Individual behavior is increasingly salient in the context of rights abuses. Human rights monitors look harder for abuse, find abuse in more places, and classify different types of abuse as human rights violations (Fariss Reference Fariss2014). As those violations are more broadly monitored and classified, holding individuals accountable is becoming a norm (Sikkink Reference Sikkink2011). The U.S. #BlackLivesMatter movement, reacting to police brutality against black citizens, both exemplifies the growing movement towards accountability and reflects the individual-level link between discrimination and other rights abuses. Nevertheless, the global rights community has far to go in battling racial injustice and other rights abuses at the level of the individual, whether directed at Dalits in India, individuals of Haitian descent in the Dominican Republic, or members of any other marginalized group. Human rights scholars rarely systematically gather individual-level data about human rights behaviors to link them to broader human rights institutions. By examining efforts to eliminate individual-level discrimination which itself often results in further rights abuses, this article aims to provide insights into macro- and microlevel processes that potentially lead to positive human rights outcomes.

Using the context of discrimination against the Roma in Slovenia and Croatia, I test the ground-level effectiveness of two strategies. The first strategy is a top-down combination of incentives and norm promotion; I ask if it affects individual behavior. I then test the same ground-level effectiveness of the second strategy, a local NGO-led effort to improve contact between those who discriminate and those who are targets of discrimination. In testing the first strategy, I conduct a border study of two closely matched towns in Croatia and Slovenia. The EU accession process provides incentives and norm promotion in Croatia, but not in Slovenia. In examining the second, I perform a within-country comparison of two closely matched Slovene towns. One has an NGO effort to improve contact and the other does not. I measure discrimination at the individual level in an experimental setup, using the trust game (Berg et al. Reference Berg, Dickhaut and McCabe1995).Footnote ¹

The top-down rights improvement strategy does not appear to reduce discrimination on the ground, but the bottom-up NGO effort does. The findings suggest that incentive-based and traditional norm-promoting strategies, while powerful at the state level, may not have effects that reach individual behavior. The results further suggest that in targeting individuals, a human rights strategy that is experiential and not merely persuasive might be particularly effective.

Human rights scholarship tends to examine the two components of the first strategy separately. Incentive-based mechanisms are generally aimed at state actors, and affect rights practices through rights conditionality in exchange for benefits like preferential trade privileges (Hafner-Burton Reference Hafner-Burton2005; Simmons et al. Reference Simmons, Dobbin and Garrett2008). Norm promotion addresses states and individuals alike, and affects rights practices and public opinion chiefly by naming and shaming (Brysk Reference Brysk1993; Davis et al. Reference Davis, Murdie and Steinmetz2012; Hafner-Burton Reference Hafner-Burton2008; Hendrix and Wong Reference Hendrix and Wong2012; Keck and Sikkink Reference Keck and Sikkink1998; Khagram et al. Reference Khagram, Riker, Sikkink, Khagram, Riker and Sikkink2002; Sikkink Reference Sikkink1993). While these two types of strategies are quite distinct, they are sometimes linked in practice. When a state commits to rights improvements in exchange for externally granted benefits or when such benefits are suspended in light of rights violations, norm entrepreneurs use the opportunity to mobilize. In fact, suspension of benefits may in part be due to naming and shaming. Human Rights Watch, for example, reports extensively on inadequate uses of the African Growth and Opportunity Act (AGOA) conditionality in improving rights practices in various African countries (Kasambala Reference Kasambala2014). When the United States revoked Swaziland’s AGOA status due to insufficient protections of workers’ and human rights, the decision accordingly received much attention (Hughes Reference Hughes2014). The two strategies, while chiefly operating at state and international levels, at least partly rely on domestic audiences to mobilize in favor of rights change (Keck and Sikkink Reference Keck and Sikkink1998; Simmons Reference Simmons2009). An individual actor observing or participating in public discourse, however, might not distinguish between incentives and norm promotion, but simply perceive that rights should be changed. In studying individuals, then, the effects of two strategies that often appear together ought to be considered together as well.

