What is the effect of gender on interethnic trust? It is human nature to classify people around us into us-them categories (Tajfel Reference Tajfel1982). This in turn matters for interethnic trust. Whether someone trusts a member of a different ethnic group can depend on an individual trait (e.g., more cosmopolitan—see Hu and Liu Reference Hu and Liu2020); an intra-ethnic group factor (e.g., whether the group is the majority or dominant in the capital—see Huang and Schuler Reference Huang and Schuler2018); or the interaction among ethnic groups (e.g., cultural distance can affect trust—see Adida Reference Adida2014). Consistent across explanations for interethnic group trust is the general absence of gender. Yet, gender and ethnicity intersect (Bird Reference Bird2003; Bratton, Haynie, and Reingold Reference Bratton, Haynie and Reingold2007; Hughes Reference Hughes2011), with distinct consequences for trust and intergroup conflict. For example, gender can moderate outgroup trust by or toward specific individuals; likewise, outgroup trust can shape changes in societal gender norms.

However, whatever attention is given to gender tends to focus on the subject—that is, the gender of who is doing the trusting. The results are often mixed: some studies find women are more hostile to outsiders (Golebiowska Reference Golebiowska1999); others find women are more tolerant (McLaren Reference McLaren2003); and then there are those that find no gendered effect (Hayes and Dowds Reference Hayes and Dowds2006). We contend one reason for these mixed results is that the object being entrusted—that is, what is being entrusted—is unspecified (Bracic Reference Bracic2020; Gries and Turcsányi Reference Gries and Turcsányi2021) or assumed genderless (Bahry et al. Reference Bahry, Kosolapov, Kozyreva and Wilson2005). We thus direct our attention to the gender of an important entrusted object in ethnic relations: children. Unlike measures that address people’s attitudes or social networks, questions about intermarriage tap into norm-based and legal dimensions of intergroup interactions. Additionally, marriage involves multiple generations, with implications for family lineage, property rights and inheritances, and the ethnic group at large (Fouka, Mazumder, and Tabellini Reference Fouka, Mazumder and Tabellini2022).

We argue people are less likely to agree to their daughters—but not their sons—marrying an ethnic outgroup. There are two possible mechanisms. The first is that most contemporary societies rely on legal authority. The family name—and related the legal status—continues through sons. This renders the ethnicity of the future son-in-law more relevant. The second is about protection and social norms. Specifically, daughters may be perceived to be weaker and therefore necessitate protection. These mechanisms manifest regardless of the individual’s own gender.

We test our argument with a survey experiment in Romania. We focus on Romania for three reasons. First, the society is still largely patriarchal: if there are gendered effects, we should observe them here. Second, the country is ethnically diverse. Romania is home to a large Hungarian population, one of the largest Roma communities in Europe, and several migrant groups including the Chinese. This diversity ensures that the noted ethnic outgroup in the experiment is one the respondent would credibly know. Third, Romanian is a Romance language where articles, adjectives, and nouns—including children—are gendered. To evaluate whether the child’s gender matters, we must use a language with such linguistic structure. In short, Romania provides the theoretical, empirical, and methodological leverage to test a gendered theory about outgroup trust.

Our findings are consistent with a multifaceted understanding of how gender influences intergroup trust. Our results suggest that intergroup trust varies—even when concerning the same individual—depending on the gendered context. Examining how a gendered lens can alter an individual’s trust levels reveals important theoretical insights for those who study intergroup trust and ethnic relations. Specifically, by noting whether the object of an individual’s trust is female, male, or non-gendered, we are able to evaluate whether norms and expectations depress an individual’s propensity to trust the outgroup.

Explaining interethnic trust

One explanation for why members of one ethnic group trust—or do not trust—those of another group rests on social identity theory. It is human nature to form us-them differences (Tajfel Reference Tajfel1982). There is the pull of ingroup favoritism. People prefer to engage with those just like them: ingroup identification and centrality mean more congruent preferences (Miguel Reference Miguel2004) and more interaction efficiency (Habyarimana et al. Reference Habyarimana, Humphreys, Posner and Weinstein2009). There is also the push of outgroup stereotyping. People want to identify positively with their group (Stewart Reference Stewart2000). Branding the other group as a threat and subjugating them to a disadvantaged social class can provide psychological satisfaction (Tajfel Reference Tajfel1982). Given the ingroup favoritism and outgroup stereotyping mechanisms, it is difficult to build trust—particularly when the cultural gap is distinct (Gradstein and Schiff Reference Gradstein and Schiff2006; Sniderman, Hagendoorn, and Prior Reference Sniderman, Hagendoorn and Prior2004). It is easy for groups to vilify the cultural unknowns of the other group and to interpret misunderstandings as ill-intentioned acts (Hainmueller and Hopkins Reference Hainmueller and Hopkins2014).

