Structure versus agency: a cross-national examination of discrimination and the internalization of negative stereotypes

Stereotypes exist, as they allow for instant orientation through classification (Fiske and Taylor 2013; Ridgeway 2011; Schneider 2005). They often emerge unconsciously and can take both positive or negative forms. Negative stereotypes in particular are extremely troublesome, as they have a tendency to lead to internalization and discrimination (Derks et al. 2008; Fiske et al. 1998; Weiner and Craighead 2010). This can happen in any context containing the possibility for marginalization based on social identity (Inzlicht and Schmader 2012; Massey and Owens 2014; Steele et al. 2002). Both discrimination and the internalization of negative stereotypes are associated with detrimental mental and physical health outcomes (Inzlicht and Schmader 2012; Massey and Owens 2014; Speight 2007; Stuber et al. 2008; Vick et al.2008).

A key mechanism in the above picture is the role of differential between-group resource availability (Smart-Richman and Lattanner 2014). This is because resource inequalities can heighten group boundaries. Heightened boundaries increase the likelihood of experiencing discrimination, enabling the internalization of negative stereotypes (Lamont and Molnar 2002; Ridgeway and Correll 2006; Runciman 1966). Through this mechanism, racial and ethnic stereotypes may become self-fulfilling prophecies (Penner and Saperstein 2008). As such, income inequalities become reified into powerful cultural and symbolic social determinants (Carter 2007; Hatzenbuehler et al. 2013; Pickett and Wilkinson 2011).

Most empirical studies treat stereotypes as individual level variables, simply relating it to other individual level phenomena. As such, it is unclear how population level structural factors – such as between-group, country-level income inequality – play a part in their persistence (Hatzenbuehler et al. 2013). While it is important to understand how individuals internalize negative stereotypes, it is also consequential to discern how macro-level conditions play a part (Angermeyer et al. 2014; Gee 2008; Pascoe and Smart-Richman 2009). It is imperative to further examine how the above pieces (between-group resource inequalities, negative stereotypes, discrimination and individual level status characteristics) work in a diverse population, leading to the internalization of negative stereotypes. As research on stereotype internalization and status processes has advanced in differing academic domains, there is a paucity of information when it comes to the examination of the tangible linkages between the two (Phelan et al. 2014). Stereotype internalization has mainly been constructed as emerging from individual level interaction (Lucas and Phelan 2012; Omi and Winant 2014).¹ As such, little is known about how these phenomena are embedded in, and are perpetuated by population level factors, such as group educational attainment levels, group-level income inequalities and country-level political and economic contexts (Nagel 1996).

The Roma population serves as a great case study for the examination of the linkages between status inequalities, discrimination, and the individual difference moderators of negative stereotype internalization. The Roma of Europe number over 12 million individuals, forming a highly marginalized, yet diverse transnational minority (Isin and Saward 2013; Spirova and Budd 2008). They experience extremely high unemployment and poverty rates, low levels of educational attainment, high rates of infant mortality and lower life expectancy than the majority populations in the countries they reside in (Isin and Saward 2013; Spirova and Budd 2008; Vermeersch 2006).

This paper uses data specifically collected to study their situation across Europe, allowing for a comparative framework. With over 54,000 respondents in my data set, I am able to examine how the markedly different social and economic contexts of the countries they reside in frame the placement of stereotypes in their contextual and structural determinants. The Roma respondents in my data come from six EU and six non-EU countries. This presents the opportunity to better isolate how macro level factors may influence both the experience of discrimination and the relative internalization of negative stereotypes. In support of this, I find that there is a great amount of variation across the countries. For example, Roma men in Slovakia make 20% less income than men from Slovakia’s majority population. In Serbia, Roma men make 55% less per month than do Serbian majority men. The novelty of my model is its focus on linking population-level stratification processes with the psychological influence of stereotypes, providing an important contribution linking the micro and macro levels of these phenomena. I show that the internalization of negative stereotypes is facilitated by the interplay of individual and structural level factors.