An alternative strategy to change discriminatory behavior, much smaller in scope than the approaches just mentioned, stems from NGO efforts that encourage contact between the majority population and the marginalized group in order to foster inclusion. While the merits of intergroup contact remain contested, laboratory-based scholarship in psychology strongly suggests that intergroup contact reduces prejudice and discriminatory intent (Allport Reference Allport1954; Dovidio et al. Reference Dovidio, Gaertner, Nier, Kawakami, Hodson and Miller2004; Miller Reference Miller2002; Pettigrew Reference Pettigrew1998; Pettigrew and Tropp Reference Pettigrew and Tropp2006; Wagner et al. Reference Wagner, Christ and Pettigrew2008). Recent evidence from field experiments offers additional support; when white children (Green and Wong Reference Green, Wong, Borgida, Federico and Sullivan2009) or college students (Boisjoly et al. Reference Boisjoly, Duncan, Kremer, Levy and Eccles2006; Laar et al. Reference Laar, Levin, Sinclair and Sidanius2005) are randomly assigned to diverse groups for short-term or extended interactions, they show lower levels of prejudice. Efforts on the part of NGOs to increase contact between those who transgress and their targets at the individual level may, then, decrease prejudice and discrimination at that level as well. Most research on intergroup contact is conducted in the American context, predominantly examining prejudice of whites towards blacks; by studying discrimination in Europe, this article offers a departure from this tradition.

I conduct a test of the intergroup contact strategy and the combination of incentives and norm promotion in the context of discrimination against the Roma in Slovenia and Croatia. The Roma, commonly referred to by the disfavored term “Gypsies” (Hancock Reference Hancock2002), are the largest ethnic minority in Europe. Historically, the Roma have been variously enslaved, deported, forcibly assimilated, and subjected to genocide (Barany Reference Barany2002). Today, many remain segregated and face discrimination in schools, on the job market, in hospitals, in police stations, and on the street (European Roma Rights Centre (ERRC) 2004). This study examines the Roma because they are a clear case of a minority that has suffered discrimination, abuse, and exclusion for centuries. In such a seemingly intractable case, identifying interventions that promote inclusion would show promise for studying and curtailing marginalization in various other communities that historically have been denied or have themselves resisted inclusion.

My empirical strategy involves two approaches: a border study and a within-country design. I use a border study to examine the combination of incentives and norms. The EU accession process is arguably the strongest mechanism of human rights conditionality, and is deemed particularly powerful in the context of Roma rights (Schimmelfennig et al. Reference Schimmelfennig, Engert, Knobel, Schimmelfennig and Sedelmeier2005). To gain membership, EU candidate states must enact legislative and policy changes that protect Roma rights, at state and local levels. Conditionality demands are accompanied by norm advocacy and naming and shaming, via NGOs and official EU channels. I leverage this process and measure discrimination against the Roma in two towns, one in Croatia and one in Slovenia, that match closely on 43 Roma and human rights related factors at state, regional, and local levels. At the time of investigation, Croatia was at the height of the EU accession process and thus faced enormous pressures to improve its treatment of the Roma. In contrast, Slovenia had been an EU member for eight years and faced no pressure to improve Roma rights.

I test the intergroup contact strategy in the context of discrimination against the Roma in two towns in Slovenia that match exceptionally well on 12 local/regional Roma and human rights-related factors, but differ in type of NGO action. Action in the first town is inclusive—targeting both Roma and non-Roma—and promotes intergroup contact. NGO action in the second town is not inclusive and focuses on service provision to Romani settlements. Accordingly, roughly half of randomly sampled non-Roma from the first town are familiar with the NGO; in the other town, the same holds for only two percent. NGO leaders’ personal idiosyncrasies completely define the organizations’ foci. This mitigates the endogeneity concern that the two types of NGOs developed because of beliefs that their types would be particularly effective in their respective environments.

To test the two hypotheses, I use simple games intended to capture other-regarding behavior, which map remarkably well onto the stereotype that motivates anti-Roma sentiment. Discrimination against the Roma is largely motivated by the belief that the Roma are cheaters and thieves. The trust game (Berg et al. Reference Berg, Dickhaut and McCabe1995) played with money elicits behavior by non-Roma that demonstrates distrust towards the Roma; given the strong link between distrust and discrimination in this particular case, the method likely captures discriminatory behavior.