There are, however, two caveats to this argument. First, a large cultural gap is not necessary for fostering outgroup distrust. In fact, it is possible that the distinctiveness makes them precisely just that—that is, they are different. Instead, the threat may come from someone closer to home (Kim and Lee Reference Kim and Lee2022; Liu, Power, and Xu Reference Liu, Power and Xu2022). When two groups are culturally similar, proximity is viewed with suspicion. There is a concern about resource redistribution: What happens when members of the outgroup claim social benefits earmarked for the ingroup? There is also the threat of defection: What happens if members of the ingroup find they derive more satisfaction as outgroup members? Here, it is the porous group borders—as opposed to the fixed rigid ones with distinct outgroups—that make it necessary to distrust the ethnic outgroup (Adida Reference Adida2014).

Second, a large cultural gap is not sufficient for breeding distrust. The gap needs to be politicized (Hopkins Reference Hopkins2010; Posner Reference Posner2004). We see politicization manifesting as economic and welfare threats (Bocskor Reference Bocskor2018), security considerations (Krzyżanowski, Triandafyllidou, and Wodak Reference Krzyżanowski, Triandafyllidou and Wodak2018; Zaslove Reference Zaslove2004), and pandemic responses (Adida, Dionne, and Platas Reference Adida, Dionne and Platas2018; Liu, Power, and Xu Reference Liu, Power and Xu2022). Conversely, a large gap can remain un-politicized or framed in a nonnegative way (Liu Reference Liu2021). In this case, a culturally distinct group is left alone. Members of each group are cognizant of group differences, but there is no barrier to facilitating interethnic trust.

Identities, however, are multidimensional; they cross-cut (Selway Reference Selway2015) and they overlap (Liu and Ricks Reference Liu and Ricks2022). What is consistent in this literature is the general absence of gender. In fact, it is fitting to note that the man who pioneered social identity theory—Henri Tajfel—denied gender as a relevant identity for group formation (Tajfel Reference Tajfel1982).Footnote ¹ While gender may not have the “minimal group” properties that Tajfel observed with ethnicity, it has the unique property of intersecting every single ethnic group (Bird Reference Bird2003; Hughes Reference Hughes2011). It is therefore highly likely that gender exerts an effect in shaping intergroup trust. In the next section, we elucidate the different mechanisms.

But let us first consider trust. Trust is something that is normatively desired (Putnam Reference Putnam1995) and of great importance (c.f., Levi and Stoker Reference Levi and Stoker2000)—yet measuring it can be difficult. One survey approach is to ask the very macro question, “Do you agree most people can be trusted?” and offer binary or a few scalar responses. Alternatively, there are those tap into an element of trust—for example,, feelings (Gries and Turcsanyi Reference Gries and Turcsányi2001), discomfort (Liu Reference Liu2020), or willingness to engage (Bracic Reference Bracic2020). But these measures often take place in an object vacuum. There is no attention to what is being entrusted, let alone the gender of what is being entrusted. If there is something to be entrusted, it is usually money in a hypothetical dictator’s game. And here, money is assumed to be genderless.

Trust, however, depends on the interaction between an individual and what they are being asked to entrust. Trust is shaped by the sense of accountability and the perceived costs of the relationship. For example, trusting the Etsy seller to deliver the items you ordered is qualitatively different from trusting your accountants to file your tax papers properly. There is a clear distinction in accountability when comparing Etsy’s reporting and request functions to fines and audits by the Internal Revenue Service. Likewise, trusting your tenants to not drill holes in your walls is substantively different from trusting the police when they have pulled you over. The entrusted object—Etsy seller versus tax accountant, tenant versus police—and accountability mechanisms that constrain them thus impart critical information about the stakes for a subject. Without identifying the entrusted object, respondents may be thinking of different items under different legal and social restraints. And even if respondents are thinking of the same item, gender matters. For example, willingness to trust the police can depend on whether the officer is male or female.