I begin by first revealing that the relationship between stratification and the internalization of negative stereotypes is more nuanced than what has been illustrated so far in controlled experimental research settings. The subsequent section examines the direct effects of status processes on the internalization of negative stereotypes. I additionally highlight the specific sociodemographic factors which may serve to protect the marginalized from psychological harm. The final section of the article offers an analysis and interpretation of these results, drawing on recent research on status and stigma processes and my own work with the Roma. My research questions include:

1. What sociodemographic characteristics influence the likelihood of negative stereotype internalization?

2. How does the interplay between structural and individual level factors affect the above internalization?

A large body of research points to marginalized individuals being more likely to replicate patterns of oppression they are experiencing. This replication leads to the internalization of negative stereotypes and projection toward other members of the marginalized groups they belong to (Bailey et al. 2011; Banaji and Hardin 1996; Berger and Zelditch 1998; Blair and Banaji 1996; Bulhan 2004; Goffman 1963; Padilla 2001; Wilkins et al. 2014). Perhaps not surprisingly, these findings are highly contested, as the results from controlled, lab-based experimental research studies have been mixed. Select social identity theorists have pushed back, asserting that minority group members should internalize lower levels of negative stereotypes (Crocker and Major 1989; Tajfel and Turner 1986). But, as it turns out, this is highly influenced by the subgroup one belongs to. For example, African Americans tend to internalize lower levels of negative stereotypes than European Americans (Gray-Little and Hafdahl 2000). On the other hand, Native Americans, Hispanic Americans and Asian Americans tend to internalize more negative stereotypes than European Americans (Twenge and Crocker 2002). However, the Roma present a conundrum: They have resided in Europe since the middle ages, but one would expect that their stereotype internalization rates would likely be very different than of the Europeans who are not routinely discriminated against in their home countries. Thus, the large body of current research still needs grounding in population-level empirical evidence, through attention on how sociodemographic variables affect this process in different contexts.

To explain the above, existing research on the topic asserts that societies are based on a small number of primary categories, which are often deployed to indicate group belonging (Fiske and Taylor 2013; Ridgeway 2011; Schneider 2005). These categories allow for instant classification, serving to either facilitate or impede class mobility (Bourdieu 1985; Fiske and Taylor 2013). A key contributor to the allocation of individuals into in-groups and out-groups is the role of differential between-group resource availability (Lamont and Molnar 2002; Runciman 1966). Stratified social differences thus emerge from structural between-group economic cleavages, leading to the formation of group status beliefs, with deeply embedded opinions regarding the competence and the abilities of the various group members (Berger and Zelditch 1998; Ridgeway and Balkwell 1997; Ridgeway and Correll 2006; Ridgeway 1991; Ridgeway et al. 2009, 1998). In a vicious cycle, differences lead to socioeconomic inequalities, yet inequalities also lead to perceived differences and stereotypes (Kimmel 2000).

Link and Phelan’s (2014) concept of ‘stigma power’ bridges the gap between micro-level constructs of stigma and agency (Goffman 1963), and the above population-level status hierarchies. The authors position stigma as a macro-level ‘resource that allows people to obtain ends they desire’ (Link and Phelan 2014: 15). ‘Stigma power’ therefore becomes a societal phenomenon, and similarly to Bourdieu’s symbolic power, serves the interests of the dominant group (Bourdieu 1986; Link and Phelan 2014). As such, social status differences and stigma processes are closely intertwined (Kimmel 2000). The stigmatizers benefit by achieving wealth, power and status through the exercise of stigma, rewarding performances that comply with the dominant world view. Concurrently, the stigmatized may partake in ‘defensive othering’ and the internalization of negative stereotypes towards one’s one group in order to make their subordination more tolerable (Pyke 2010). Link and Phelan (2014) assert that there are ‘close parallels between processes characterizing stigmatization and the status processes that contribute to systemic stratification’ (20). They assert that like status, stigma is also rooted in shared social expectations, positing that stigma can occupy multiple levels, from interpersonal to macro-levels.

In this framework, internalization of negative stereotypes can occur as soon as one is categorized by status, framing future expectations regarding performance and behavior. This internalization exists on a continuum: marginalized group members individually encounter different levels of discrimination, leading to varying amounts of negative stereotype internalization (Crocker and Major 1989; Poupart 2003). Existing research examining how marginalized groups cope with prejudice notes that individual reactions to discrimination vary. The magnitude of the psychological distress one may experience after discrimination depends on multiple factors: the recognition of the act itself (Cooley 1902), feeling in control (Ruggiero and Taylor 1995), the length of time one is exposed to discrimination (Williams et al. 1998, 2003), whether one feels that the stigma against their own group is justified (Major and Crocker 1993), and the individual coping mechanisms deployed after the incident (Branscombe et al. 1999; Burkley and Blanton 2008). Here, it is important to note that despite being related constructs, stereotypes and discrimination are distinct. Stereotypes often emerge unconsciously and can be both positive and negative (with the negative form often referred to as stigma). Discrimination is the behavior that stems from these internal attitudes (Fiske et al. 1998; Lippmann 1922; Oakes and Haslam 1994; Weiner and Craighead 2010). Thus, between-group resource availability, stereotype internalization and discrimination are closely intertwined, yet very distinct phenomena.