I find no support for the hypothesis that a high-level combined strategy of incentives and norm promotion affects individual behavior. At the height of the EU accession process, Croats discriminated against the Roma, while Slovenes just across the border, under no pressure to improve their treatment of the Roma, did not. I find support, however, for the intergroup contact hypothesis. Slovenes from a town with inclusive NGO action treated Roma no differently than non-Roma, whereas Slovenes from a town with noninclusive NGO action discriminated against the Roma.

My findings suggest that (1) while powerful at the state level, incentive-based and traditional norm promoting strategies inherent in the EU accession process may not necessarily have effects that reach the individual, and (2) NGOs promoting intergroup contact between Roma and non-Roma can help reduce discrimination. In the context of changes at the individual level, then, microlevel strategies that engage individuals as participants in normative behavior may be preferable to (otherwise powerful) macrolevel strategies that merely advocate for it.

Quantitative human rights scholarship tends to view rights violations in a top-down manner, with the state as an abuser and the citizen a victim. Some cases of abuse are not as clear-cut. State actors may be the primary perpetrators, but their behavior may reflect the general tenor of the environment. For example, citizens may implicitly or explicitly support abuse. In India, government food security programs are often situated in locales that exclude Dalits; if not, Dalit children are often chased away from schools when they try to take advantage of the midday meal scheme to which they are entitled (Thorat and Lee Reference Thorat and Lee2005). Actions of parents, teachers, and local administrators are intertwined; parents mobilize in favor of Dalit exclusion (Thorat and Lee Reference Thorat and Lee2005), while the officials curb the children’s right to food. Further, citizens can themselves become state actors. Consider American juries. In murder cases of white victims, black defendants are significantly more likely than white defendants to receive the death sentence (Baldus et al. Reference Baldus, Woodworth, Zuckerman, Weiner and Broffitt1998), particularly so if they look more stereotypically black (Eberhardt et al. Reference Eberhardt, Davies, Purdie-Vaughns and Johnson2006). Some cases of human rights violations, then, demand that we examine individual actors, both state and nonstate, if we aim to understand and improve human behavior. This article offers a small contribution to that effort.

In what follows I first introduce the Roma. Next, I present the two strategies of human rights change along with expectations about their efficacy at the level of the individual. The empirical strategy, results, and a broader discussion of implications follow. The last section concludes.

CASE SELECTION

I measured discrimination levels in two overlapping pairs of towns, one pair per hypothesis. While differing on the crucial covariate—EU accession process or inclusive Roma NGO activity,—the two town pairs match closely on a set of factors that may influence human and Roma rights, at state, regional, and town levels. To select the towns I performed nearest neighbor matching (Ho et al. Reference Ho, Imai, King and Stuart2007a, Reference Ho, Imai, King and Stuart2007b; Nielsen Reference Nielsen2014),Footnote ¹⁰ using seven covariates, on all towns in Slovenia and Croatia that did not experience war violence in the Yugoslav wars in the early 1990s (The Central Intelligence Agency 2002) and have a Roma population of at least 50.Footnote ¹¹ I excluded locations that experienced war violence in order to control for the aftermath of ethnic contention that may have been more, though not uniformly, present in certain Croatian locations.Footnote ¹²

Of the seven covariates used in matching, town population, Roma inhabitants as proportion of total population, and ethnic majority as proportion of total population were used to achieve balance in town sizes and their ethnic compositions. Proximity to Slovene/Croat border and a triborder region dummy (Hypothesis I) were included as an additional control for variation in proximity to outgroups. People from borderlands are significantly more likely than inlanders to have contact with people across the border (Mirwaldt Reference Mirwaldt2010), which may increase the saliency of ingroup sentiment (Branton et al. Reference Branton, Dillingham, Dunaway and Miller2007; Mirwaldt Reference Mirwaldt2010). Ingroup favoritism can, in turn, give rise to intergroup discrimination irrespective of attitudes toward specific outgroups (Brewer Reference Brewer, Kruglanski and Higgins2007). Controlling for increased proximity to a border therefore accounts for a potential source of variation in intergroup discrimination more generally.

