The abductive logic of ethnographic research

Ethnography is a case study account of the cultural knowledge and behavior characteristic of a particular group or setting [1, 38, 39]. The “culture” piece relates to the fact that humans as a social species have the evolved ability to think and behave jointly or collectively [40–42]. That is, humans transmit information to each other, which becomes the basis of collective attention, intention, and action [43–45]. This is important because it implies that humans are incomplete without such socially learned information and behavior [10, 46]. Methodologically, this means that to fully explain human behavior in a specific setting researchers would need to understand local culture—that is, patterns of socially learned thought and practice [43]. And ethnographers developed techniques to do just that. Those methods included qualitative modes of inquiry such as participant-observation and interviews, which typically included open-ended phases of study to reveal among other things unexpected patterns that were not yet known to the researcher [47]. However, as an ethnographer’s knowledge of a particular place and culture grows, they might include more structured and even quantitative methods like field surveys to further clarify local thinking and behavior [47–51].

Ethnography, then, is more a particular mode of inquiry and logic than any specific method or technique, be it qualitative or quantitative, open-ended or structured [24]. Key to our understanding of ethnographic method is the (often unrecognized but ever present) role of logical abduction [23–25, 28]. In Peirce’s writing, abductive inference can be characterized as follows:

The surprising fact C is observed.

But if A were true, C would be a matter of course.

Hence, there is reason to suspect that A is true [23, 27].

Novelty, then, is critical to ethnographic inquiry [24]. An ethnographer is confronted with surprising observations in the field that cannot be easily explained according to current understandings of the world. These are surprising because they are based on socially learned structures of knowledge and practice that are not yet known to the ethnographer. Even ethnographers originating from a given culture group would still not understand all dimensions of this particular form of thinking and practice in this specific setting. The ethnographer subsequently draws on whatever resources available—their experience, current theory—to identify some A that would make C a matter of course. As Peirce noted, this form of logical inference—what he called informed guessing—relies on human creativity and invention [23]. Importantly, compared to both induction and deduction (discussed later), “A is neither assumed before the fact (as it is in deduction) nor observed (as it ideally is in induction). Rather, the proposition is guessed at, presumed after the fact to explain observations we cannot easily explain away” [27].

As explanations invented after the actual occurrence of an event, which was not actually witnessed, abductive inferences are neither rigorously true in a logical sense (as they are in deduction) nor necessarily statistically common (as they are in induction). Rather than representing a logical weakness per se, Peirce instead thought of abduction as the most creative of human logical processes [23]. Further, he thought, as do we, that abduced explanations can be further refined and complemented via inductive and deductive logic. This has been described as abduction’s iterative and recursive logic [24]. Ethnographers repeat (iterate) this form of logic until arriving at a satisfying conclusion. And prior explanations become embedded in earlier ones (recursion). As we will show, logical induction is important here: for an abduced explanation to be compelling, ethnographers should see repeated data points consistent with the proposed inference. Deduction also has its place: to gain greater confidence in an abduced explanation, ethnographers can test whether such explanations lead ethnographers to deduce subsequent propositions from the ethnographic findings that are logically consistent with each other.

To summarize, abduced explanations, as invented, can originate from various places. That could be the experience of the ethnographer, which, we would add, suggests that socially and culturally diverse ethnographic teams, such as those combining relative insiders and outsiders to a given setting, would have access to a greater reservoir of potential explanations. Or, alternately, those explanations could come from current theory, though necessarily adapted in some form to the new cultural circumstances. This has also been conceived of as a strength of abductive logic: in contrast to “inductive” approaches like grounded theory, abductive explanations can more effectively build upon and extend current theoretical understandings characteristic of a community of researchers [28, 52].

Assessing strength of evidence via the ladder of causality: Association, intervention, counterfactuals

Another piece of our argument, besides recognizing the relevance of logical abduction, is what Judea Pearl refers to as a three-level ladder of causality, with each level or rung of the ladder identifying a stronger causal process [29, 53]. The lowest level, association, is linked with seeing and observing regularities or patterns in the world. In statistics, one object is associated with another if observing the first object changes the probability of seeing the second one. Pearl notes how association alone is weakly indicative of causality in such instances, given a first object might be the cause of a second, the second might cause the first, or a third object or process (referred to as a confounder) might cause both of the two to occur. The middle level is intervention, which refers to how actions produce changes in the world. Experiments are the classic example of this level, and random assignment to the treatment or control group helps to ensure that the treatment, rather than some other non-random characteristic of the individuals in the two groups, produces the change. Ethnographers are more likely to observe naturally occurring events—which we refer to as interventions, though typically without conscious design—that change the value of a predictor variable in ways that can impact a study’s outcomes. The highest counterfactual level or rung of the ladder involves building theories of the world that explain why specific factors like a treatment have effects and what might happen in the absence of those factors [54, 55]. This can involve thinking through how different versions of past events (that were not observed) could have produced alternate versions of the present. In quantitative analysis, this means building models that allow a researcher to set key variables (such as a mediator) at different levels and observe how other parts of the model change, like how a main predictor is associated with an outcome.