Here, we focus on a very high-stakes object for entrusting when it comes to intergroup relations—and one that allows us to clearly identify gender: the child. When a child marries, the parents are entrusting the future spouse to take care of their child and build a life together—with implications for grandchildren specifically and ethnic group survival generally. While there are mechanisms to bring families together and protect the longer-term interests or treatments of both parties—for example, prenuptial agreements, dowries, and bridewealth payments—many often function in suboptimal ways. For example, studies on bridewealth payments in Zulu culture indicate that a man paying the bridewealth does it to signal his value for his wife. Yet, this can result in women feeling or being trapped in abusive marriages because their families are unable to repay in the case of a divorce (Ansell Reference Ansell2001; Rudwick and Posel Reference Rudwick and Posel2015, 297). Unlike houses, cars, and money, children are not inanimate objects; thus, we can leverage how gender affects interethnic trust.

Why gender matters for interethnic trust

Parents may have gendered relationships with their offspring such that (1) mothers and fathers may raise their “children” in distinct ways and/or (2) “parents” may raise their daughters and sons differently. Additionally, mothers may raise their sons in one way but their daughters in another; likewise, fathers may treat their sons differently from their daughters. As a result, there are numerous possible paths for gender—of the parent and/or the child—to potentially influence (1) the parent–child relationship and (2) how it in turn affects outgroup trust.

There are a number of explanations for why parents may be inclined to entrust daughters differently from sons to outgroup marriages. One focuses on gender—especially via social norms and ensuring behaviors. Fathers are often associated with less involvement in raising children and devoting time to childcare (Lamb Reference Lamb2000). Yet simultaneously, they are more likely to be engaged when raising a son (Raley and Bianchi Reference Raley and Bianchi2006). This is due to factors such as traditional, socialized gender ideologies that stipulate maternal rather than paternal caregiving (Kluwer et al. Reference Kluwer, Heesink and Van de Vliert2000), women’s greater emotional expressiveness (Goldschmidt and Weller Reference Goldschmidt and Weller2000), and structural—but gendered—division of labor that keeps women in caregiving roles (Presser Reference Presser1994). Even so, fathers have been traditionally assumed to occupy the role of the family disciplinarian. Thus, fathers are (perceived as being) required to take more active roles in matters related to misbehavior and schoolwork (Holland Reference Holland1994). Conversely, women are (perceived to be) both more involved and more controlling parents. This is often the result of being more involved in the first place. Women generally devote more time to caregiving regardless of paid work status (c.f., García-Mainar et al. Reference García-Mainar, Molina and Montuenga2011)—a long-established phenomenon made starkly clear during the COVID-19 pandemic (Petts et al. Reference Petts, Carlson and Pepin2021; Zamarro and Prados Reference Zamarro and Prados2021).

More recent work shows emerging shifts in the traditional, gendered socialization of parents. The degree and type of shift, however, largely depend on national and cultural context. The conventional style of “breadwinner” fathers has been replaced in recent generations by more nurturing and engaged fathers (c.f., Jarska Reference Jarska2021; Kaplan and Knoll Reference Kaplan and Knoll2019)—particularly where parental leave policies enable fathers to take more substantial time off from paid employment (Almqvist and Duvander Reference Almqvist and Duvander2014). While shifting gender attitudes and economic conditions lead to some evidence of “daughter preferences” in contemporary Japan (Fuse Reference Fuse2013), in many societies, “son preferences” still prevail for economic and/or behavioral reasons, especially for immigrant and economically vulnerable populations (Blau et al. Reference Blau, Kahn, Brummund, Cook and Larson-Koester2020; Filmer et al. Reference Filmer, Friedman and Schady2008). Taken together, this points to differences in parental preferences for daughters and sons, how parents spend time engaging with daughters and sons, and the overall value accorded to daughters and sons.

Beyond preferences, goals, and roles in childhood, parents are often involved in their children’s choice of partners. This influence can be communicated by overt expressions of preferences or via passive transmission of norms and values. We see parental influence on marriage choice when it comes to religious and ethnic identities. Studies have shown that religious and ethnic groups may proscribe expectations regarding the (in)appropriateness of marriage to a member of a different religious or ethnic group (c.f., Perry and Whitehead Reference Perry and Whitehead2015; Van Niekerk and Verkuyten Reference Van Niekerk and Verkuyten2018). In Suriname, members of the Hindustani and Maroon ethnic groups are still significantly more likely than other large ethnic groups to favor parental control of their children’s choice of partner and objection to intermarriage (Buunk et al. Reference Buunk, Leckie and Pollack2020). Parental involvement in communicating and (re)enforcing these expectations can lead to gendered differences that constrain who parents believe to be acceptable for daughters in countries as diverse as Belgium (Blycq 2012), Egypt (Van Niekerk and Verkuyten Reference Van Niekerk and Verkuyten2018), and the United States (Perry and Whitehead Reference Perry and Whitehead2015).