A regional capital dummy was included as a proxy for resources and institutional capabilities at the municipal level, both general and those pertaining to the Roma. Of the 24 Slovene municipalities in which Roma live, for example, only four have a municipal strategy for addressing the needs of their respective Roma communities (Vlada RS 2014). Both regional capitals in the sample have such a strategy. Finally, dummies for the EU accession process (Hypothesis I) and inclusive Roma NGO action (Hypothesis II) selected for the two main categories tested.

The pair of towns used to test Hypothesis I was selected first; the pair used to test Hypothesis II was selected second. To test the two hypotheses with the lowest possible number of subjects, participants were recruited from three towns; data from one town is therefore used to test both hypotheses. Matching singled out Čakovec in Croatia and Murska Sobota in Slovenia as the best pair to test Hypothesis I; Murska Sobota and Novo mesto, also in Slovenia, were selected as the best pair to test Hypothesis II (see Figure 1 for a map). Beyond the factors used in matching, the locations match on 38 additional relevant characteristics. The remainder of this section first discusses how the towns differ on the crucial covariates, and then lists the shared characteristics.

FIGURE 1.

A Map of the Three Towns

Hypothesis I: The EU Accession Process

A meaningful study of discrimination in the context of EU accession and membership would necessarily span several years if limited to one country. To test Hypothesis I within a short time span, I looked at cross-sectional snapshots of discrimination in towns of which one was undergoing the accession process and the other had already completed it. As Roma rights experts strongly emphasized that any beneficial effects of the accession process were short lived, I constructed Hypothesis I and chose the towns with the aim of evaluating that claim. Conclusions drawn from this test will therefore speak to levels of discrimination in the context of EU accession and membership, but will not speak to discrimination in the preaccession stage.

At the end of June 2011, the EU closed negotiations for membership with Croatia (European Commission 2011). Chapter 23, the negotiation chapter most relevant to Roma rights, was among the last three to be closed (European Commission 2010); in early summer 2011, when Croatia was still reminded of the challenges faced by the Roma minority in the context of accession negotiations, this project was already in place. I was therefore able to capture people’s attitudes towards the Roma during a particularly critical period: when Croatia’s treatment of Roma rights was among the last few things keeping Croatia from the EU.Footnote ¹³ For reasons listed below, the cleanest and most compelling cross-national comparison to a town in Croatia is a town in its northern neighbor, Slovenia. Slovenia acceded to the EU in 2004 and, since accession, has experienced little pressure, if any, to improve its treatment of the Roma (Vachudova Reference Vachudova2005).

Hypothesis II: Inclusive NGO Action

Testing Hypothesis II, on the other hand, utilizes a within-country design. Novo mesto and Murska Sobota in Slovenia see different types of Romani NGO action.

Roma NGO activity in Murska Sobota aims to improve Roma/non-Roma relations and is inclusive: it effectively engages the non-Roma just as much as it engages the Roma. For example, fairly early in its tenure Romani Union–Zveza Romov in Murska Sobota began organizing events that would improve local awareness and relations between Roma and non-Roma. Initially, it was challenging to convince local non-Roma that the events are intended for all, but the organization succeeded in 1992 with Ciganska noč (“Gypsy night”), an annual concert of Romani music and dances that traditionally evolves into a lively party (Horvat-Muc Reference Horvat-Muc2010; Sandreli Reference Sandreli2012). Once that barrier was breached, non-Roma began attending book launches, plays, workshops and the Romani summer camps as well. A sister organization runs a Romani radio, Radio Romic, and there, too, non-Roma listeners are invited to tune in. Connecting Roma and non-Roma through culture and awareness is the organizations’ chief objective, which they have apparently attained.Footnote ¹⁴ Not only do Roma and non-Roma attend the events together, but the number of non-Roma among event participants and radio listeners recently surpassed the number of Roma (Sandreli Reference Sandreli2012). As it connects Roma and non-Roma in a friendly, nonthreatening context, Romani Union likely lowers Roma/non-Roma anxiety—doing precisely what recent literature on contact, intergroup anxiety, and prejudice finds particularly effective (Davies et al. Reference Davies, Tropp, Aron, Pettigrew and Wright2011; Page-Gould et al. Reference Page-Gould, Mendoza-Denton and Tropp2008).

