially representative practices always produces excess [emphasis added]. Aspects of the person elude the frame within which he or she is placed. Dimensions of their behavior reveal a resistance to or subversion of patterns that could be seen as typical.

Removing excess limits a reader’s ability to push back against the authoritative claims of the researcher or gain an understanding of the people they studied beyond the frame of their theoretical arguments.

The downside of compositing is that biographical excess is not simply removed but replaced with biographical data from other sources, which can affect our interpretation of situational excess. A theorist of gender might, for instance, take my account in Fixing Stories of a male reporter character mocking a female fixer for her emotional connection to a source as evidence of misogyny and of the gendering of pro-source bias as perceived by client reporters. In reality, however, that account is based on a female reporter mocking a male fixer. That remixing of biographical information in my composite narrative could substantially change the way the reader interprets situational data and so could mislead them if they attempt to use my account as a primary source for their own argument.


Any means of representing society incorporates trade-offs between empirical exactitude and theoretical breadth. Journalism sacrifices grandly imaginative ideas for hard facts and literature does the opposite; quantitative methods of social science sacrifice data richness for statistical generalizability and qualitative methods do the opposite. The trade-offs of fictionalization must be acknowledged: creating a composite character narrative as outlined above sacrifices–as its central ethical justification–biographical excess for robust anonymization. The SBC method also loses fact-check-ability even as it encourages systematic attention to building theory and replication through analogical comparison.

The potential for misrepresentation corresponds to the extent that a sociologist is unconstrained in turning realities into narratives, data into theory. Realistically, qualitative sociology in general and composite character ethnography in particular face little external constraint when it comes to selection and arrangement of data. The constraints I have outlined to enhance reliability are tools for internal accountability. The purpose of such a rigorous methodology is not to salvage a claim to scientific authority for oneself as the authority of facticity crumbles, but to maximize the prospect of transcending one’s initial assumptions and coming up with useful and possibly new ways of understanding social interactions (that is, theory) grounded in participants’ lived experiences.

Those compression and remixing methods are for me to hold myself accountable to my own data, while the “empirical fiction” label is for readers, a warning that external audit cannot be limited to fact-checking. Framing research transparently as the author’s imaginative interpretation encourages readers to think about theory utility instead of fact-checking, and to verify theoretical claims through replication by analogical comparison rather than bowing to the power of the personal true story or of scientific authority. When research procedures are necessarily idiosyncratic and fact-checking is impossible, the best possible method is one that both encourages the writer to be rigorous and the reader to be skeptical.