Minority language recognition in Europe often follows a territoriality principle—i.e., the belief that minority communities in a given locality, if sufficiently large, should be linguistically accommodated (Csata and Marácz 2016; De Schutter 2008). This accommodation includes the use of the minority language in all official matters, including at city offices, in judicial proceedings, and in all government communications. In theory, this recognition should have socioeconomic implications. On the one hand, recognition could encourage further segregation of the minority group from the majority. This segregation in turn stunts socioeconomic development. On the other hand, affording minorities the right to use their language sends a message of their cultural worth and facilitates interactions with the state. Thus, accommodation can be economically efficient and seen as socially egalitarian—both of which facilitate economic well-being. In this paper, we ask: What are the socioeconomic implications of minority language recognition when realized through the territoriality principle?

We answer this question by focusing on Romania—home to one of the largest ethnic minorities in Europe (Hungarian). We leverage a quasi-experiment: In Romania, the territoriality principle is enshrined in the legal framework: If a minority group constitutes at least 20% of an administrative area, the minority language is recognized as a language of public administration. Using data at the municipality level, we find that—consistent with the literature—ethnic diversity has a negative effect on socioeconomic outcomes. This effect, however, manifests only in areas where Romanian is singularly recognized. In places where minority languages are also recognized, the negative economic effects of diversity are attenuated.

The paper proceeds as follows. In the next section, we review the literature on the purported link between ethnic diversity and socioeconomic outcomes. We focus on the mechanisms that underlie this relationship, and on how ethnic demography and institutional choices moderate this effect. We then discuss the moderating effects of the territoriality principle—i.e., how recognition of a minority’s language can attenuate the “growth tragedy” (Easterly and Levine 1997: 1203). In section three, we present the research design, detailing the municipality-level data used for our analyses. We then present the empirical results in section four, with robustness tests in section five. We conclude by discussing the theoretical contributions and policy implications of these findings.

The Negative Effects of Ethnic Diversity

Scholarship has identified different pathways—from civil war to fertility rates—linking ethnic diversity to negative economic outcomes (see Gören 2014 for an overview). In this same vein, we suggest ethnic diversity limits economic growth due to the patterns of cooperation and conflict that characterize ethnically heterogeneous societies (e.g., Alesina et al. 1999; Easterly and Levine 1997; Habyarimana et al. 2009; Miguel and Gugerty 2005). In this section, we outline specific mechanisms—particularly those relevant to our research context (Romanian localities)—before discussing (1) the importance of choosing an appropriate measure for ethnic diversity, and (2) how contextual factors—e.g., ethnic demography and institutional arrangements—could moderate the negative effects of diversity.

While early work focused on how “primordial” ethnic differences can limit development by stoking conflict between groups (e.g., Fishman 1966: 152), later studies specified concrete mechanisms related to economic inefficiencies that characterize interactions in multiethnic societies. For example, Easterly and Levine (1997) suggest that salient group demarcations lead to low levels of intergroup trust and inefficiencies in information exchange. These inefficiencies are likely to be worse in multilingual societies, where language barriers can further limit the ability of groups to communicate (Liu 2015). At a minimum, group differences—especially those rooted in language—increase transaction costs, and thus create economic inefficiency (Arcand 1996).

Ethnic diversity can also limit economic development through its effect on political competition. In multiethnic societies, political competition can incentivize group members to seek office for the purposes of capturing and distributing rents to their own ethnic group (Chandra 2007). Rent seeking, in turn, lowers investment in other potentially productive aspects of the economy (Easterly and Levine 1997). Finally, while ethnic differences do not necessarily beget violent conflict, governments may prioritize consumption spending to reduce intergroup tensions instead of investing for economic growth (Montalvo and Reynal-Querol 2005).

Many of the outcomes we consider in this study—education, health, and infrastructure—are directly related to the effective provision of public goods. Thus, we proceed with an in-depth discussion of why ethnically diverse communities underperform in the provision of these goods relative to more homogeneous communities. According to Habyarimana et al. (2009), three families of mechanisms connect ethnic diversity to the under-provision of public goods: (1) preference divergence; (2) the ease/difficulty of collective action; and (3) differential behavior among/toward coethnics.

First, different ethnic groups may have varying preferences over both the types of public goods and their beneficiaries. Since coethnics are more likely to value processes and outcomes in a similar way, they are more likely to develop a consensus and work together—which leads to positive socioeconomic outcomes. Conversely, individuals from different ethnic groups are more likely to have divergent preferences concerning public goods provision—e.g., location or intended beneficiary. Early work generally employed a strict version of this assumption (e.g., Rabushka and Shepsle 1972). However, the type of ethnic cleavage (e.g., language versus religion, see Bormann et al. 2017); the degree of differences between groups (Fearon 2003); the extent of cross-cuttingness (Liu and Ricks 2022; Selway 2011); and the institutional context (Posner 2004b) can all moderate the extent of preference divergence—and thus the extent of public goods provision and economic development.