Romano Veseli in Novo mesto, in contrast, is a service provision NGO and generally does not address relations between Roma and non-Roma. The organization has a very strong presence in the Romani communities and focuses on socioeconomic aid provision and efforts related to education of both Romani children and adults (Tudija Reference Tudija2012). Aside from an occasional volunteer, non-Roma are not involved in the organization’s activities. The leading activist is well known and respected among the Roma in Novo mesto, however; every randomly sampled Romani subject who participated in the study was able to identify her by name and many profusely praised her efforts. The organization in Novo mesto is of somewhat lower capacity than that in Murska Sobota, but is growing.

Identification

As neither was randomly assigned, expansive inclusive organizing in Murska Sobota and somewhat less expansive non-inclusive organizing in Novo mesto suggest an omitted variable bias, namely the possibility that any difference in discrimination today and the scope and type of organizing in each town could be related to a difference in respect for Roma rights prior to organizing. The possibility of bias is weakened, however, by several decades of Yugoslav policies on minorities and, more importantly, of equality in employment imposed upon all citizens (Baluh Reference Baluh2012; Šiftar Reference Šiftar1989). Before Romani organizing began in Murska Sobota in early 1991 (Horvat-Muc Reference Horvat-Muc2010), the Romani experience—ranging from kindergarten (Balažek Reference Balažek2012; Horvat-Muc Reference Horvat-Muc2011a; Tancer Reference Tancer, Tišljar and Širec1997) and primary school attendance (Šiftar Reference Šiftar1989; Tancer Reference Tancer, Tišljar and Širec1997) to a severe drop in employment (Balažek Reference Balažek2012; Klopčič Reference Klopčič2012; Šiftar Reference Šiftar1989)—in the two towns was as similar as ever.

Further, the nonrandom assignment of the type of organizing in Murska Sobota and Novo mesto also suggests a possible endogeneity problem, namely that the activists from both towns might have chosen their foci because they believed that those would be effective while any others would be ineffective in their respective towns. Extensive interviews with the leading activists, however, suggest that this is not the case. Activists in Murska Sobota focus strongly on Roma/non-Roma relations because they are themselves passionate about the issue and have been so from the very start (Horvat-Muc Reference Horvat-Muc2011a; Sandreli Reference Sandreli2012). Likewise, activists in Novo mesto focus on providing socioeconomic and educational aid because they believe it to be by far the most important cause (Tudija Reference Tudija2012). The choice of focus appears entirely driven by the idiosyncracies of the NGO leaders’ personalities and preferences, rendering the assignment of inclusive organizing almost “as if” random. This substantially mitigates endogeneity concerns.

To establish the extent to which the non-Roma population of each town is familiar with local Romani organizing, I asked 100 randomly sampled non-Roma from each town to identify it. Forty-six percent of randomly surveyed individuals in Murska Sobota were familiar with local Romani NGO action.Footnote ¹⁵ In Novo mesto, only two percent were familiar with the same. This finding confirms the expectation from comparing types of organizing in Murska Sobota and Novo mesto: activism in Murska Sobota reaches almost every other individual from the non-Roma population, whereas activism in Novo mesto—understandably—reaches only a few. In testing the effectiveness of inclusive organizing at the ground level, Novo mesto is therefore a suitable counterpart to Murska Sobota.

Factors Common to the Test Locations

While the towns differ on the crucial covariates, they match very closely on the state, regional, and local characteristics that most strongly influence human rights generally and Roma rights specifically, ranging from average income to bigotry in the media. National and EU laws regarding Roma rights are self-evidently the same in Novo mesto as in Murska Sobota, as are state level factors that may have historically shaped the relationship between Roma and non-Roma. The match of Murska Sobota and Čakovec at the state level, in turn, is very close. Most importantly, the locations match on the level of Roma rights protections, both before accession and while I collected data for this project (2011/2012). Before accession, police brutality against the Roma was observed a few years prior to accession in both states (Slovenia 2000, Croatia 2010).Footnote ¹⁶ Discrimination in education, employment, difficulties in housing, and societal maltreatment were consistently present in both states—before accession (Slovenia 2000-2003, Croatia 2009-2012), right after (Slovenia 2005, Croatia 2014), and during data collection (Slovenia 2011/2012, Croatia 2011).¹⁴ Table 1 lists the factors common to the locations.

