CHALLENGING DOMINANT INTERPRETATIONS

It may seem a trivial point to state that data interpretation requires caution and critical thought. However, some types of error are easily hidden by baseline assumptions, thereby presenting a problem so elusive that even careful scholars might not see it. Thus, scholars can fall prey to errors in interpretation that reproduce the same incorrect insight—which we believe to be accurate—until someone challenges our assumptions. Members of the researched community can raise such challenges. I illustrate this with an example from my own work on Roma/non-Roma relations in Slovenia and Croatia.

The Roma are Europe’s largest ethnic minority, and they are known derogatively as “Gypsies.” Of the approximately 14 million Roma around the world, 10 million to 12 million live in Europe (Matache and Mark Reference Matache, Mark and Bhabha2014). The Roma minority is diverse, complex, and constantly evolving (Matache and Mark Reference Matache, Mark and Bhabha2014). Although its members occupy all walks of life, many face widespread discrimination and systemic exclusion. In my work, I focus on Roma who are excluded and non-Roma who contribute to that exclusion.

Almost universally, the Roma are stereotyped as cheaters and thieves. This stereotype is so pervasive that it is used in everyday language. In the United States, for example, people often claim they were “gypped” when they feel cheated—sometimes without realizing the true meaning of the word. In my conversations with non-Roma, I often heard the following: “We don’t discriminate against the Roma because we hate them; we treat them differently because of how they behave.” When they claim this, they are using the concept of statistical discrimination. This concept is typically contrasted with animus (Becker Reference Becker1957), or hatred, and suggests that disparate treatment can stem from rational optimizing behavior that relies on group characteristics used as proxies for unobserved individual attributes (Arrow Reference Arrow, Ashenfelter and Rees1973; Phelps Reference Phelps1972). Thus, the commonly employed rationalization of disparate treatment is as follows: if one expects that the Roma will steal or cheat whereas non-Roma will not, it makes sense to treat the Roma differently.

My first study of Roma/non-Roma relations in Slovenia focused on discrimination against the Roma and therefore explored non-Roma behaviors (Bracic Reference Bracic2016). To capture individual engagement in discriminatory behaviors, I used the trust game (Berg, Dickhaut, and McCabe Reference Berg, Dickhaut and McCabe1995). In other contexts, this game has been used chiefly to measure trust and trustworthiness; in this study, however, the game can be used to capture discrimination. The trust game maps well onto the damaging stereotype about the Roma because defecting in the game—although allowed under the rules and not actually cheating—often is interpreted by non-Roma as cheating. Consider the following rules.

The trust game has two players: a sender and a receiver. Each player receives an initial endowment (i.e., 6 euros) and learns about the rules of the game. The sender then has the opportunity to send some, all, or none of her endowment to her receiver. Whatever she does not send, she keeps. The amount that she sends, if any, I double; the receiver then receives this doubled sum in addition to his initial endowment. He then has the opportunity to send some, all, or none of the total pot back to his sender. Whatever he does not send, he keeps. His partner receives whatever he sent, and the game ends.

I paired non-Roma senders with randomly chosen Roma or non-Roma receivers to determine whether non-Roma senders would systematically send less to Roma receivers, based on the expectation that Roma receivers would not return anything or a large enough share of the total pot. In some localities, non-Roma senders discriminated, sending about 30% less to Roma than to non-Roma (Bracic Reference Bracic2016). When I describe these results, whether in an academic or a social setting, people often respond: “Yes, but how did the Roma behave? Did they send anything back?”

The answer to this question has three parts. First, the test was not designed to compare behaviors of Roma and non-Roma; rather, it was designed to compare behaviors of non-Roma senders. Comparing how much Roma and non-Roma receivers return to their non-Roma partners would be inappropriate because non-Roma receivers are returning money to in-group members, whereas Roma receivers are returning the same to members of an outgroup that has excluded them for generations. Second, Roma receivers sent money back to their partners; however, on average, they sent less than non-Roma receivers. Third, branding the behavior of non-Roma senders as statistical discrimination is deeply problematic.