Through and beyond the preferences and relationships that evolve during the pre-infancy and childhood stages of life, offspring have value to and bonds with parents that are influenced by gendered social and legal norms. These social and legal norms shape how a parent considers entrusting their offspring to another individual, family, or community. Systemically, transmission of familial interests could occur through two separate channels: by blood or by legal distinction. The observation of matrilineal descent practices—sometimes codified in laws preventing intermarriage between adherents of different faiths—is an example of transmission by bloodlines. Here, parents worry about continuing their ethnic group’s genetic material; the focus is on how marriage and subsequent production of offspring carry on their legacy via the family bloodlines. In this context, marriage concerns should matter less for whom daughters marry. This is because when the daughter bears the grandchild, the parents know that the grandchild’s blood is at least 50% from their ethnic group. This largely follows the same logic and sometimes coincides with matriarchal legal systems in which property and inheritance rights pass through the eldest female child. As sons do not necessarily pass on their genetic material once they marryFootnote ² —nor will they inherit the family’s property—remaining part of the group may necessitate marrying someone from the ingroup. In such systems, parents should be (1) more willing to permit daughters to marry outgroup members, but (2) concerned that sons marry within the ingroup.

In general, however, bloodlines matter less frequently than the legal status conferred via traditional, patriarchal systems where sons inherit property and the family name, thus carrying on recognized membership in the ingroup. Here, whose blood runs through the veins of subsequent offspring is of secondary relevance. Instead, what is of primary importance is that an ingroup ethnic name—and thus ingroup membership—is carried by said offspring. In this context, marriage concerns should matter less for whom sons marry, because the son’s offspring will bear the family name, inherit the family’s property, and remain a recognized ingroup member. This is regardless of the mother’s ethnic origins. As daughters do not typically retain the family name once they marry—nor will they inherit the family’s property—to remain part of the ingroup may necessitate marrying someone from within the ingroup. In such systems, parents should be (1) more willing to permit sons to marry members of an ethnic outgroup but (2) more insistent about their daughters marrying within the ingroup.

There is a second distinct gendered mechanism. It is about a parent’s general sense of innate protectiveness and specifically, the gender biases that shape how it manifests. While parents may want to protect all their children, societal gender biases suggest there are more concerns when the child is a girl. Since many societies still rely on a male breadwinner/male head-of-household model that governs domestic environments, the assumption is that one’s daughter will be subjected to the wills of whatever her husband wants—for example, when to have children, where to live, and how often to visit with family. Put differently, parents may worry that allowing a daughter to marry someone from an ethnic outgroup will result in her being subjected to the cultural whims of that group. Conversely, the son is assumed to be able to withstand the challenges of an intermarriage because he will become the head of his household, and thus he will not be forced into any undesirable traditions of the ethnic outgroup. Under such considerations of protectiveness, parents should be less willing to permit daughters to marry outgroup members.Footnote ³ Though we do not test these mechanisms directly, they should lead to the same observable outcome: parents will oppose outgroup marriage prospects for their daughters more than their sons.

Not all ethnic outgroups, however, are considered equally “distant” from the ingroup. Religious beliefs, degree of religiosity, and nationalism can affect whether a specific group is more acceptable than others (Perry and Whitehead Reference Perry and Whitehead2015; Van Niekerk and Verkuyten Reference Van Niekerk and Verkuyten2018). Given that our theoretical lens focuses on when differences from the ingroup drive opposition to intermarriage, we expect ethnic outgroups that are perceived to be culturally distant from the ingroup should elicit more pronounced opposition from a parent and that this effect should be more pronounced for daughters. Given this discussion, we predict the following hypotheses:

Research design

To test our argument about the effects of gender on interethnic trust, we focus on Romania. There were three considerations for case selection. First, from a theoretical standpoint, we needed a case where society is still largely patriarchal. If there are gendered differences when it comes to parental preferences over intermarriage, we should observe them in this case. Evidence of this patriarchy manifests in public attitudes. For example, as per the World Values Survey, 38.1% of the Romanian respondents agreed that men make better political leaders than women. In comparison, only 5.2% of Swedish respondents, 8.8% of Spanish respondents, and 16.4% of respondents from the United States similarly agreed. Note, however, that Romania is far from being an outlier: the global average of the full sample (N = 80 in Wave 7) is 33.4%.