The second class of mechanisms focuses on the ease of collective action within ethnic groups. Not only can ethnic groups draw upon shared cultural material to produce public goods, coethnics also have the benefit of a shared linguistic repertoire, thereby aiding communication. These shared understandings also minimize the likelihood of misunderstandings arising from cultural references (Liu et al. 2014: 140). In these ways, bargaining and decision-making in homogeneous communities are more efficient than in their diverse counterparts. Several scholars have also noted that social sanctioning is easier within ethnic groups due to findability—or the ability of coethnics to locate one another (Miguel and Gugerty 2005). Findability reduces an individual's ability to shirk. Both the capacity to punish and the resulting increases in trust facilitate public goods provision. Again, however, these accounts underestimate the capacity of institutions to facilitate collective action among different ethnic groups.

The final class of mechanisms concerns cooperation between coethnics on the one hand, and between individuals from different ethnic groups on the other. Ethnically homogenous communities may develop norms and cooperative behaviors that only apply within the ethnic group. This family of mechanisms concerns what individuals will do as opposed to what they can do. For example, as Habyarimana et al. (2009) note, in a situation where everyone can sanction shirking—but only coethnics do so—cooperative outcomes between coethnics are more likely. These patterns of behavior facilitate the provision of public goods by spurring cooperation between coethnics. At the same time, they limit cooperation between individuals from different ethnic groups. Taken together, these findings suggest that ethnically diverse communities will underperform relative to ethnically homogeneous ones when it comes to public goods provision. Next, we turn to a discussion of how to best measure ethnic diversity.

Measuring Ethnic Diversity

How scholars measure and operationalize diversity has important consequences. First, scholars must decide what type of diversity is relevant to their theoretical account: ethnic, linguistic, cultural, or religious (see Bormann et al. 2017). This decision often depends on the specific context of the study—i.e., which ethnic cleavages are politicized or relevant to the outcomes being studied (Posner 2004b). Next, scholars must decide how to aggregate diversity into a specific measure. If the focus is on how majoritarian politics matter, an indicator of whether one ethnic group composes more than 50 percent may be sufficient. However, if an overall depiction of diversity is the aim, then scholars have multiple indices from which to choose—e.g., fractionalization, fragmentation, or polarization. Each of these indices operationalizes diversity in a slightly different manner. For example, fractionalization does not consider the relative size of groups, while polarization does (Montalvo and Reynal-Querol 2005). At the same time, these measures also connote different meanings. In the context of our study, we focus on polarization; we discuss this choice in greater detail below.

Ethnic Structures and Institutional Arrangements

There are two factors that could moderate the relationship between diversity and economic growth: (1) ethnic group characteristics, and (2) institutional structures. First, we know ethnic demography matters: Factors such as the relative size of groups (Posner 2004a); the economic inequality between groups (Baldwin and Huber 2010); or the degree of difference between them (Fearon 2003) can moderate—if not mediate—the strength of the relationship between diversity and economic growth. Likewise, the presence of cross-cutting cleavages or multiple identities can mollify the negative impact of diversity on economic growth (Liu and Ricks 2022; Selway 2011). In short, ethnic groups are not like units. Groups A and B may differ from one another in ways completely unrelated to the differences between groups C and D. Likewise, group A and group B could cooperate in one context but not another. Accounting for some of these differences shows that the relationship between diversity and economic growth is not so straightforward.

Whether diversity has a negative impact on economic growth also depends on whether institutions promote social inclusion, and if so, the efficiency of their work. Institutions can be as broad as regime type: Collier (2001) shows that the negative effects of diversity are more pronounced in autocracies than democracies. Alternatively, institutions can be thought of more narrowly: Putnam (2007) suggests that an inability to modernize social solidarity institutions is the reason for inequality between different ethnic groups. Similar worrying developments are reported in European studies that draw attention to the negative economic impacts of increasing ethnic polarization and residential segregation (Marcuse and Kempen 2000; Musterd and Ostendorf 1998). In sum, policy choices matter.

Language can also moderate the relationship between diversity and development. If ethnic groups share a common tongue (Liu 2015; Liu and Pizzi 2018), or if a government officially recognizes a minority language (Medeiros et al. 2019), the negative effects of diversity may be attenuated. Language can be both a characteristic of ethnic demography and a facet of institutional design (see Ringe 2022). Multilingualism, for example, can arise when an individual holds multiple ethnic identities (e.g., when a child’s parents are from two different ethnic groups); it can also be the result of an institutional arrangement (e.g., the adoption of an official lingua franca). We proceed by discussing how a policy choice related to language—official language recognition through the territoriality principle—can moderate the relationship between ethnic diversity and socioeconomic development.