Exercising statistical discrimination is predicated on the ability to compare one entity to another. In constructing comparisons, we are inclined to determine a baseline to which everything is compared; in group comparisons, the dominant and privileged group inevitably becomes the baseline. The behaviors, opinions, and preferences of its members are normalized, whereas those of the outsiders are understood in reference to that baseline. Implicit in the question “But how did the Roma behave?” is such a comparison: behaviors of Roma are considered and interpreted against the baseline of non-Roma. But they shouldn’t be. Whereas the overall unemployment rate in Slovenia was between 7% and 9% in 2016 (Statistični urad Republike Slovenije 2017), among randomly sampled Roma participants, it differed by an order of magnitude: 80%.

Recall that the highest possible sum that a receiver in a trust game could have is 18 euros (i.e., the sender sends all 6 euros, which are doubled and then added to the receiver’s initial 6 euros). For a typical Roma receiver, the 18 euros hold a different value than for a typical non-Roma receiver. The former might be able to feed his family, whereas the latter might buy an extra movie ticket. Although this difference is clear, directly comparing Roma and non-Roma behaviors assumes that giving up 10 euros and keeping 8 would be as easy for the Roma as it is for the non-Roma.

The idea that something so obvious would not immediately occur to a scholar might seem unlikely. Yet, among the scholars who suggested that I was observing mere statistical discrimination, as well as those who participated in subsequent discussions, not one questioned whether the concept was appropriate within these circumstances (and neither did I, at first). The concept of statistical discrimination is a familiar one. In reaching for a familiar theoretical framework in which to contextualize the difference in Roma and non-Roma behaviors, it is easy to lean on it. Perhaps the reason lies in the automatic baseline assumption: to uncover the limitations of statistical discrimination in this context, we would first have to question the tendency to perceive the privileged group as the baseline. To many members of dominant groups, this insight is not obvious (Haney López Reference Haney López1996). To most—if not all—members of non-dominant groups it is.

After I completed the fieldwork, a Roma participant remarked (referring to the game) that “we Roma often cannot send a lot, but we’ll always send something.” I thought nothing of it at the time; however, as I tried to figure out why statistical discrimination seemed limited, I kept coming back to this statement. I returned to the Roma community for a conversation about this only after I thought I had worked out the answer. My aim was to present it to them and see if they agreed; they did. Most notably, one said that “comparing Roma and non-Roma like that is unfair, as our circumstances in life are so different.”

The insight is as follows. Statistical discrimination and discrimination based on animus tend to be viewed as concepts that are entirely separate and contrasting: the former is often seen as acceptable, even rational, whereas the latter is unacceptable. Sometimes, however, the two concepts are not divorced at all. When non-Roma actions are considered the baseline and the Roma are assumed to make their decisions as if their circumstances mirrored those of non-Roma, systemic inequalities are rendered invisible. Yet, those systemic inequalities—largely based on past and present exclusion and animus-based discrimination—render many Roma unable to return as much as non-Roma in the trust game.

Statistical discrimination would consider each interaction between two people unrelated to any past or present context. Where profound inequalities exist between the people interacting, the implicit judgment that statistical discrimination is acceptable provides a guilt-free path to maintaining inequality. The non-Roma sender can use rationality to justify current disparate treatment without any reference to historical and present marginalization of the Roma, or his own privilege as a non-Roma. The Roma receiver, however, can neither escape historical trauma (Sotero Reference Sotero2006) nor interact with the non-Roma sender on the same terms as a non-Roma receiver would. In such contexts, statistical discrimination fails to describe reality—that is, interactions between individuals do not happen in a vacuum—while, at the same time, it provides an easy justification for maltreatment.