Second, from an empirical standpoint, we needed a case where there was ethnic diversity—not only in the number of groups but also in their dynamics. In Romania, we are able to leverage four groups. There are the Moldovans: coethnics who live in a neighboring state because of Joseph Stalin; public attitudes toward Moldova and Moldovans are routinely positive. There are the Hungarians: a large territory-based and self-professed “indigenous” minority that became part of Romania after WW1; relations with Romanians have not always been amicable. There are the Romas: a spatially dispersed minority population that has been in Romania for centuries; they are subject to extreme discrimination in Romania and across Europe. And finally, there are the Chinese: a large, new migrant population concentrated in the cities; Romanians have generally tolerated them—even during the pandemic.

Third, from a methodological standpoint, we needed a case where the language is heavily gendered, thus allowing us to manipulate the text for the treatments. Romanian is a Romance language. Like Spanish, all articles, adjectives, and nouns are gendered. If we are interested in whether individuals entrust daughters and sons in different ways, it is imperative that we field a survey experiment in a language that forces respondents to consider the child’s gender. One potential concern, however, is that if we do find a systematic difference, we cannot differentiate whether the result is due to some cultural value—a value that is transmitted through the Romanian language—or some linguistic feature in the language itself. When a language like Romanian is heavily gendered, it demarcates the masculine from the feminine in every part of the sentence. Conversely, when a language is genderless, it is possible that the distinction is not as pronounced. There is evidence that when people use a gendered language, this can constrain more gender-egalitarian attitudes (Liu et al. Reference Liu, Shair-Rosenfield, Vance and Csata2018; Pérez and Tavits Reference Pérez and Tavits2019)—potentially dampening parents’ perception of the suitability of intermarriage, particularly for their daughters. Here, the Romanian case allows us to test this distinction as the national language is heavily gendered, but the Hungarian language—the language of the largest minority—is genderless.

Survey experimental design

We worked with a survey company in Romania (Kantar TNS) to field our survey experiment (University of Texas at Austin IRB Study 2019100041). The surveys were administered in November 2019 to a nationally representative online panel of 505 respondents. The sample included quotas on age, gender, regions, and urbanization level. By using an experimental language manipulation embedded into the survey, we can pinpoint specific aspects in the intersection of gender and interethnic group trust. Specifically, we manipulate the word for “child”—and the gendered nature of the child—to determine how gendered considerations prompt differences in intergroup trust levels.

The survey was administered primarily in Romanian (the survey was also available in Hungarian to the ethnic Hungarians—more below). There are two parts to the survey. In the first part, there is a series of demographic questions, including the respondents’ age, education level, employment status, household income, location type, county of residency, and—of relevance here—gender. In the second part, we introduce the treatment. There is one control group (child) and two treatment groups (treatment 1: daughter; treatment 2: son). The prompt and distribution for each group are shown in Table 1. Respondents are asked whether they are okay with their offspring marrying someone in the outgroup—with responses ranging on a five-point scale from strongly disagree (0) to strongly agree (4). The outgroup list includes four ethnic groups: Moldovans, Hungarians, Chinese, and Romas (ordering randomized per respondent).

Table 1. Survey experiment prompt per group

A question about neighboring Moldovans allows us to tap into a large coethnic, but distinctly non-conational, population. The vast majority of Moldovans are ethnic Romanians who speak the Romanian language and practice Christian Orthodoxy. The country itself was the product of Soviet politics. In 1940, Stalin took territories from Romania—including the eastern portion of the Moldavian principality—and created the Soviet Socialist Republic of Moldova. The country of Moldova emerged with the collapse of the Soviet Union. Public attitudes among Romanians toward Moldovans are generally positive. In a 2014 survey about who people would not want as neighbors, fewer than 10% said Moldovans (Liu Reference Liu2020). These attitudes have remained stable: even during the strictest COVID-19 lockdown (April 2020), only 9.9% said they did not want a Moldovan neighbor (Liu, Power, and Xu Reference Liu, Power and Xu2022). In short, Moldovans are the control ethnic group in our sample. Here, we expect agreeability for marriage to be highest for this group—whether in the aggregate or disaggregated by the child’s gender.