The Moderating Effects of Language Recognition

Generally, work on ethnic diversity and economic development assumes one of two types of monolingualism. The first is individual monolingualism—i.e., people only know one language. And thus, when two people speaking two different languages interact, their inability to communicate can heighten mistrust and hinder any meaningful socioeconomic exchange. But people can be multilingual; and people do learn second languages (Laitin and Ramachandran 2016; Hu and Liu 2020). When people know a second or third language, they are able to interact meaningfully with people from different ethnic groups. These interactions can allow for the development, understanding, and communication of congruent preferences—e.g., the development of a paved road here versus there. Moreover, if people can switch between languages in their communication—even if they are not fluent in their non-native tongues—this facilitates efficiency. People understand what the implied unit of transaction is; people are comfortable clarifying misunderstandings; and people trust that they will be able to interact with the same individuals again in a future iteration (see Ringe 2022). This efficiency is important for socioeconomic development.

The second assumption common in the literature is of territorial monolingualism—i.e., only one language is recognized as official. Even countries that are de jure bilingual (or multilingual) at the national level, often manifest as monolingual at the subnational unit—e.g., Belgium, Canada, and India. Monolingualism can be a problem when there is a sizable minority population; if the community cannot engage meaningfully with the state, they are socioeconomically disenfranchised (Ginsburgh and Weber 2011; Liu 2015). But governments do recognize second languages as working languages within subnational units. This allows the minority community to substantively engage with the state—e.g., to fill out requisite paperwork and interact with civil bureaucrats in their own language.

Language recognition, however, is not just about the practical matters of filling out forms and communication. There is also an element of symbolic recognition—i.e., a signal for “parity of esteem” (Van Parijs 2011). When the government recognizes a minority language, it suggests the minority is of worth and importance. This in turn can legitimize the government in the eyes of the minority group (see Liu and Baird 2012; Marquardt 2018; Ricks 2018)—and even possibly the majority group (Liu et al. 2015). In turn, confidence in political institutions is important for economic development (see Mishler and Rose 1997).

In this paper, we suggest that the ability to interact across ethnic boundaries, the ability to interact with the state, and the value of symbolic recognition associated with language recognition are consequential for socioeconomic outcomes. Specifically, we contend that while the economic effects of ethnic diversity—all else being equal—may be negative, government policies can attenuate—if not ameliorate—the consequences. For example, when governments allow for minority languages to be used in public administration in areas with a sizable minority population, this not only acknowledges that there is ethnic diversity, but it also signals the value of the minority group. Recognizing equality across different ethnic groups can promote socioeconomic development. Even if we remain pessimistic about the positive effects of language recognition on minority trust and state attachment, existing work shows that a failure to meet linguistic demands can spur frustration and anger (Chriost 2004; Liu et al. 2015; Moormann-Kimáková 2015). In fact, studies focusing on our research context, Romania, emphasize the contentious nature of language recognition, noting negative consequences for intergroup relations and the extent to which the Hungarian minorities feel respected by the state (Csergő 2007; Kontra 1999; Stroschein 2012). In short, evidence shows that language recognition certainly matters to minority communities. Given this discussion, we hypothesize the following:

  • Hypothesis 1: Ethnic polarization has a negative effect on socioeconomic development if a minority language is not recognized in a region.

  • Hypothesis 2: Ethnic polarization has no effect on socioeconomic development if a minority language is recognized in a region.

Research Design

We look at the conditional effects of minority language recognition in Romania. Romania is home to multiple minority ethnic groups—who collectively constitute more than 10% of the country’s population. The census recognizes 19 minority groups, including a large Hungarian population—a legacy of World War I (Csata et al. 2021b). In addition to the Hungarians, there is a significant Roma population—subject to systemic discrimination, as elsewhere in Europe (Csata et al. 2021a). There is also a German minority population. While their numbers have steadily declined, the group is recognized and remains politically relevant. For example, the current Romanian president Klaus Iohannis is an ethnic German from Transylvania (Sibiu).