In the current study, we show how ethnography can help reveal processes linked to each of Pearl’s causal levels. Ethnographers commonly identify associations between different factors at their field sites. Those linkages can help clarify causal processes. Ethnographers also observe naturally occurring events in the field that result in other things happening. Their careful observation of these events—naturally occurring experiments or interventions of a kind—can reveal how those events result in the changes. Here, it is critical to document, as we discuss next, how other factors at the field site might confound or mediate how the process being observed unfolds. Finally, ethnographers can work in a comparative counterfactual mode where they can observe and document how slightly different social and personal situations (established at some point in the past) have led to alternate present circumstances. This is challenging, and documenting the role played by confounders, mediators, and the like is again critical. Having respondents reflect on their experiences in the different situations and what might have produced them can also help ethnographers build more valid theories to explain the causal processes at play.

Ethnographically refining and testing causal understandings: The role of graphical causal models (DAGs)

Social science researchers have developed a language and approach for representing causal assumptions graphically, based in part on foundational work by Sewell Wright [56, 57]. One class of these graphical causal models are referred to as directed acyclic graphs (DAGs). In epidemiological and social science applications of such graphs, study variables are represented as nodes or vertices and their causal connections as directed arrows, edges, or paths that move in specific directions, as seen for example in Fig 1. The graphs are “acyclic” in the sense that following the directed edges will never result in a closed loop. Bidirectional causality can be represented in DAGs by creating predictor or outcome vertices at different timepoints. Fig 1 presents in abstract terms a hypothetical research project’s primary causal assumption that the study Predictor influences the Outcome. Further, the Predictor might also impact the Mediator, also shown in Fig 1, which in turn influences the Outcome. Such an intermediary variable might help researchers further explain the mechanism that connects the Predictor to the Outcome. Other causally prior factors that influence both the Predictor and the Outcome are referred to as Confounders. And causally downstream variables that can open illusory causal associations between the Predictor and the Outcome—because they occur after the causal process of concern—are referred to as Colliders. In general, such graphical models help researchers determine the identifiability of causal effects from observed data (i.e., whether the data includes all relevant factors). And they direct them on how to treat various classes of study factors. In reference to the latter, social science researchers would generally: examine how the Predictor-Outcome relationship changes under various Mediator values or conditions, to clarify that potential causal mechanism [29, 58]; account or “control” for Confounding variables (covariates) that might bias an estimated association between the Predictor and Outcome and thus create spurious causal relationships; and be wary of Colliders that if accounted for might deceptively magnify the purported causal association between the Predictor and Outcome [29, 30, 32].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. A graphical causal model expressed as a DAG.

https://doi.org/10.1371/journal.pone.0302857.g001

Of note, DAGs are qualitative models of how the world works: they only encode, for example, that there is a directed causal relationship between a predictor and an outcome; they do not say anything about the magnitude or functional form of that causal relationship. As qualitative models, DAGs are highly amenable to being refined through qualitative modes of ethnographic analysis of qualitative data (like using abductive or other forms of logic to code fieldnotes and interview transcripts). (Here, we follow others in using qualitative and quantitative to refer either to modes of analysis or to forms of data [59, 60].) But qualitative understandings of the world derived ethnographically and represented in DAGs can also be further refined and tested via quantitative analysis of quantitative data. For example, based on preliminary work, an ethnographer might develop and distribute a field survey to test (via regression analysis or some other strategy) causal processes encoded in an ethnographically derived DAG. This would form a multiphase research strategy consisting of exploratory and confirmatory stages of study. Nevertheless, to our mind, the project would remain thoroughly ethnographic if the derived understandings and concerns were all closely tied to culturally informed experiences, thinking, and behavior characteristic of the field site. All in all, graphical causal models provide ethnographers with a powerful tool to represent abstractly their evolving understandings (iteratively and recursively abduced) about what influences what at their specific study site. Representing their understandings in this way provides other researchers with a more transparent window into an ethnographic account, as a reflection of ethnographers’ expert knowledge and beliefs but also potential biases. And rendering causal assumptions more visible (to themselves and to other researchers) clarifies what is unique (or not) about a particular field site, which helps to advance theory. (For more detail on graphical modeling in relation to interdisciplinary work on causal inference, see S1 Appendix).

Ethnography’s abductive, iterative, and recursive logic in relation to gaming and player well-being

Early on in our ethnographic research, we learned that some players of the video game World of Warcraft (WoW) claimed to be addicted to that game. Such reports were surprising to members of our research team who had only recently begun playing that game, as they had not considered that a game could be experienced in this way. In part, reporting to be addicted to WoW revealed that a player found a game well-constructed, compelling, and fun. But it was more than that for certain players, who also found themselves playing compulsively in ways that eroded commitments to their families, partners and spouses, schooling, and jobs. Reporting WoW “addiction” was also a way of communicating the distress that this game created in their lives.