TABLE 1.

Factors Common to the Test Locations

^a Mesojedec (Reference Mesojedec2012), Republika Hrvatska - Državni Zavod za Statistiku (2001), Statistični urad Republike Slovenije (2002); ^b Horvat-Muc (Reference Horvat-Muc2011a), Klopčič (Reference Klopčič2012), Vugrinčić and Siladi (Reference Vugrinčić and Siladi2008); ^c Ajdič (Reference Ajdič2008), Šlezak (Reference Šlezak2009); ^d Bajrić (Reference Bajrić2012), Horvat-Muc (Reference Horvat-Muc2011b), Tudija (Reference Tudija2012); ^e Karba (Reference Karba2010), Vican and Litre (Reference Vican and Litre2006); ^f Republika Hrvatska - Državni Zavod za Statistiku (2001), Statistični urad Republike Slovenije (2002); ^g Kenrick (Reference Kenrick and Guy2001); ^h Kolaković (Reference Kolaković2013), Statistični urad Republike Slovenije (2011); ⁱ Eurostat (2015). EU average is 100.; ^j Barany (Reference Barany2002); ^k See entries on Slovenia and Croatia at The University of Minnesota Human Rights Center (2011); ^l See Hafner-Burton and Tsutsui Reference Hafner-Burton and Tsutsui2007; ^m Klopčič (Reference Klopčič2007), Tancer (Reference Tancer, Klopčič and Polzer2003); ⁿ Klopčič (Reference Klopčič2007); ^o Polzer-Srienz (Reference Polzer-Srienz, Klopčič and Polzer2003); ^p Pokos (Reference Pokos and Štambuk2005), Tancer (Reference Tancer1994); ^q Erjavec et al. (Reference Erjavec, Hrvatin and Kelbl2000), Kanižaj (Reference Kanižaj2004); ^r See US State Department Human Rights Reports, Slovenia (2000-2003) and Croatia (2009-2012); ^s See US State Department Human Rights Reports, Slovenia (2011-2012) and Croatia (2011); ^t European Commission (2009), RS Urad za Narodnosti (Reference RS Urad za2015); ^u Vratuša (Reference Vratuša, Klopčič and Vratuša2008); ^v Šiftar (Reference Šiftar1970).

RESULTS

The results can be summarized as follows. First, non-Roma from the EU candidate town discriminated against the Roma, while non-Roma from the EU member town just across the border did not. Second, non-Roma from the town with inclusive organizing did not discriminate against the Roma, but non-Roma from the town with noninclusive organizing did.

As this article discusses levels of discrimination against the Roma in Slovenia and Croatia, I focus on the behavior of the subjects who were senders in the trust game—in total, 303 randomly chosen individuals. The main quantity of interest—the dependent variable in the model—is the amount participants sent to their respective partners in the trust game. As the currencies in which participants were playing were not the same, the relevant variables are coded as proportions of total endowment.Footnote ²⁵

Figure 2 presents the average proportion of total trust game endowment that senders in each town sent to Roma and non-Roma partners. The figure shows a negligible average treatment effect in the case of the EU member (Murska Sobota), where senders on average sent 57 percent of their endowment to non-Roma and 58 percent to Roma partners, a statistically insignificant difference-in-means. The average treatment effect in the EU candidate (Čakovec), on the other hand, is statistically significant (p < 0.05). Senders there sent 75 percent of endowment to non-Roma and 65 percent to Roma partners.

FIGURE 2.

The Average Proportion of Total Endowment Sent to Partner in the Trust Game, by Town and Partner’s Ethnic Identity

Notes: The difference-in-means or average treatment effect is statistically significant where marked. The average treatment effect in the case of the EU member (Murska Sobota) is negligible, while the average treatment effect in the EU candidate (Čakovec) is statistically significant (p < 0.05). The average treatment effect is also statistically significant at p < 0.05 in the noninclusive NGO town (Novo mesto), but is not statistically significant in the inclusive NGO town (Murska Sobota).