Per the Romania Constitution (Article 120)—and further clarified by the 2001 Law on Local Public Administration—if a minority community constitutes more than 20% of the population in an administration-territorial unit, they have the right to use their mother tongue in all public matters (Horváth 2009; Salat and Novák 2015; Toró 2020). This threshold is quite high—and does not regard the absolute number of the minority population (Csata and Marácz 2016; Wickström 2019). Take Cluj, for example. The city—the second largest in Romania—has a total population of 325,000, of which 50,000 is Hungarian (2011 census). Yet, because of the 20% threshold, Hungarian is not a recognized administrative language in Cluj (see Liu et al. 2018). In this paper, we leverage this threshold to assess whether minority language recognition can attenuate the socioeconomic effects of ethnic diversity.

Within Romania, we look at the municipality level—inclusive of communes (villages) and towns (N=3181). Per European Union’s (EU) Nomenclature of Territorial Units for Statistics (NUTS), municipalities are the equivalent of the Local Administrative Unit (LAU) 2. We focus on municipalities as the Romanian legal framework specifically designates them as the administrative unit at which the territoriality principle should be applied. In other words, if a municipality has an ethnic minority that amounts to over 20% of the population, said municipality must ensure language recognition.Footnote 1

Dependent Variable: Socioeconomic Development

Conventional measures for development include GDP per capita (Arcand and Grin 2013) or economic growth (Liu 2015). Unfortunately, we are not able to use either one here because the data is not available at the municipality level. As such, we use an alternative measure: the local human development index (LHDI). The first version was created in 2007 by Dumitru Sandu who aggregated data from ten variables along four dimensions: housing infrastructure, community financial resources, individual (family) economic capital, and age-adjusted community human resources (Sandu 2011). Since then, the measure has been improved multiple times. The one we use in this paper is from 2011 and looks at six indices on four dimensions: human capital—the aggregate level of education in the municipality; health capital—life expectancy at birth; vital capital—the average age of the adult population; and material capital—average living space per dwelling, gas consumption per capita, and the number of cars per 1000 inhabitants (Ionescu-Heroiu et al. 2013). The final index is the factor scores from a principal component analysis—where higher values suggest more development. For our sample, the local human development index ranges from 21 to 140.

Official recognition of a minority language at the municipality level should have a positive effect on each of the components of LHDI. When it comes to human capital, for example, a minority who lives in a municipality with minority language primary schools should also, all else being equal, have better educational outcomes than a similar individual who does not have access to education in their own language (see Csata 2014; Laitin and Ramachandran 2016). The logic is similar for health capital. When an individual can communicate with medical personnel (as well as the healthcare bureaucracy) in their own language, they are more likely to receive the requisite care (see Laitin and Ramachandran 2016). In terms of material capacity, the recognition of a minority language facilitates economic exchange by reducing transaction costs. At the same time, recognition boosts the self-esteem of the minority group, increasing trust in the government, and thus facilitating economic growth generally. As a result of economic growth, material resources like the size of apartments or the number of cars in a municipality should increase (see Liu 2015; Liu et al. 2015). Finally, improvements in these three components should, in the long term, increase vital capital. When individuals are more educated, can efficiently access healthcare, and possess more resources, they are more likely to live longer.

Figure 1 shows the local human development index across Romanian municipalities and suggests that developed settlements appear to concentrate around larger towns.Footnote 2 This is consistent with Ionescu-Heroiu, Burduja, and Sandu (2013: 111) who find that large towns act as catalysts for development in less-developed regions.

Fig. 1
figure 1

Local Human Development Index in Romania Municipalities (2011)

Full size image

Independent Variable 1: Ethnic Diversity

While the Romanian census recognizes twenty different ethnic groups (inclusive of the majority Romanians), demographic data show that many of these minorities live in areas where they are the ethnic majority—thereby increasing the territorial separation of ethnic groups. Moreover, daily interactions are often structured through ethnically homogeneous networks (Csata 2018). Given this discussion, we believe ethnic polarization—versus other conventional measures, e.g., fractionalizationFootnote 3 or fragmentationFootnote 4—best captures the ongoing process of increasing ethnic parallelism and institutional pillarization in the Romanian context. Specifically, we use the ethnic polarization index (EPI) developed by Montalvo and Reynal-Querol (2005), which is calculated according to the following formula:

$$Q=1-{\sum}_{i=1}^N{\left(\frac{0.5-{\pi}_i}{0.5}\right)}^2{\pi}_i,$$

where Πi is the ratio of ethnic group i compared to the total population. We use data from the 2011 census. The value of EPI ranges from 0 to 1, where 1 is a society divided into two equal groups—i.e., a state of perfect polarization. Deviations from this value can suggest either one ethnic group is larger than the other or that there are more than two ethnic groups. As we see in Fig. 2, the counties in Transylvania—home to the sizeable Hungarian minority—generally have higher levels of polarization than the rest of Romania. Within Transylvania, the counties with the highest levels of polarization are Satu Mare, Mureş, and Bihor.Footnote 5 The sample average is 0.22 with a standard deviation of 0.28.Footnote 6

Fig. 2
figure 2

Ethnic Polarization in Romanian Counties (2011)

Full size image

Figure 3 disaggregates the ethnic polarization measures to the municipality level. Darker shades indicate settlements with higher polarization. Here, Zagon (0.999) in Covasna; Lopadea Nouă in Alba (0.993); and Tărcaia (0.987) in Bihor are of note. In each of these municipalities, there is another minority group—usually but not always the Hungarians—who are of the same size as the Romanians. Note that the standard deviation at the municipality level is higher than at the level of the county (0.32).