Spurred by an offhand comment from a Roma participant, I eventually arrived at this insight on my own and only consulted the Roma community afterwards. But what if I had asked them earlier? Indeed, in light of this experience, the question of whether we should share our puzzles and research questions with the communities we study seems misplaced. Rather, the better question to ask is when we should share. Should we consult with the community after we have already developed our own theories and explanations? Or should we consult as soon as possible after the puzzle or a challenge emerges because their answers (regardless of what they are) may set us on the correct path faster? Or should we perhaps consult with the community even earlier, before the study begins, so that the scholar and the community develop the research question together? The principles of community-based participatory research provide a good starting point for thinking about this possibility (Minkler Reference Minkler2004).

In my case, conversations with the community shaped subsequent work. The trust game captured disparate treatment, but I was unable to determine to what extent statistical discrimination, as opposed to animus, was responsible. In a follow-up study, I created a videogame in which participants repeatedly interacted with Roma and non-Roma avatars whose behaviors were identical. This provided no justification for statistical discrimination but created instead an opportunity to learn and adjust. The videogame and expressions of anti-Roma sentiment I gathered through surveys and conversations with non-Roma participants (e.g., “theft is in Roma blood” and “the Roma steal our dogs and eat them”^{Footnote 1}) allowed me to more directly target the issue of statistical discrimination and animus.

ASSESSING THE RELEVANCE AND FIT OF THE TEST AND DELIVERY METHOD

Engaging the researched community in the process of research design can help determine whether our tests are appropriate for the local context. Before fielding the trust-game study, I spoke to members of Roma and non-Roma communities to see what they thought of the game approach more generally and of the trust game specifically. To clarify, the individuals I consulted were not those who participated in the study, although the Roma were from the same locality. At those meetings, I described the game and its delivery and explained the underlying logic. Because I used the games to capture and analyze behaviors of non-Roma senders, my questions for the Roma were primarily about whether they thought the game would capture disparate treatment. The Roma I spoke with thought that the game presented an excellent way of eliciting the type of behavior that they face in daily discrimination. They also thought that directly asking non-Roma whether they discriminate would not be as effective because some may not admit it but might engage in it unwittingly as trust-game senders. Had I asked them at the time whether the trust game was appropriate for comparing behaviors of Roma and non-Roma, they likely would have said that it was not; that conversation, however, followed much later.

Exploring whether a test is appropriate before fielding it is sometimes more challenging. Concerns of contagion—wherein details about the study are spread through interpersonal communication—for example, might restrict when and with whom an experiment can be discussed before fielding it. When concerns of contagion are substantial, how can we gain insight from community members? We might consult with members of the community but hold back a few details. We might coauthor with a scholar who is a member of the community. We might talk to members of a community from a different but similar locality, distant enough but likely to offer similar insight.

This is what I did when I spoke to non-Roma in Slovenia before fielding the trust game. Because I am a Slovene non-Roma myself, obtaining access to Slovene non-Roma that were not from the researched localities was easy. If I had tried to do this with a different Slovene Roma community, from a locality where I have no established connections, I would have needed more time and resources, which unfortunately were scarce.

Resource constraints can prevent scholars from conducting the best possible field or lab-in-field experiments. With fewer or no constraints, however, we can pursue designs that allow both theory building that actively engages the community and subsequent theory testing in other localities, with sufficient contagion precautions. This approach may only work for some research questions and not others. But we should at least consider the possibility.

In addition to validating our methods, members of the researched community can be indispensable regarding delivery. The reason is obvious: they know their own communities much better than we do. For example, to determine which method of administering a survey would be most appropriate, I created with members of a Roma community several versions and piloted them. This addressed the variation in literacy levels among the Roma: elderly Roma and women in particular are more likely to be illiterate and I did not want to systematically exclude this segment of the population from participating. We tested several versions; the most elaborate involved a recording of the questions that a participant would listen to on headphones and answer by pressing differently colored squares on a tablet. However, to my surprise but not theirs, I ultimately decided that the classic paper-pen-and-interview option was best.