As the literature above illustrates, between-group resource inequality is an important contributor to the formation of negative stereotypes. Yet, its consideration is also important for more macro-level reasons. We can think of long-standing, state-sanctioned between-group resource inequalities as a form of structural discrimination (Bourdieu et al. 1994; Wacquant 2009) or structural violence (Farmer 2010). Both concepts reflect the institutional reproduction of long-term marginalization, and the ways in which poverty and inequality ‘get under the skin’ (Singer and Erickson 2011). Collectively, the literature implies that population level structural factors (such as highly entrenched between-group income inequality) play a substantial part in how marginalized populations experience discrimination and stereotypes. Therefore, we must pay attention to country-bounded status structures. The specific topography of a state’s economic system, with its symbolically sanctioned resource inequalities, plays a central role in the formation and persistence of status and stigma processes. With this in mind, we must consider how resources acquired through both physical and symbolic capital – such as income, education, occupation, migration status, geographic location, age, gender and health – may influence the relationship between stratification and the internalization of negative stereotypes.

The data are comprised of in-person interviews with 41,334 Roma and 13,326 non-Roma individuals in twelve European countries conducted in a partnership between the United Nations Development Programme, World Bank, European Commission and the European Union’s Agency for Fundamental Rights. In 2011, these organizations coordinated efforts to examine the situation of the Roma in Albania, Bosnia and Herzegovina, Bulgaria, Czech Republic, Croatia, Hungary, Macedonia, Moldova, Montenegro, Romania, Serbia and Slovakia. These countries represent a wide range of political, economic and social contexts, in both EU and non-EU states. The sample was collected through random sampling and is nationally representative for Roma living in areas more densely populated by members of their ethnic group. The non-Roma sample serves as a benchmark for the Roma.² Of the Roma population, a randomly selected adult sub-sample (N = 4651) was asked questions about their views on stereotypes and experiences with prejudice.

The interviews were carried out face-to-face in the respondent’s homes by trained fieldworkers in the national language. Up to three household members were interviewed in every household: the head of household answered questions regarding the demographic profile of each household member and the overall status of the household; the children’s primary caregiver was queried about childcare and educational details; and a randomly selected adult respondent was interviewed concerning individual attitudes. The response rates for the Roma varied from 56% in Moldova to 90% in Croatia. The survey contains questions covering the socioeconomic situation of every member in the household. Additional topics include health, housing circumstances, neighborhood infrastructure, civic pride, citizenship status, stereotypes, human rights awareness, experience of discrimination, and migration history. For this study, I restricted the sample to only the adult Roma randomly selected to be queried about stereotypes and discrimination (N = 4651).

The dataset contains several variables that can be used as proxies for negative stereotypes toward the Roma. The interviewer prefaced them with: ‘Below is a set of statements reflecting certain opinions, stereotypes and prejudices about the Roma. We would like to know your opinion about them. Please tell us which of them you find justified and which – not.’ The statements include: 1. ‘Roma are dirty/not clean.’ 2. ‘Roma are lazy.’ 3. ‘Roma steal.’ 4. ‘Roma are abusing the system.’ Respondents had the option of responding to the negative stereotypes presented with ‘finding them totally unjustified’ or ‘finding them justified’. I consider respondents who report finding at least one of the four negative stereotypes ‘justified’ as having internalized a negative stereotype towards their own ethnic group. I base my internalization measure on established research that suggests that those who recognize other’s negative view of their own ethnic group as justified are more likely to also confirm, experience and internalize negative stereotypes (Davies et al. 2005; Massey and Owens 2014; Ruggiero and Taylor 1995; Steele and Aronson 1995).