Fig. 3
figure 3

Ethnic Polarization in Romanian Municipalities (2011)

Full size image

Independent Variable 2: Language Recognition

Recall, we are interested in the conditional effects of minority language recognition in public administration. To this end, we assign the variable Language Recognition a value of “1” if a minority population constitutes more than 20% of the municipality population—per the 2011 census; and a value of “0” if otherwise. Almost 12% (N=363) of the municipalities in Romania recognize a minority language. Here, recognition includes municipalities such Gălăuțaș where the Hungarians are 20% and the Romanians are 76%; Bahnea where the Hungarians are 30%, the Romanians are 33%, and the Roma are 35%; and Cârța where the Hungarians are 99%. While the vast majority of these cases involve the recognition of the Hungarian language (N=326), other recognized languages include Bulgarian (Dudeștii Vechi; Старо Бешеново), Czech (Gârnic; Gerník), Serbian (Pojejena; Пожежена), Ukrainian (Bistra; Бистрий), and others—but not Romani (more below). The correlation between ethnic polarization and language recognition is positive and significant (0.51; p value=0.00). As we see in Table 1, we have the full mathematical range for the ethnic polarization index in areas that do and do not recognize a minority language.

Table 1 Ethnic Polarization According to Minority Language Recognition Status
Full size table

The 20% threshold is a legal obligation: Municipalities where a minority group exceeds 20% of the population are legally required to recognize the minority language. And while de jure recognition is not sufficient for de facto equality of languages, we argue the former is necessary for the latter. Thus, while there may be some error in our chosen operationalization of minority language recognition, it is the best available option. Moreover, while concerns about census data quality are relevant for measures of ethnic identification (particularly among Roma), these concerns are less acute for LHDI. Measures such as age and size of dwelling are both less subjective and less sensitive than questions related to ethnicity. Importantly, census participation is mandatory in Romania. Thus, we expect the LHDI measure to be relatively bias free.

Control Variables

There may be other conflating factors. Geography, for example, can matter for development (see Ionescu-Heroiu et al. 2013). For example, a municipality with a small population is less likely to affect the supply-and-demands of the market than one with a large population. Likewise, a large distance from the nearest market—or a lack of roads—can hinder development. Here, we control for the size of the municipality (per 2011 census). We also look at the distance of the municipality center from the nearest town (kilometers) and the existence of a European road (a road that crosses an international border, possibly facilitating economic development) crossing through the municipality (data source: Sandu 1999). The quality of democratic institutions can also affect development (Bracic 2020). A vibrant democratic environment, one where participation in public life is high, can facilitate both cross-cultural understanding and adjudicate conflicts—both necessary ingredients for economic development. To measure civic participation, we use two proxies: the number of non-government organizations per 1000 inhabitants (as reported by the Ministry of Justice, National Register of NGOs)Footnote 7 and the participation rate in the 2016 local elections.Footnote 8 Finally, we include a dummy for whether the municipality is in Transylvania. The large region is home to a diverse community, including a sizable Hungarian minority—one of the largest ethno-linguistic minorities in Europe. Table 2 lists the descriptive statistics for the variables.

Table 2 Descriptive Statistics—Romanian Municipalities
Full size table

Empirical Evidence

Given the continuous nature of the local human development index, we estimate the model using ordinary least squares.Footnote 9 We begin with a simple bivariate regression. The coefficient for ethnic polarization is statistically significant and negative (β=-5.06; SE=0.90). While the marginal effect may be small (i.e., 1% of the variation in local human development is attributable to ethnic polarization), this pattern is consistent with the literature. The association attenuates once we include controls (see Table 3). In model 1, the coefficient for ethnic polarization decreases to -2.90— but retains statistical significance at the 0.01 level. Again, the negative coefficient is consistent with the literature on ethnic diversity.