Figure 2 also shows that senders from the non-inclusive NGO town (Novo mesto) on average sent 73 percent of endowment to non-Roma and 59 percent to Roma partners (the average treatment effect is statistically significant at p < 0.05). In contrast, senders from the inclusive NGO town (Murska Sobota) on average sent almost the same proportion of their endowment to Roma and non-Roma partners.

Regression analysis provides further insight. The appropriate specification for a model in which the dependent variable is a proportion is a generalized linear model with the binomial variance and the logit link function (McDowell and Cox Reference McDowell and Cox2004; Papke and Wooldridge Reference Papke and Jeffrey1996).Footnote ²⁶ To estimate the treatment effect, the model includes interaction terms between the control function and the treatment variable (Roma) (Morton and Williams Reference Morton and Williams2010). Since partner ethnicity in the trust game was randomly assigned within each town, the observations were accordingly weighted.Footnote ²⁷

Table 2 presents results from the model that includes the lottery chosen, the proportion of the endowment sent to the family in need in the dictator game, town dummies, and the main population controls—age group, gender, education level, and income bracket—as well as the interaction terms between these covariates and the treatment (Roma) as independent variables.Footnote ²⁸

TABLE 2.

Effect of Game Partner Ethnicity (Roma or non-Roma) in Croatia and Slovenia on the Amount Sent in the Trust Game: Generalized Linear Model with Weighted Data

* p < 0.1; ** p < 0.05; *** p < 0.01; standard errors in parentheses.

Table listing the coefficients and standard errors of control covariates is in the Online Appendix (Table 5).

Conditional marginal effects demonstrate the effect on the predicted mean proportion of the total endowment sent in the trust game as partner ethnicity changes from non-Roma to Roma.Footnote ²⁹ The marginal effect for Murska Sobota is not statistically significant: essentially, there is no evidence that a randomly chosen sender from Murska Sobota would send any less or any more to a Roma partner than she would to a non-Roma partner. In Čakovec, on the other hand, the change in the proportion of the endowment sent is statistically significant (p < 0.01). A randomly chosen sender from Čakovec (EU candidate) would send about 30 percent less to a Roma partner than to a non-Roma partner. The marginal effect is also statistically significant in the case of Novo mesto (non-inclusive NGO). There, too, a randomly chosen sender would on average send 30 percent less to a Roma partner (p < 0.01).Footnote ³⁰

Curiously, senders from Čakovec and Novo mesto on average sent more to their partners than did senders from Murska Sobota.Footnote ³¹ They also sent substantially more to non-Roma receivers, clearly exhibiting in-group favoritism (Hewstone et al. Reference Hewstone, Rubin and Willis2002). In-group favoritism is a manifestation of discrimination generally seen as a milder form of bias when not connected to outright out-group derogation (Brewer Reference Brewer1999). In the case of study participants from Novo mesto and to a lesser extent from Čakovec, however, out-group derogation was present as well.

Unexpectedly, participants offered unsolicited statements regarding the Roma. In most cases the statements were given after or during the exit survey; a few participants commented on the Roma during the trust game. One participant, tellingly, mistook a research team for a pair of Roma going door to door asking for money. We received comments from senders paired with Roma and non-Roma alike; whether or not they were assigned the treatment had no effect on the substance of the comments. Thirty-six percent of senders from Novo mesto made comments; one percent was positive, 11 percent were neutral, and 24 percent were negative. The most remarkable comment included a half-joking threat that “we will send a couple of bus-loads of our Roma to your hometown and you’ll see what it’s like.” Unsolicited statements were not nearly as common in the other two towns. In Murska Sobota, one subject remarked upon receiving the treatment that she did not see why it would matter that her partner was a Roma or a non-Roma. In Čakovec we received three negative comments; the most remarkable, “The hell will he send,” referred to the possibility that a Romani partner may return some of the endowment to his partner in the trust game.

A concern arises in light of these findings. Senders from Čakovec and Novo mesto might have sent less to Roma partners because the Roma in Čakovec and Novo mesto are actually less likely to reciprocate in such a context, compared to the Roma in Murska Sobota. A comparison of responses by Romani receivers, however, indicates that this is not the case. There is no statistically significant difference between what the Roma from all three locations returned to their partners, as a proportion of the total pot. The Roma played consistently.