Of the over 4600 randomly asked adult Roma respondents, nearly 27% report internalizing at least one of the negative stereotypes listed above. The first three categories (dirty, lazy, steal) are most prevalent (with approx. 30% of the respondents internalizing one or more). The category of system abuse is the least internalized. The internalization of negative stereotypes is coded as a categorical variable, comprised of those who have internalized at least one negative stereotype and those who have not. This makes sense for multiple reasons. While the individual stereotypes are all negative, they are qualitatively slightly different from each other. The various factors that may cause respondents living in different contexts to run the risk of internalizing one over the other are unknown. Thus, it would be irresponsible to lump them all together as a continuous measure representing low to high rates of internalization. It is also undeniable that internalizing at least one negative stereotype toward your own ethnic group is troubling, and warrants examination.

In In order to account for individual level factors that could influence the internalization of stereotypes, I control for variables that were shown by previous literature as influential. These are: gender (categorical: male/female), age (adults only, categorical: 18–29, 30–49, 50+), household income (continuous on log scale), employment status (categorical: employed, unemployed), educational attainment (categorical: incomplete primary, primary education only, incomplete secondary, secondary only, higher than secondary school), occupation (categorical: farmer, unskilled worker, skilled trades, professional), marital status (categorical: married, divorced, separated, widowed, cohabiting, single), self-reported health (categorical: good, fair, poor); migration status (categorical: household migrated in the last 5 years, household has not migrated) and geographic location (categorical), measured by urban or rural residence and current country. Table 1 shows the distributions of these controls.

The statistical analyses proceed in three stages. First, I present descriptive statistics of variations between respondents who have experienced ethnic discrimination and those who have not. I provide both simple descriptive (unadjusted) comparisons and group comparisons adjusted for differences in covariates through binomial logistic regression analyses. Logistic regression analysis interrogates whether the log odds of the internalization of at least one negative stereotype is associated with the recent experience of ethnic discrimination, while controlling for the effects of the covariates above. To account for non-independence among observations and the assumption of independence in sampling required by logistic regression analysis, in the third stage, I use hierarchical linear modeling to investigate the association between individual and structural economic discrimination, as this is captured by long-standing, state-enabled population level income inequality (Hox et al. 2010; Raudenbush and Bryk 2001). This method is appropriate, as my data is hierarchically structured into individuals belonging to different levels, with respondents within the same group sharing the same country-level economic environment (Goldstein et al. 2002; Hox et al. 2010; Merlo 2003).

My multilevel logistic model is grounded in a fixed component which measures the magnitude of associations between the variables, allowing for a random intercept showing the differences between second-level components and the variances in the different levels, following the guidance of Sniders and Bosker (2012) and Kreft and de Leeuw (1998). The random coefficients are measures of the random effects derived from variability between units, shown as variation between the country level intercepts in fitted regression lines (Kreft and de Leeuw 1998). At the individual level, this model examines how the log odds of internalizing stigma is modified by the predictor variables. At the country level, to interrogate how state structures enable the existence of income inequality between specific populations, I introduce a variable measure of income inequality between the Roma and the non-Roma.

The results are presented in four parts. First, I describe the overall composition of the adult Roma sample in my dataset. The following section examines how the recent experience of discrimination affects the internalization of negative stereotypes and how this relationship varies by context. Next, I identify the sociodemographic factors of influence when it comes to the likelihood of internalizing negative ethnic stereotypes. Last, I interrogate how the joint repercussions of structural and individual level discrimination influence this internalization.

Table 1 shows the descriptive statistics for the Roma adult sample. The sample population is nearly evenly split between the sexes, with 50.4% being female. Reflecting current knowledge, we find that the population is younger: 41.7% are between the ages of 18–29, approximately 37% between the ages of 30–49, and 21% over the age of 50. Over half of the adult sample is currently married. Only 23% have never been married. 60% live in an urban location. The educational attainment levels of the sample are quite low, with approximately 11% having completed secondary education or higher. Only 24.5% of the sample is employed. Of these, the largest percentage are unskilled workers (64.28%). 27.78% consider themselves in a skilled trade. Contrary to popular stereotypes about the population, only 3.5% of the total Roma households have moved (from a different city or from a different country) in the past five years. Nearly 61% of the sample report being in good health, while the rest are in poor or fair health.

Figure 1 illustrates the between-group, country level income inequality between the Roma and non-Roma. There is a great amount of variation, both between the genders and across the countries. In Moldova, Romania and in Bulgaria, Roma men and women make the same amount of money. However, they make 40–50% less than the majority men in these countries. On the other hand, Roma men in Slovakia make 20% less money than men from Slovakia’s majority population. On average, Roma women in the same country make only 48% of what non-Roma men make.