Table 3 Explaining Local Human Development in Romanian Municipalities
Full size table

In model 2, we consider whether a minority language is officially recognized in the municipality. Here, we see the substantive effects for ethnic polarization remain unchanged. But what is of interest is the negative coefficient for Language Recognition (β=-0.94; SE=0.70). At first glance, this suggests areas that recognize a second language have lower levels of human development. In fact, it looks like the effects of minority language recognition are statistically no different from that of complete ethnic polarization. Given these numbers, the policy recommendation would be to shy away from minority language recognition.

But we caution against this. Since the effects of polarization on human development can manifest differently depending on whether a minority language is considered official in a given area, we must consider the conditional effects of recognition. In model 3.1, we rerun the model but on a restricted sample—specifically, only areas that do not recognize a minority language. And then in model 3.2, we delimit the sample to municipalities that do recognize a minority language. This side-by-side comparison offers a first glance at how recognition matters—for the positive. In areas where a minority language is not used in public administration, human development levels are significantly lower when ethnic polarization is high (β=-3.79; SE=0.82) However, in places where a minority language is considered official, we see ethnic diversity has no significant bearing whatsoever on the well-being of the community (β=-0.83; SE=1.61).

In model 4, we run the full model with ethnic polarization and language recognition interacted. While the coefficients for the two constituent terms are still negative, the coefficient for the interaction is positive and significant (β=4.58; SE=1.85). To help with ease of interpretation, we plot the marginal effect of EPI when a minority language is and is not recognized (see Fig. 4). The results mirror those we saw in the split sample: recognition of a minority language can attenuate the otherwise negative effects of ethnic polarization on human development. Specifically, ethnic polarization only has a statistically significant, negative effect (β=-3.58; SE=0.81) on LHDI when a minority language is not recognized. In areas where minority languages are recognized, the effect of EPI is nonsignificant (β=1.00; SE=1.64).

Fig. 4
figure 4

Marginal Effects of Ethnic Polarization on Development (95% CI)

Full size image

It is possible these results are driven by some latent unobserved variable. To consider this, we run the full interaction model with county fixed effects in model 5 (to consider possible UDMR party effects in Hungarian-majority counties, e.g., Covasna and Harghita) and then with region (e.g., northwest, west, and center) fixed effects in model 6. The results remain highly robust. Ethnic polarization has a negative effect on socioeconomic development—but only when a minority language is not recognized. In fact, when we account for time-invariant, regional factors (model 6), the negative association between ethnic polarization and LHDI in municipalities without language recognition becomes even stronger (β=-4.52; SE=0.82). And, in places where minorities are afforded language recognition, the association between EPI and LHDI is positive and significant at the 0.10 level (β=2.99; SE=1.65).

As expected, geographic factors matter. A large population size has a positive effect on development. In fact, of all the variables included in the model, the marginal effects of population size (0.31) are the second largest—three times the magnitude of either ethnic polarization (-0.09) or minority language recognition (-0.10). Having a European road that runs through the municipality can improve development although the marginal effects are relatively small (0.07). Likewise, proximity to the closest town matters for development—more so for a small town (-0.22) than a town with more than 30000 inhabitants (-0.01). Widespread civic participation is also important for development. The socioeconomic effects of having many NGOs are significant and positive (0.38). Likewise, higher levels of turnout have a positive effect (0.05)—suggesting possible widespread clientelism (Barany 1998).

Robustness

In this section, we subject our findings from the previous section to several robustness checks. We first examine alternative measures of socioeconomic development. Instead of local human development, we shift our attention to indicators of private economic production. One is about business activities—where the logic is that the number of enterprises will increase as an economy grows. We measure this using data from a balance sheet of firms and cooperative associations from 2018 (per Ministry of Finance). We focus only on Romania-based companies. We exclude firms engaged in financial activities, sole traders, and family partnerships because they are subject to a separate statistical accounting record. We standardize the number of active businesses per 1,000 inhabitants. The average business density in Romania is 12.98—with a wide range from 0 to 155. We also look at housing stock. Here, housing is a good proxy for economic growth: People are likely to build or buy houses as their incomes increase. We measure stock by taking the difference between 2011 and 2017 (per National Institute of Statistics), and then standardizing the difference by the 2011 numbers. The number of houses built increased by 2.24%—with 152 municipalities seeing shrinkage.

The results in Table 4 suggest that while ethnic polarization does not have a significant association with either business density (model 1) or housing stock (model 2)—this is the case only when we consider its effect unconditionally. Once we consider its effect conditional on whether the municipality recognizes a minority language, we see results that are consistent with those using the local human development index. When minority language is recognized, we see that diversity has a significant and positive effect on private goods production (β=3.65; SE=1.55 for business density; β=3.34; SE=1.14 for housing stock). Conversely, when minority languages are not recognized ethnic polarization has no effect on either of these outcomes.