Our ethnographic observations and interviews showed us how important WoW play was to some players, which helped to explain its addictive qualities. As we advanced in our study of WoW and other games, we began to see how gaming avatars or characters could be akin to secondary or proxy selves for some players. That is, games were motivating in part because players experienced themselves as personally present in some powerful way in the games they were playing. A character’s challenges, victories, and defeats, then, were the players’ as well. And many players could not easily separate themselves from their characters. To put it more strongly, we learned how digital gaming avatars allowed players to enact and experience in virtual game-worlds their best and most ideal selves. WoW guild members, for example, developed in online characters more highly accomplished selves, which provided them with elevated social standing in this virtual world and guild setting. And those avatars, as idealized second or proxy selves, had the potential to help players compensate for their perceived life failings, thereby bolstering their psychosocial well-being.

The existing scholarship partially guided us throughout this newer phase of ethnographic study. In one well-known early formulation, Bessière and colleagues spoke of how gaming avatars or characters were akin to ideal elves (in the sense of a digital gaming pixie or elf) [63]. Other researchers investigating avatar processes in video games and other digital environments have used Edward Higgins’s self-discrepancy theory (SDT) framework to help explain well-being in these contexts [64]. First formulating his ideas in the late 1980s, Higgins, a health psychologist, identified three separate aspects of the self: what he called the actual self (the traits one actually possesses), the ideal self (the traits one would ideally like to possess), and the ought self (the traits someone believes one should or ought to possess) [65, 66]. Research in the SDT tradition shows that discrepancies between an individual’s perception of their actual and ideal selves are associated with dejection, sadness, disappointment, and depression, whereas discrepancies between perceived actual and ought selves are linked to agitation-related states such as anxiety, nervousness, and guilt. Adapted for virtual worlds research, the relevant self-dimensions are the actual self (defined in this context as the characteristics of the offline user), the ideal self (the traits one would optimally possess), and the avatar self (characteristics of the user’s online personae). A recent review of the literature suggests that the avatar self is typically created as “better than the user, resembling the ideal self more closely than the actual self” [64]. Even more relevant to what we were learning, a now substantial body of research suggests that gamers suffering from low self-esteem—predicted by SDT as resulting from felt tensions emerging from the perceived disjunction between the actual and ideal self [67]—tend to create avatars even more closely resembling their ideal selves, possibly in order to compensate for perceived shortcomings and thereby bolster psychosocial well-being. These adaptations of SDT, then, helped us to clarify what we were learning ethnographically about avatar therapeutic mechanisms: as approximations of individuals’ ideal selves, avatars provide opportunities for the enactment of one’s best self.

We might summarize these findings in the following proposition:

Proposition 1. Some gamers are highly motivated to play—even describing their gaming as “addictive”—because their character-avatars come to feel akin to self-objects or self-proxies, even ideal selves, which can help psychosocially compensate players for their felt personal deficiencies.

This idea was logically abduced, in Peirce’s terms [23], in the sense that what was once puzzling to us (WoW players’ deeply motivated and even compulsive and “addictive” game play) is clarified. The explanation is causal because a felt experience—my avatar-character is in some important sense also me—explains how gaming can come to play therapeutic (or in other situations distressing) roles in players’ lives. Proposition 1 was unexpected to us, but once we identified it, puzzles we encountered in the field (like addictive game play) became a matter of course. Of note, in an abductive approach, we allowed ourselves to make use of any ideas—be they from our own observations, interviews, and experiences or from the extant scholarly literature—that helped explain the puzzles we encountered in the field. This meant that we were able to engage current scholarship in ways not possible by approaches like grounded theory that aim to be more purely inductive. Our findings did echo and thus confirm in important ways the research of Bessière, Mancini, and others.

As we pushed further ahead in our inquiries, we learned that players in fact could relate differently to their avatar-characters. Some did relate to them as self-objects or proxies—like a me—as we described previously. But others treated them like distinctive other persons separate from themselves. And still others used them more impersonally like a tool or object. In relation to the latter, the lead author still remembers vividly an early interviewee correcting his assumption that this player’s relationship to his avatar was personal and deeply felt. Instead, the interviewee said that he used his gaming character more like he would a chess or backgammon piece—that is, as an object or tool to accomplish his in-game goals.

As in the previous example described in this section, there is literature on this topic, though we learned about them later. For example, research shows how video game players can relate to gaming avatars as object, me, symbiote, and other [68]. Each of these vary in terms of how players perceive their emotional investment in avatars, the autonomy of avatars from players themselves, the suspension of disbelief in relation to avatars, and their relative degree of control over avatars [69]. Experiencing an avatar as an object is associated with low emotional investment, low avatar autonomy (as players feel that digital avatars simply do players’ bidding), low suspension of disbelief (avatars are not imagined to be anything other than a digital tool), and a high degree of felt control. By contrast, feeling that the avatar is the self or me would be linked to high emotional investment, little avatar autonomy (the player and avatar are felt to be one), a high degree of suspension of disbelief (to imagine that the avatar is in fact identical to the self), and less of a sense of external control (because the player and avatar are a single entity). Further, players feeling they have a symbiotic relationship with their avatars—that is, merged to some extent as a single being and yet still distinctive, so that players feel that they “bleed” into their characters [70]—are likely to show somewhat less emotional investment and suspension of disbelief relative to players feeling that the avatar is me, and somewhat higher avatar autonomy and felt control compared to avatar is me kinds of players. Finally, perceiving the avatar to be a distinctive being or other is associated with high degree of emotional investment (because players establish human-like relationships with avatars), high avatar autonomy (the avatar is separate from the player), high suspension of disbelief (to imagine that the avatar is a “real” person), and less of a degree of felt control (the avatar does what it wants).