Figure 2 presents the country level percentages of ethnic discrimination and internalized negative stereotypes among the Roma. As the figure shows, country-level ethnic discrimination and the internalization of negative stereotypes closely parallel each other. In countries with high levels of ethnic discrimination, the Roma also internalize more negative stereotypes. There is a considerable amount of variation across countries. Of the overall adult Roma respondents, 27% have internalized at least one negative stereotype toward their own ethnic group, ranging from 25% to 37% of the total respondents per country.

Over 12% of the Roma report experiencing ethnic discrimination in the past year. There are vast country-level differences in reported discrimination levels. In nearly every country, Roma respondents report having recently experienced discrimination due to their ethnicity. The internalization of negative stereotypes and discrimination are closely intertwined, even across markedly heterogeneous populations and contexts. Montenegro is the only exception here. 26.33% of the Roma respondents have internalized negative stereotypes here, but only 3.39% report experiencing ethnic discrimination.

Next, I turn to the task of identifying sociodemographic factors of influence when it comes to the likelihood of internalizing negative ethnic stereotypes. Table 2 contains the results of logistic regression analyses, interrogating whether the log odds of the internalization of at least one negative stereotype is associated with the recent experience of ethnic discrimination, while accounting for the effects of sociodemographic variables. To provide a more intuitive understanding of the results, I convert these log odds into odds ratios.

Results show that when it comes to the internalization of stereotypes, the recall of ethnic discrimination is detrimental. The odds of internalizing negative stereotypes toward one’s own ethnic group are over 11 times higher when a person experiences ethnic discrimination. Additionally, the odds of internalizing at least one negative stereotype increase as the Roma progress through the educational system. This is consistent with existing work on the subject of educational institutions playing a part in larger systems of dominance, through which marginalized populations experience both explicit forms of discrimination and persistent microaggressions, increasing the likelihood of stigma internalization (Dixson and Rousseau 2005; Feagin and Sikes 1995; Kohli and Solórzano 2012; Sue et al. 2007). Migration is not a significant predictor in the model.

Taking a closer look, we find that while women in general have a higher likelihood of internalizing negative stereotypes, the statistical significance of this relationship is influenced by geographic context. Specifically, holding other variables constant in the model, of the twelve countries, this relationship is statistically significant only in Albania, Croatia, Macedonia, Moldova, Bulgaria and Serbia. Those in poor health also internalize more negative stereotypes toward their own ethnic group.

Table 3 shows results of the multilevel models, illustrating that the consideration of between-group income inequality matters when it comes to discrimination shaping the internalization of negative stereotypes. This is true for both Roma men and women. To honor the differences in income inequality between the genders, I run 2 models: one considering the income inequality between Roma men and majority men, and the other considering the same between Roma women and the men of the majority population. While we cannot directly compare logit coefficients across these models, it is interesting to note that the probabilities of internalizing stigma remain significantly positive, reaffirming that having experienced recent ethnic discrimination leads to a higher likelihood of internalizing negative stereotypes. However, structural inequality shapes this process by modifying the magnitude of this effect, with the overall influence of individual level discrimination decreasing. This reinforces previous findings that ethnic stigma and racial perceptions are mediated to a large degree by macro-structural and socioeconomic processes (Ladanyi and Szelenyi 2001; Penner and Saperstein 2008).

The multilevel results also show, that with country-level income inequality in the mix, women have a much higher probability of internalizing negative stereotypes than men. Women aged 30–49 are also significantly more likely to internalize negative stereotypes. Being widowed or never married remains a risk factor for both genders. Those in poor or fair health are more likely to internalize negative stereotypes. The odds of internalizing negative stereotypes are 1.04 times higher for those living in urban areas. Additionally, as the Roma progress through the educational system, their risk of internalizing negative stereotypes increases.

One of the most destructive effects of discrimination is the internalization of negative stereotypes. However, our understanding remains incomplete without an examination of the interplay between the macro and micro-level factors that influence the likelihood of internalizing negative stereotypes. This study made an effort to address this gap, by systematically linking the concepts of discrimination, negative stereotype internalization and status. This was accomplished through a specific focus on how individually experienced discrimination and structural income inequality shape the internalization of negative ethnic stereotypes.