These findings have three implications. First, the EU accession process does not necessarily lower discrimination such that it will be lower in an accessing state than in an EU member. Second, discrimination can be remarkably low, even absent, in an EU member state. Third and finally, inclusive organizing that aims to improve Roma/non-Roma relations helps reduce ground level discrimination.

A few caveats are in order. Crucially, the findings do not assess the absolute ground level effect of the EU accession process. The sample does not include observations from a location that has not yet been affected by the accession process or one that is entirely outside the purview of the EU. Without such a baseline, an evaluation of the efficacy of the process as a whole is nearly impossible to make. Therefore, while the findings demonstrate that the EU accession process does not necessarily lower ground level discrimination below that in an EU member, they do not address the claim that the EU accession process as a whole is ineffective in reducing ground level discrimination against the Roma.

Next, while the results demonstrate that discrimination can be remarkably low in an EU member state, this is not universally the case. It is abundantly clear that many Roma who live in EU member states, Eastern and Western, do not consistently enjoy enviable rights and equal treatment. At the state level, French deportations of immigrant Romani populations are self-evidently intolerant (Erlanger Reference Erlanger2010), as are Italy’s efforts to destroy makeshift dwellings in settlements outside Milan (ERRC et al. 2008). Hateful acts inspired by personal bigotry range from creating fictional administrative barriers that impede obtaining social benefits (State Department 2009a) to participating in impromptu (State Department 2009b) or planned (State Department 2007) demonstrations that usually take place in Romani settlements and frequently involve Molotov cocktails (Tkach Reference Tkach2010). To this collection of specific events, this study adds statistically significant evidence that ground-level discrimination can still exist within EU member states, and that it varies within countries, likely at the town level.

Finally, the results do not ensure that we would observe the same effect of inclusive Roma organizing throughout the EU space or beyond it, or that we would observe the same if inclusive organizing were randomly assigned. They also do not speak conclusively to the degree of a direct link between the EU and Roma organizing. Roma issues are a high priority within the EU space (European Commission 2015). In the context of violence against women, the combination of local autonomous feminist activism and regional or international normative mechanisms is remarkably effective at spurring change (Htun and Weldon Reference Htun and Weldon2012). Drawing on that, and considering the depth of EU involvement in the transnational Roma advocacy network, possible EU effects cannot be ruled out. Activists and experts from across Slovenia, however, are divided on the depth and importance of the EU in local organizing (Tahirović Reference Tahirović2011). Some claim that the EU plays a strong role in motivating activists (Balažek Reference Balažek2012; Klopčič Reference Klopčič2012), while others maintain that its role is negligible (Horvat-Muc Reference Horvat-Muc2011a; Rošer Reference Rošer2011). Whether or not the EU, in any form, is the primary driver behind the observed lack of discrimination is at present uncertain, and more work is required to clarify that relationship.Footnote ³² In the meantime, the absence of animus in Murska Sobota is striking.

This finding speaks to a facet in the literature on the promotion of human rights norms (Brysk Reference Brysk1993; Keck and Sikkink Reference Keck and Sikkink1998; Khagram et al. Reference Khagram, Riker, Sikkink, Khagram, Riker and Sikkink2002; Sikkink Reference Sikkink1993) that focuses on softening the public mood (Amenta et al. Reference Amenta, Caren, Chiarello and Su2010). The result thus contributes to scholarship that establishes the necessity of a local presence in norms promotion (Htun and Weldon Reference Htun and Weldon2012; Murdie and Davis Reference Murdie and Davis2012), and offers a clarification: not just any type of organizing will do. That does not mean that the only type of local action that helps improve human rights is inclusive organizing; rather, it means that local action ought to be tailored to its target audience. In targeting individuals from Murska Sobota, inclusive organizing appears to be appropriate.

Focusing on the behavior of individuals, this finding also bears upon literature in psychology on contact, prejudice, and discrimination. Crucially, and unlike those in many psychological studies, the subjects in this study were not treated with direct contact during the course of participation. The study instead offers an assessment of long-term intergroup contact—as it develops naturally on the ground—as a measure intended to reduce discrimination, and so provides a link between scores of convincing results from intergroup interactions in laboratory settings and theories of human rights norm promotion.