Further, we began to understand ethnographically that the kinds of relationships players established with their characters could impact the positive or negative quality of their overall gaming experiences. We thought this because those four kinds of player-avatar relationships bring players into either more or less intimate and meaningful contact with their avatars, thus potentially magnifying both gaming-related well-being benefits and distress. Thus, for example, greater emotional investment in avatars would mean that players’ achievement, social, and immersion experiences would be felt more keenly, thus potentially magnifying (for better or worse) the positive or negative consequences of play. In this sense, players relating to their avatars as me or as symbiotes (with high personal and emotional investment in those alter egos) would therefore more likely experience both greater gaming-related well-being gains and be at higher risk of problem play. Likewise, experiencing avatars as more autonomous would point to how players feel distinct from their avatars and thus more distanced or even protected in some sense from whatever achievement, social, or immersion experiences those gaming avatar-characters have. But following that logic, players feeling that their avatars are a me or a symbiote (with less autonomy) would again more likely experience the highs (and lows) of gaming-related pleasure and distress. Further, a high suspension of disbelief in relation to avatars would again suggest more intensely felt achievement, social, and immersion gaming experiences, in the sense that what-if gaming worlds are felt to be real in some important way. This would again mean that players relating to their avatars as a me or a symbiote (and as other) would potentially experience both greater gaming-related pleasure and distress. Finally, feeling a higher degree of control over avatars could be experienced as quite pleasurable, especially for players who experienced little to no control over their actual-world lives (connected, for example, to achievement-related feelings of mastery). Such feelings could promote a pleasurable sense of flow [71] and immersive escape. Or, instead, they might lessen the felt independence and thus reality of imagined others in the game, which might instead reduce the felt pleasures of immersive role-playing.

We summarize these ideas as follows:

Proposition 2. Players’ relative gaming-related pleasure and distress emerges in part from their specific relationships to their character-avatars, with some relating to characters in highly personal terms as extensions of the self or alternately as separate living persons or as impersonal objects or tools that help one accomplish gaming goals.

This second proposition (also logically abduced) helped us to explain new puzzles in the field. In this case, those puzzles related to differences we were noticing across gaming genres. For example, players of role-playing games—MMOs like WoW, but also tabletop roleplaying games (TTRPGs) like Dungeons & Dragons (D&D) and live-action roleplaying games or LARPs dramatically enacted in person—tended to relate to their characters as a me or symbiote. Players of other games like shooters or MOBAs, by contrast, instead related to avatars in more impersonal terms. And those differences also seemed to help us predict patterns in players’ gaming-related well-being.

This second conclusion also points to the iterative and recursive quality of ethnographic research, which Agar highlights. We iterated or repeated our ethnographic observations and interviews—and our engagement with relevant scholarship—until we arrived at a new conclusion that we felt advanced our understanding of how and why gaming could be so motivating for some players. Further, our prior conclusion—Proposition 1—was recursively included in this second conclusion. Earlier conclusions about players’ more personal relationship to their avatar-characters—as self-proxies or substitutes—are referenced in and even foundational to Proposition 2. However, in this case, current theory could only take us so far. The link between player-avatar relationships and gaming-related well-being had not yet been systematically addressed.

So, we set out to do that, as we describe in the next section. As readers will see, we present those findings in a different way. Specifically, we do so in relation to Pearl’s ladder of causality, showing how different ethnographic observations and interviews allowed us to move up in causal certainty from association to intervention to counterfactuals. That analysis shares many features with what has been called analytic induction and grounded theory [15, 17]. Following that presentation, we illustrate how we presented our understanding of those causal processes in a graphical causal model or DAG, which was amenable to more formal testing via a field survey and quantitative analysis. Confirming our earlier observations and interviews in that way would give us even greater confidence in our findings.

Strength of ethnographic evidence and the ladder of causality: Player-avatar relationships shape gaming-related well-being

Interventions

Naturally occurring events that that change the value of a predictor variable in ways that can impact a study’s outcomes—or what we refer to as interventions—can provide firmer evidence (compared to mere associations) for causal processes. With over a decade of experience in Dungeons and Dragons (D&D), our respondent Ash’s relationship to this game and their characters changed dramatically when they suddenly found themselves unemployed and decided to take up a career as a professional D&D game master (GM). This naturally occurring intervention (in Pearl’s terms) moved Ash from relating to their characters in more personal me terms to less personal terms as other persons or objects, which in turn reduced some of the pleasures and psychosocial benefits they originally derived from gaming.

Ash described their early characters as follows:

I was very much like most people who get into it. I was very attached to the characters. I just enjoyed playing a character that spoke to me at that moment in time. Whatever I am feeling, whatever shows I’ve recently watched. Ideas and sparks. And I like exploring that.