The dynamics discussed require the consideration of agency and power: stereotypes require buy-in, and persistent income inequalities require deep societal control over employment and educational opportunities. Paying homage to Bourdieu’s multilevel model of the social world provided an apt framework for the interrogation of the ramifications of structural discrimination. This study enriches the oft individual-level focus of the existing psychological literature on stereotype internalization. My results support Bourdieu’s assertions when it comes to the benefits from education being unequally distributed between groups, as different groups possess varying amounts of cultural and social capital required successfully navigate and manipulate this system (Bourdieu 1986). The odds of internalizing at least one negative stereotype increase as the Roma progress through the educational system. In order for educational institutions to cease to be a mechanism of symbolic violence against the Roma, future research and policy efforts need to focus on ensuring that educational achievements enable class mobility and health for all.

Researchers and policy makers have long-grappled with how to best ensure that education can serve as a route to socioeconomic mobility for the Roma too. Currently, only one percent of the Roma are able to enter a university, and this is not because of a lack of want, nor is to because they value education less (Bhabha et al. 2018). As a recent study conducted in collaboration between the François-Xavier Bagnoud Center for Health and Human Rights at Harvard University (Harvard FXB) and the Center for Interactive Pedagogy in Belgrade notes, Roma pupils across Europe still face significant discrimination while attending school. This can occur as often as nearly every day according to the respondents queried (Bhabha et al. 2018). Yet, this work also points at important sites of intervention when it comes to achieving more positive higher education outcomes. These include access to discrimination-free early childhood development services, adequately funded schools for Roma pupils, increasing parental education levels, and having non-Roma allies to help when Roma pupils do experience discrimination (Bhabha et al. 2018). While none of the above ensures that educational institutions will fully cease to ensure that the Roma do not internalize the negative stereotypes they are subjected to, nevertheless, many countries in Europe are actively taking steps in the right direction to allow for eventual equity (Fuller et al. 2015).

Though this study is a great first step, it needs to be supplemented by further ethnographic work exploring the reasons as to why particular Roma respondents internalize prevalent negative stereotypes toward their own ethnic group. My ongoing qualitative research with Hungarian Roma refugees points to the reasons for this internalization being both socioeconomic status and context dependent. Exploring this ‘why’ question in the future will be imperative for a more complete understanding of the processes that link structural and individual level discrimination and the internalization of stigma.

It is important to remain mindful that in environments characterized by high levels of inequality, both the stigmatizers and the stigmatized are adversely affected when it comes to their overall social, psychological and physical health outcomes (Castro and Farmer 2005; Hatzenbuehler 2011; Parker and Aggleton 2003; Phelan et al. 2008; Ridgeway and Balkwell 1997; Ridgeway and Correll 2006; Wilkinson and Pickett 2009). It is vital that future work pays equal attention to the processes that also influence the majority population’s internalization of negative stereotypes.

While the data was collected through a random sampling process following current best practice standards for dealing with the sensitive subject of ethnicity, it cannot fully capture the experiences of all the Roma (Ivanov et al. 2012). Though only 2–3% of the Roma population complete university,³ future researchers also need to study these outliers, while also continuing to focus on the many ongoing successful educational initiatives targeting the Roma.⁴

Lastly, we need to ask what would be a ‘low’ level of stereotype internalization and discrimination experience in a population, and what could be considered high. In my sample, over 12% retroactively report experiencing ethnic discrimination, and nearly 27% have internalized negative stereotypes. This is troubling, as I interpret this as nearly a third of my sample risking the curtailment of their life chances due to stereotype theat. In an ideal world, no one would experience any discrimination. However, in order to dig into these results a bit more, future data collection efforts need to ask more nuanced questions when it comes to the discrimination measure. In order to ensure that the recall of discrimination is not under-reported, the surveyors themselves need to come from the Roma community. Though this dataset represents the highest quality multi-country data currently available on this population, future data collection efforts need to directly enlist Roma leaders and academics, focused on collecting longitudinal data on the population.