Table 4 Alternative Variable Specifications
Full size table

Next, we consider an alternative measure for ethnic diversity. We used polarization because we believed it best captured ethnic parallelism in Romania generally and Transylvania specifically. For robustness, we also look at ethnic fractionalization—constructed as a Herfindahl Index (see Alesina et al. 2003). The measure reflects the probability of drawing two people from two different ethnic groups. A minimum value of 0 suggests a completely homogeneous municipality; and conversely, a maximum value of 1, a completely heterogeneous municipality. The results in model 3 remain substantively unchanged. In areas where a minority language is not recognized, a one-unit increase in ethnic fractionalization can decrease local human development index by over 6 points. However, the same fractionalization shift in areas where a minority language is recognized has no significant effect whatsoever.

Language Recognition Versus the Hungarian Effect

While the principle of language recognition applies theoretically to all minorities, there is no doubt that the law has the most implications for the Hungarians—the largest minority group in Romania. There are at least two reasons why the Hungarians may be different from the other minority groups. First, the Hungarians are politically relevant. Yet, relations with the Romanian majority have not always been amicable. In Transylvania, the ethnic strife for social positions goes back for centuries (Brubaker et al. 2006; Livezeanu 1995; Verdery 1983). There is still evidence of intergroup distrust, perceptions of large social distances, and tendencies to symbolically delimit the other (Csepeli et al. 2002; Veres 2011).

The second reason why the Hungarians may be different from the other minority groups has to do with diaspora politics. As part of its foreign policy, the Hungarian government under Prime Minister Viktor Orbán has considerably increased funding to Hungarian communities abroad—whether directly through public finances or indirectly through private firms partially owned by the Hungarian state. As for the former, investigative journalists suggest that Transylvania received about 500 million euros between 2011 and 2020 in financial support from Hungary.Footnote 10

While this is a substantial amount of funding, it is important to disaggregate the numbers by when: Of the aforementioned 500 million euros, 100 million was allocated only in 2020—suggesting limited influence on a local human development index from 2011. It is also critical to disaggregate the numbers by for what: Most of the funds have been directed to cultural matters, e.g., cultural heritage centers, churches, childcare centers, media, and sports clubs (see Keller-Alánt 2020). While these are important institutions, their impact on economic development is more indirect. And finally, it is also essential that we remain cognizant of where the money goes. If funds are meant to support the diaspora and encourage transnational engagement (Kovács 2020), financing numbers are likely to correlate with the absolute number of Hungarians in a locality. Put differently, larger cities with more Hungarians will get more financing independent of language recognition status.

To test the Hungarian effect, we employ a three-prong strategy. First, we rerun our full model with a dummy control for the Szekler region. Two counties in Szeklerland (Covasna and Harghita) are Hungarian dominant. Additionally, 38.1% of Mureș County is also Hungarian. It is possible the municipalities from these three counties are driving the results. Second, we rerun the full model again but with a control for the Hungarian population number. Since the number ranges from 0 to almost 60,000—with a very skewed distribution to right—we log transform the variable. And third, we employ a different measure for Language Recognition. Instead of assigning a municipality that recognizes any minority language with a value of “1,” we restrict it exclusively to the Hungarian language. Thus, a municipality that recognizes Ukrainian—but not Hungarian—would now be assigned a value of “0.” We then compare the models to see how much the results differ.

The results in Table 5 suggest that our substantive findings remain unchanged even if we focus strictly on the Hungarian minority. Ethnic polarization decreases development, as does Hungarian language recognition—but this is only the case from an unconditional standpoint. Once we look at their conditional effects on one another, we see language recognition can attenuate the negative effects of diversity. When accounting for the Hungarian-dominant counties (model 1), the association between EPI and LHDI is negative in municipalities that do not recognize a minority language (β=-3.58; SE=0.81). The same association is nonsignificant when a minority language is recognized. The same pattern holds for the other two robustness checks. In model 2, a one-unit increase in EPI decreases LHDI by 4.5 points (SE= 0.86) when a minority language is not recognized; in model 3, the same one-unit increase in EPI associated with a 3-point decrease in LHDI (SE=0.81). In sum, across these models, the moderating effects of language recognition remain unchanged.