In this early period, Ash used the medium of the tabletop roleplaying game (TTRPG) to explore less obvious aspects of their identity and to experience life how they imagined themselves or wanted to be. Later, however, pressures from their gaming group led Ash to begin gaming less and less often as a player and more as a game master, who was responsible for running and arbitrating games. If Ash didn’t take up this role, they feared that his group would disband. Then, after five years of participating more casually in TTRPG play as both player and GM, Ash unexpectedly lost their job. To pay his bills, Ash decided to try his hand as a professional paid GM. By the time Ash was interviewed by a member of our research team, they were working online as a paid game host, helping to produce a D&D podcast, running their own online streaming channel, and guest starring in the podcasts and streams of other professional TTRPG content creators.

The transition from player to GM, and then from casual GM to professional player and GM, dramatically changed how Ash interacted with their own characters, as well as with other players and the game itself. The game had gone from, in Ash’s words, a “hobby and passion and turned to essentially my job now.” When talking about their characters after this transition, Ash described the new relationship as follows:

I don’t have any attachment to them anymore, because a lot of times those characters end up becoming one shot characters [i.e., temporary characters created for a particular gaming session or campaign]. Or it’s a very small series where we’re recording or streaming for, like at most ten episodes. So, I only get to be this character and in that character’s shoes for at most three months. There’s a disconnect as well, especially with safety tools and a lot of that stuff [rules that help protect players from experiences and content felt to be harmful]. I think that for me personally it’s made that bleed happen less and less often [i.e., less blurring of the boundary between the player and character]. I am me, the player, and this is my character, almost a third person perspective. I am no longer like ‘this character is me.’

Ash described the work of playing characters online in these professional groups as akin to putting on a theatrical production, where Ash was less able to invest emotionally in their characters and thus felt less connected to them.

By pursuing work as a guest player, meeting other content creators online, and inviting guests to their own shows, Ash also began playing with strangers. To accommodate the varied interests of players and their diverse backgrounds, professional GMs like Ash turn to safety tools to ensure they are not harming their players. Often safety tools involve players and GMs negotiating ahead of the game itself as to what topics and conflicts are acceptable in play. According to Ash, common topics of avoidance are racism, gender discrimination, and police brutality. As Ash mentioned above, the addition of safety tools contributed further to their feeling of emotional disconnection from their characters. Despite repeatedly mentioning in the interview that safety tools are an important part of their professional toolkit, it has also in part caused them to bleed less and less into the character in ways that earlier felt satisfying.

In those earlier periods, then, when playing casually, Ash was able to identify with their characters and pursue their personal passions through them. When their hobby became a profession and the game became work, Ash found themselves unable to immerse themselves in the game. The character was no longer me, but a “third person” or even an object. And the game was no longer relaxing and beneficial temporary escape from life’s stresses, but work itself and thus one of the sources of those stresses.

In this vignette, circumstances push Ash to shift from relating to their characters as a me to a more distanced and impersonal “third person” or even object kind of relationship. And this shift made it more difficult for Ash to enjoy and benefit from the game in ways that earlier came easily and naturally to them. These changes are brought about in part by what Pearl and others refer to as naturally occurring interventions. Members of Ash’s early play group pressure Ash to take on the role of GM, without which the group might dissolve. In that role, Ash becomes a game arbitrator who still manipulates other characters, but they are largely non-player characters—or “NPCs”—who, as bit players, are largely there as props who help advance the story and drama of which the player-characters are stars. Ash became an expert at using those NPCs as tools—objects—to advance the game and story. Their relationship to the game’s characters had changed, as had their original satisfaction playing their own characters. Then, financial pressures lead Ash to begin building a career as a professional GM. This furthers the erosion of earlier felt connection to their own characters, which was the foundation of how and why D&D was originally meaningful to them. Circumstances, then, intervene in ways that change Ash’s relationship to their characters, which in turn erodes many of the pleasures and psychosocial benefits they originally derived from gaming.

According to Pearl and others, naturally occurring interventions like these provide stronger causal evidence because many potentially confounding factors are held constant before and after the interventions. Ash is still (roughly) the same person before and after these changes, and they’re still playing the same game, in some cases with the same players in an identical gaming group. Ash, then, is their own statistical control, and this makes it more likely that the changes in their relationship to their characters—from me to other or object—cause the changes in the satisfaction they derive from gaming.

Of course, Ash’s reality is more complicated than this. It is more accurate to say that Ash shifted their gaming role from that of player to GM to eventually professional player and GM. Central to those transitions, we would argue based on our ethnography, is a change in how Ash related to their characters. But other factors were also associated with that shift in roles. For Ash, these included greater time investment in D&D (they learned how being a GM is hard work), more responsibility, and in some instances new gaming groups, situations, and expectations. We cannot rule out the possibility that these other factors—bound up as they were with the changed manner Ash related with their characters—erode Ash’s gaming pleasure and satisfaction. The interventions we describe, then, provide us with greater causal certainty compared to mere associations, because we can eliminate certain potential confounders as being causally determinant of gaming-related well-being in such situations. But the evidence is far from perfect, and we cannot rule out the potential importance of other factors. However, counterfactual evidence, as we show in the next section, can provide even greater degrees of causal certainty.