Despite the above limitations, this study presents multiple contributions to the literature. I show that when it comes to a better understanding of the lived experience of disadvantage, structural economic marginalization – as measured by between-group inequality and framed by geographic context – matters greatly. There is a considerable amount of heterogeneity between the 12 countries under examination when it comes to the internalization of stigma. Yet, I caution against grouping these countries into stable categories of more or less stigma toward the Roma, as there is considerable ebb and flow in this phenomenon across time. Romania serves as a good example for this. Before 1864, the Roma were enslaved in Romania, then murdered by the Nazis. During the Ceausescu dictatorship, they were included in both the employment and educational systems, improving their overall economic standing.⁵ However, after the fall of the dictatorial regime, over 90% of Romania’s Roma again found themselves in extreme poverty.⁶

The examination of the multiple social, political and economic factors that influence changes in population level internalization of negative stereotypes across time and geographic context serve as fertile ground for future research on the subject. A worthwhile next-step would be the consideration of how migration might play a part in discrimination and stereotype threat. As only 3.5% of my respondents have migrated any time in the past five years, due to the temporal mismatch between how the migration related question and the discrimination related question was asked, it was hard to determine if their experience of discrimination occurred in their current country of residence. Future data collection efforts focusing on this population need to ask a more comprehensive set of migration-related questions. The low migration rate of my sample points to two possibilities: It potentially pushes back against a very prevalent public discourse about the Roma, which often portrays them as being perpetually on the move. Alternatively, it is also possible the overall sampling strategy of the survey is biased against migrant populations. Both possibilities require further research with more complete longitudinal data sources on the subject of migration and ethnic discrimination.

Future research needs to re-visit the issue of gender when it comes to how structural variables affect the experience of discrimination and the internalization of negative stereotypes. People at the intersection of multiple disadvantages are particularly vulnerable: women in general have a higher likelihood of internalizing negative stereotypes, but women who are divorced, widowed or never married have a much higher probability of internalizing negative stereotypes than those who are married. However, when holding other variables constant in the model, of the twelve countries, this relationship is statistically significant in six of the twelve countries. This calls for future comparative qualitative research for a better understanding of the differences between women’s lived experiences in the various European Union member countries.

My findings support recent theoretical work on stigma and processes of social stratification being closely intertwined (Link and Phelan 2014). Multilevel analyses results show that the interplay between structural and individual level factors is important when it comes to understanding stereotype internalization. I find that between-group income inequality directly and significantly influences the likelihood of experiencing discrimination and the internalization of negative stereotypes.

In short: (1) We know that minority group membership serves as a marker of lower status, while lower status also leads to a higher likelihood of being categorized as belonging to a marginalized minority group (Ladanyi and Szelenyi 2001; Penner and Saperstein 2008). (2) Structural economic marginalization additionally leads to stigma, and the higher likelihood of the stigmatized directly being discriminated against – linking the tangible memory of ethnic discrimination to the macro level phenomena giving rise to it (Marmot 2005; Wilkinson and Pickett 2009). (3) Negative stereotype internalization emerges from the direct knowledge of the stereotype in question. It is shaped by experiences with individual level discrimination, which in turn is connected to status processes. Therefore, it is reasonable to expect structural between-group income inequality to directly influence the social psychological process of stereotype internalization (Castro and Farmer 2005; Hatzenbuehler et al. 2014; Parker and Aggleton 2003; Phelan et al. 2008).

This study rests on the conviction that when conducting research on the internalization of negative stereotypes, data permitting, it is important to maintain focus on both individual and population level processes, as this is the only way to identify holistic intervention strategies. This perspective serves as a contribution to the existing literature. My results show that in highly unequal environments, factors that we often think of as protective – such as higher education – may carry unintended consequences when it comes to the internalization of negative stereotypes (Mirowsky and Ross 2003). In such contexts, policies that only focus on individual level factors without also addressing structural inequality-based processes will fall short in providing equally protective psychological health benefits to the marginalized populations they seek to target.

My results provide evidence showing that structural discrimination has an effect on many of the sociodemographic factors influencing the likelihood of internalizing negative stereotypes. This analysis serves as an important first step in tracing the contours of the simultaneous effects of individual and structural discrimination on the internalization of negative stereotypes, drawing attention to the necessity for an explicit inclusion of theory to guide study design and interpretation. My results underscore the need for further population level research on psychosocial processes as they are connected to processes of stratification. Research on this subject has significant health and social policy implications.

No potential conflict of interest was reported by the author.

Boróka B. Bó is a sociologist, mathematical demographer and data scientist. For the last 10 years, she has worked on the statistical modeling of population-level processes as they are internalized by individuals. Her research interests broadly encompass areas of stratification, migration, gender, health and demography. She specializes in incorporating a mixed method approach to understand complex social phenomena, combining multiple qualitative methods with ‘big data’ and digital demography. Currently, she is a PhD candidate in the joint PhD Programs in Sociology and Demography at the University of California, Berkeley.

Boróka B. Bóhttp://orcid.org/0000-0002-2675-2169