Table 5 The Hungarian Effect
Full size table

Language Recognition Versus Roma Effect

It is hard to talk about ethnic diversity and local development in Romania without discussing the Romas. The discrimination facing the Romas is well-documented (Vincze 2014; Vincze et al. 2018). This is also supported by our data—e.g., in municipalities with higher proportions of Romas, the unemployed numbers are substantially higher. The inequities, however, do not manifest just for the dependent variable (i.e., local human development index); they also matter for the independent variable (i.e., language recognition). While the Romas are the second largest ethnic group in Romania, they are not afforded many of the same rights as the other minority groups. While ethnic Hungarians may lament the de facto inequities in language support, the Romas are de jure shunned—i.e., Romani is not a language recognized by the state. The 2001 Law on Local Public Administration enumerates language recognition for ethnic minorities generally, but does not note which minorities specifically. Likewise, the law for the ratification of the European Charter for Regional or Minority Languages in Romania (No. 282/2007) recognizes a list of minority languages (Article 10). Alas, Romani is not one of the listed languages. As a result, in areas with sizable Romani populations, Romani is largely absent from public use (Pop 2009: 72). In short, the effects we find may have little to do with language recognition and may be more about the Roma population.

To consider this, we employ three strategies. The first is to recalculate the ethnic polarization index with the Romas removed. This is not to say the Romas do not matter. But if we are still able to find a similar effect, then we know the purported mechanism about ethnic group relations manifests between any non-Roma minority group and Romanians as well. The results in Table 6 (model 1) demonstrate we still see similar dynamics independent of the Romas. In areas where a minority language is not recognized, increasing ethnic polarization between the Romanians and other minority groups—but not the Romas—is associated with a 1.58 decrease in LHDI. This decrease is statistically significant. However, in areas where a minority language is recognized, ethnic polarization has no significant effect.

Table 6 The Roma Effect
Full size table

As an alternative, we run the standard full interaction model with ethnic polarization; but this time, we also include the Roma population—not as a proportion but as an absolute number. We do not use proportions because they are constituent parts of the ethnic polarization index in each locality; instead, an absolute number would allow us to roughly control for a large (or small) Roma population. The results in model 2 show that the inclusion of the Roma population explains about 12% of the local human development variance. Specifically, every additional 269 Romas in a municipality (sample standard deviation) decreases human development by as much as three points. But even so, the marginal effect of ethnic polarization remains negative (positive) in municipalities without (with) language recognition.

Our third strategy is to do matching analysis. Matching can reduce possible bias in model estimates by ensuring a balanced set of covariates between the controlled (no language recognition) and the treated (language recognition) units. Matching can also help account for omitted variables, although only to the extent that these variables correlate with the included covariates. Since covariate balance is our primary aim and given that the treatment assignment is deterministic—i.e., all municipalities where minorities constitute more than 20% of the population are treated—we rely on one-to-one, nearest-neighbor matching with the purpose of minimizing Mahalanobis distance between observations. We deal with missing data through listwise deletion. This results in a matched dataset of 498 observations (249 per control and treatment group). Table 7 presents covariate values prior to and after matching. In general, we see balance improvements across most of the variables, including Roma population. Results from the matched dataset remain consistent (model 3). The marginal effect of diversity in municipalities where a minority language is not recognized remains negative and significant; in municipalities without language recognition, the marginal effect is positive and nonsignificant. While a large Roma population decreases development, the recognition of another minority group’s language can attenuate the otherwise negative effects of diversity.

Table 7 Balance—Matching Analysis
Full size table

Discussion

This paper examined the socioeconomic effects of ethnic diversity—conditional on minority language recognition—in Romania. We find that in areas where only Romanian is used in public administration, ethnic diversity—whether we measure it as polarization or fractionalization—has a negative effect on human development. However, when a second or third language is recognized, diversity has no effect—if not a positive one—on the economy. From a theoretical standpoint, this effect runs counter to the likes of Easterly and Levine (1997) who talk about ethnic heterogeneity as a sort of “growth tragedy.” Instead, it corroborates the arguments put forth by Liu et al. (2015) that the recognition of non-majority languages can help standardize diversity.

Theoretically, this paper addresses a gap in the literature on ethnic minorities in Romania. On the one hand, there are scholars who look at the effects of ethnic diversity on regional planning from a geographic perspective (Benedek 2009; Benedek et al. 2013; Benedek et al. 2018). And then on the other hand, there are the sociologists (Csata 2017; Kiss 2010, 2014) and economists (Andrén 2012) who examine the social and economic positions of the Hungarians vis-à-vis the other ethnic groups. There has been little work looking at the socioeconomic effects of ethnic diversity. This paper seeks to bridge these two bodies of scholarship.

The policy implications of this paper are clear. If the Romanian government cares about improving the socioeconomic conditions in Transylvania, it is imperative that it maintains the territoriality principle. However, it would behoove the government to lower the threshold from 20%. Doing so would allow more minorities to feel they are valued by the state. It would also incentivize Romanians to learn the minority language to some degree—thereby facilitating inter-ethnic cultural understanding.