Counterfactuals

Video games like World of Warcraft (WoW), where players inhabit online virtual worlds, provide players opportunities to create many different characters, with each crafted in ways that meet players’ wants and game-related goals. Players of MMOs like WoW often have their primary character—called a main—often treated like an alter ego or me-symbiote. And they have secondary characters—alts—manipulate in ways that allow for exploration of other identity dimensions or to achieve particular gaming demands. The latter might include, for example, competitive player-vs-player mini-games inside WoW, amassing virtual resources (farming them) that help players achieve their goals (say, making potions that enhance players’ armor so they can collaborate with others in dungeons or raids where the aim is to defeat potent enemies called bosses), or simply serving as a mule that holds a players’ gaming property (serving as a kind of gaming bank). Online RPG players frequently—but not always—relate to their alts more as impersonal tools or objects that help them accomplish specific gaming tasks.

Most WoW players, then, have deep experience playing a variety of mains and alts. This allowed members of our research team to learn from players (via interviews and observations) how their relationships with their various characters might differentially shape the satisfaction and benefits they derived from WoW. Of particular interest, players we knew would typically settle on a main character, with whom they often had a deep and personal connection. And based on players’ prior experiences, they could (on our prompting) imaginatively contrast how playing that character was inherently different in terms of well-being outcomes compared to if that character were somehow different. Such imaginative what if explorations allowed us to contrast players’ actual current experiences with what Pearl refers to as counterfactual realities, which did not actually occur but could have. According to Pearl, such comparisons provide an even higher degree of causal certainty because the comparisons between what did actually occur and what instead might have occurred allow for a degree of precision in the isolation and manipulation of the predictor in relation to the outcome not possible in naturally occurring experiments or interventions.

Thus, take for instance the experiences of another respondent, Sam, an avid 24-year-old gamer, who played another video role-playing game called Neverwinter Nights. Sam plays a variety of computer role-playing games, regularly clocking in more than 60 hours a week. Until recently, Sam played predominantly male characters. Sam, however, had trouble relating with these male avatars. “Trying to connect with my male characters is hard… [I feel] disconnected from them.” It wasn’t until Sam realized that she was transgender that she was able to form a closer bond to her avatars by making them female like herself.

Because of my recent life choice, one that I’ve been debating on for a long time…I finally ended up saying ‘this is what I wanna be.’ Just being able to say that. Like my whole mindset for how my characters are or how my character should be ended up completely changing. It was because of this choice that made me think that maybe I want to be a female in these games, because maybe they would make me feel like I’m actually in the game.

Creating female avatars, and in some cases, changing those characters’ gender from male to female, allowed Sam to connect with them in a deeper, more meaningful way, and to see herself and her journey reflected in them. As she explains:

One of my favorite avatars is from Neverwinter Nights. Her name is Luna, she is a warlock that is a follower of the sun god. She is what I imagine myself to be. I’ve wanted to be a woman for a very long time, and wanting to practice magic, wanting to learn more esoteric wisdom to help and assist the people around me, these things are how me and Luna relate to each other. Actually, Luna was a (re)creation from an old character, a male one.

Sam ended up creating another female character as well, this one an evil necromancer, as she tells us:

In a word, I would describe her as Internal Darkness. She is everything I have yet to heal in myself, everything I have yet to let go. She is the demon I have yet to fight. I would characterize her as blind rage… I’m bad at letting my emotions go. I tend to bottle a lot up and anger management has always been something that I’ve had trouble with. Without check, the people around me could be hurt, so I would make it into something physical or “digital” so that I could have my anger in something other than myself.

Reflecting on her two favorite characters, she realizes that they are both “me” avatars, though different sides of her, like Jung’s ego and shadow:

One holds the image of me toward the future and the other is what I leave behind. Luna and the necromancer were creations of my current emotional state.

As Sam goes through the process of discovering her own femininity, her female avatars become embodied expressions of her psyche. Being able to connect and identify with them has allowed Sam to work through her past “demons” and work towards actualizing the woman she wants to be.

In some ways, Sam’s experiments with her characters resemble what Pearl calls interventions: Sam personally connects in variable ways to her evolving main and her various alts; these feelings of connection (or lack of connection) in turn shape the degree to which Sam derives emotional and well-being benefits from her gaming. In fact, comparing how Sam plays her main and her alts would be illuminating. However, Sam’s deep gaming experience also allows us to imagine with her (in interviews) a counterfactual world where her main Luna is not female (as she is in reality) but instead male (as she might be counterfactually). And Sam tells us in no uncertain terms that such a counterfactual reality would be much less satisfying, as it would feel inauthentic and “fake.” That is, she would never be able to personally connect with such a male avatar, nor successfully immerse in the game via such an avatar in ways that felt satisfying to her. Ethnography, then, provided us with opportunities to jointly imagine with Sam a counterfactual world where Sam never came to understand how her characters should be female instead of male and instead continued to play male avatars. In this alternative reality, Sam would never have created her warlock nor her necromancer. And as a result, she never would have found ways to work through her inner struggles via her characters.

Both Pearl’s interventions and counterfactuals provide greater degrees of causal certainty compared to associations. However, counterfactuals, unlike interventions, allow for a more precise isolation and manipulation of a key causal factor. In this case, simply changing Luna’s gender from male to female is the key to allowing Sam to connect in a deeper and more personal me way to her character, which in turn leads her to experience deeper forms of gaming immersion and satisfaction. Then, imagining what might have been if Sam had not made that simple gender change provides us with insight into how that single factor impacts Sam’s gaming satisfaction. In our prior intervention example, by contrast, Ash’s diminished personal me relationship to their D&D characters was inextricably bound to their changing gaming roles (from player to GM to professional GM) in ways that made it more difficult to isolate the former’s causal force compared to the latter’s. And likewise, simply comparing how Sam plays her main and her alts is bound up with other factors. However, ethnographically informed, counterfactual thinking also allows us to imagine the impact of potential confounders to Sam’s experiences. We can ask if Sam’s experiences would differ if she played less than 60 hours a week? Perhaps, but many players, who game far less, still report meaningful connections to avatars. What about Sam’s age? Again, there is nothing theoretically or ethnographically to make us think that Sam’s relationship to her avatar would be meaningfully different if she was 5-years older or younger. And Sam’s gender? Yes—we can imagine that if Sam was not transgender, playing female avatars would not have the same effect. So, it’s the combination of both Sam’s and the avatar’s gender that allows this experience to happen. In other words, counterfactual thinking allows us to manipulate the variables in the model to determine which drives the causal relationship.

Representing our ethnographic findings in a graphical causal model (DAG): Clarifying causal processes and mechanisms with mixed methods research

Fig 2 presents graphically our ethnographic findings discussed in the prior sections. Our ethnography convinced us that the way players relate to their avatars can shape the quality of their gaming experiences and how those experiences contribute to their overall well-being. So, player-avatar relationship is represented in the model as the key predictor variable, and gaming-related well-being is the principal study outcome. That graphical model includes variables that might bias or confound estimates of the path between our key predictor and outcome. As revealed to a large extent in our ethnography, those variables include: gaming genre (RPGs vs. other games); propensity to have immersive experiences (both Ash and Sam highlight this); degree of involvement with gaming; social support for gaming; age; and gender. Our ethnographic research also led us to expect that distinctive emotional experiences associated with avatars would shape gaming-related well-being, including by partially mediating between one’s relationship with their avatar (our primary predictor) and positive and negative gaming-related well-being (our outcomes). That relationship is also encoded in this graphical model.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Research model.

https://doi.org/10.1371/journal.pone.0302857.g002

Particularly important to causal explanation is not only whether a change in a predictor caused a change in an outcome but how and why a predictor effected that change. Such variables or factors that constitute the mechanism by which some cause produces an effect, as just mentioned, are commonly known as mediators. Connected as they are to naturally occurring events as they occur in the field, ethnographers are well-positioned to identify such factors. Of note, in social science research, mediating factors may include how study participants interpret and respond to changes in the predictor of interest [19–21, 73]. That is why understanding native points of view via long-term fieldwork—as interpretive anthropologists like Geertz remind us [10, 74]—is so important to anthropological understanding and (in our case) causal explanation. In fact, this concern with the connection between native experience and causal explanation is broader than the interpretive anthropological literature we cite here. For example, Ed Hutchins’ “cognitive ethnography” [75] highlights how cultural perceptions of how the world works—what are sometimes referred to as “cultural models” [44, 45, 76]—shape human motivations in ways that direct behavior and thus should be accounted for in causal explanations [77]. And outside of anthropology, Jon Elster’s influential work on “intentional explanation” highlights the important way that motivated social action—an agent’s perception that certain actions will lead to desirable changes in the world—importantly influences how social processes unfold in the world [37].

Relevant to this discussion on causal mediation, we also identified in our gaming research how the causal connection between player-avatar relationships and players’ well-being could be mediated by distinctive emotional experiences. Via avatar-characters, roleplaying game (RPG) players experienced intense emotions, which we documented extensively in fieldnotes and explored further in interviews. Those included social and moral emotions like pride (in their characters’ acts of heroism) and empathy (with their own avatars’ struggles, with others’ characters, and with other players), as well as other feelings like joy, wonder, hope, and gratitude. But conversely, some RPG players also reported that they could emotionally invest too much in avatar-characters. That could lead to too strong of a blurring of fact and fiction, intense negative and even toxic emotional experiences (e.g., feelings of deep shame, anger, and frustration), with the potential to disrupt their non-gaming lives. Players of other game genres could also experience similar highs and lows. For example, shooter games like Counterstrike—Global Offensive could be highly competitive, and sometimes felt to be “toxic” for these reasons, thus shaping players’ well-being in distinctive ways, as we learned in India [78]. But overall, we found that RPG players’ highs and lows were more often linked to the deeply personal relationships they formed with their avatar-characters as compared to some other cause, whereas gamers who more typically related to avatars as objects were more shielded from such avatar-related effects.

In the current discussion, we would emphasize that the causal processes shown graphically in Fig 2 only represent a small slice of reality: how player-avatar relationships impact gaming-related well-being. Other factors have also been shown to influence how games can come to contribute positively or negatively to players’ overall well-being, including particular games’ reward systems and the social context of play [78–81]. However, Fig 2 only presents the main causal relationship of interest to our study, along with any other factors (like confounders and mediators) that could affect or bias how we understand that primary causal pattern. Furthermore, Fig 2, as a qualitative representation of causal processes, does not say anything about the magnitude or functional form of any of the postulated causal pathways we show.

Nonetheless, representing our ethnographic findings in this way does allow the possibility of further testing with quantitative data the relationships we posit. Such tests could help us confirm our ethnographic findings and provide estimates of the magnitude of causal relationships we posit. With this in mind, and based on our ongoing ethnography, we constructed and distributed a field survey to North American gamers, and also to gamers from other parts of the world, which included: Argentina, Europe (Finland, France, and Italy), and Asia (China and Singapore) [82]. Linear regression analyses, then, allowed us to test hypotheses of relationships between different avatar relationships and gaming-related well-being, with us modeling relationships between our study predictors and outcomes using an ordinary least squares (OLS) approach [83]. To minimize bias in the estimation of the relationship between our key predictors and gaming-related well-being, we included other factors in our analysis as control variables that our ethnography and current theory suggested were likely to be causally prior influencers of both our main predictors and outcomes. For our North American sample, initial analyses reveal our survey data to be consistent with key causal patterns we postulated. For example, players reporting a more personal relationship with their avatars, as compared to relating to avatars as an object, show a substantial positive relationship to both positive and negative gaming-related well-being. (This working manuscript [83] can be made available upon request by emailing the lead author).

Here, we would emphasize that field surveys and their accompanying quantitative data and analysis, while sometimes serving as useful tools in an ethnographic toolkit [50], are not necessary to ethnographic analysis [51]. By this, we mean that ethnographers can use field surveys to sample relatively quickly and efficiently a broad range of respondents, though with less depth, for example, compared to qualitative interviews [60, 84]. Quantitative analysis of field survey data using techniques like regression analysis can also allow ethnographers to consider how the relationship between a predictor and outcome might change when considered alongside multiple confounders, all considered simultaneously together in the analysis [85], in ways that remain challenging in analyses of qualitative fieldnotes and interview transcripts. Further, field survey data can be analyzed via new methods in causal mediation using graphical models and counterfactual logic [29, 53, 86]. These statistical routines calculate what Pearl refers to as natural direct effects (DE), natural indirect effects (NIE), and total effects (TE). Pearl uses a counterfactual logic that mimics experimental research design and thinking where he examines how the relationship between a predictor and outcome varies with the mediator set at different (not actually observed and thus “counterfactual”) values. For his natural direct effects, say, the predictor can vary between 0 (no treatment) and 1 (treatment). Pearl then examines how moving from 0 to 1 in the predictor is associated with changes in the outcome when the mediator is set or fixed in the analysis ("set" logically and thus not actually observed) at the naturally occurring value it would have for each individual when the predictor takes on the value of 0 (the no treatment condition). For his natural indirect effects, the predictor can again vary between 0 (no treatment) and 1 (treatment). In this instance, Pearl examines how moving from 0 to 1 in the predictor is associated with changes in the outcome when the mediator is set or fixed in the analysis at the naturally occurring value it would have for each individual when the predictor takes on the value of 1 (treatment). In Pearl’s framework, the total effect is the sum of the natural direct and indirect effects. These new counterfactual causal mediation routines are found in Stata 18, for example [87].

Nonetheless, we have shown throughout this article how qualitative analysis of fieldnote and interview data can clarify in different ways the relationship between predictors and outcomes, while accounting for confounding and mediating factors. We would thus encourage ethnographers not to think that the only causal explanation worth presenting is an ironclad and standalone one, say, that accounts for all confounding factors in all their multiple combinations. Rather, ethnographers working primarily with qualitative data might instead consider how the work they do can complement and contribute to research carried out using other methods, including in team-based collaborative work [51, 72]. Indeed, they might profitably think instead how single ethnographic studies of causality—however imperfect—might be subsequently combined with and synthesized into larger “meta-ethnographies” [88–90].

Prospects

We have aimed to extend current ethnographic approaches to causality by highlighting the abductive logic of ethnography and by connecting more directly to interdisciplinary ways of assessing the strength of causal evidence and using graphical models to clarify causal patterns. Nevertheless, other topics need further clarification and explicit discussion of how they might be incorporated into current ethnographic approaches to explaining causal processes. These include: the role that logical induction and deduction play in relation to abduction in ethnographic analysis; the extent that induction, deduction, and abduction approaches are necessary or sufficient in the logical definitional sense for developing causal ethnographic arguments; how ethnographic data might be coded and analyzed to clarify potential moderation effects, where one factor changes the relationship between other factors; greater attentiveness to selection bias and collider effects in ethnographic data; using ethnography to develop culturally valid measures that can be employed in causal studies; integrating the current approach with field experiments or longitudinal data; more refinement of ethnographic analysis of interventions and counterfactuals, say, by documenting ethnographically naturally occurring experiments or developing interview protocols to explore more effectively counterfactual realities; attention to how different kinds of ethnographic data—say, observations versus interviews—differently inform causal analysis.