Next, we ask about the Hungarians—that is, the largest ethnic minority in Romania (Csata et al. Reference Csata, Hlatky and Liu2021). Although Hungarian minorities are regularly included in the government cabinet, Romanian–Hungarian relations have had episodic periods of tensions (Csergo Reference Csergo2007) with strong ethnic parallelism today (Kiss et al. Reference Kiss, Székely, Toró, Bárdi and Horváth2018) —whether in education (Culic Reference Culic2019), economic activities (Csata Reference Csata2020), or media consumption (Csata et al. Reference Csata, Compton, Liu and Attila2023). Yet, mixed marriages are not uncommon (Gyurgyik et al. Reference Gyurgyík, Horváth and Kiss2010): 18.1% of Hungarian marriages in Transylvania 1992–2007 (the area with the large Hungarian population) are interethnic (Kiss Reference Kiss and Apró2016). We expect respondents to be less supportive of their child marrying a Hungarian than a Moldovan.

Next, we ask about the Chinese. We focused on the Chinese for multiple reasons. First, they are one of the largest migrant populations in Romania, making them socially salient (Liu Reference Liu2021). Second, they are a culturally distinct outgroup. And third, their relatively recent arrival (post-1989) means less time has passed through which multigenerational engagement could occur, as is the case with the Hungarians. This multigenerational element matters when the dependent variable asks about marriage of a child. A priori, we would expect respondents to be less supportive of their child marrying a Chinese than a Moldovan. We are, however, agnostic about the difference between Hungarians and Chinese. On the one hand, while the Hungarians may be more culturally proximate, historical tensions may make intermarriage less acceptable. On the other hand, Romanian attitudes toward the Chinese have been generally positive. Before COVID-19, only 14.2% of Romanians objected to having Chinese neighbors specifically (Liu Reference Liu2020). While this number increased to 18.0% in April 2020, it was still lower than the 21.9% for coethnic Romanians returning from Italy when that country was a hotspot (Liu, Power, and Xu Reference Liu, Power and Xu2022).

Finally, we ask about the Romas. The Roma population in Romania is also sizable; Romania is home to almost 20% of Europe’s Roma population (Bilefsky 2013). Both in Romania specifically (Csata, Hlatky, and Liu Reference Csata, Hlatky and Liu2021) and in Europe generally (Bracic Reference Bracic2020), Romas are subject to some of the worst discrimination. Consider that 71% of Roma households live in poverty, 67% of the Roma population cannot afford basic medical care, and a Roma individual has only a 29% chance of completing secondary education (The World Bank 2015). Despite different government bodies—from local municipalities to the national government to the European Union—taking steps to address this discrimination, Romas are simply “unwanted” (Astier Reference Astier2014). In a 2014 survey, 58.9% of the respondents did not want a Roma neighbor (Liu Reference Liu2020). The number remained high during the pandemic (59.7%; Liu, Power, and Xu Reference Liu, Power and Xu2022). Of the four groups, we expect responses for this group to be the least positive.

Sample balance

The samples are balanced (see Table 2). Half of the respondents (51%) are female. Likewise, about 67% of the respondents indicated they have at least one child. The gender distribution of parental status is relatively uniform. Respondents are on average about 40.5 years old. More than 60% of the sample have at least a university education and are employed full-time (including self-employed). Income is a self-reported categorical scale where the average respondent has a monthly net income between 2001 and 3000 Romanian RON (478–717 USD). The Hungarian minority constitutes about 2.5% of the sample. There is some variation in location type, with a ten-percentage point gap between urban respondents in the daughter and son treatment groups; however, the difference is not statistically differentiable. Moreover, we see similar figures for the four largest cities and surrounding regions in Romania (Bucharest, Cluj, Iasi, and Prahova).

Table 2. Sample balance

Empirical evidence

Before examining the effects of the child’s gender, we first examine the average effects across the four ethnic groups. The results (Fig. 1) suggest that Moldovans are the most acceptable outgroup for marriage (2.94 out of a maximum of 4). The numbers decline for the other three groups. For example, approval for marrying a Hungarian and a Chinese is 2.66 and 2.41, respectively. Note that while these two numbers are (1) both statistically different from that of the Moldovans, they are (2) not statistically different from each other—thereby lending support for our prior agnosticism. As expected, responses for marrying a Roma are largely negative. With a mean of 1.86, more than half of the respondents definitively said they disagreed with the statement that “it is okay that my child [son/daughter] marries a Roma.”

Figure 1. Means test: effects of outgroup ethnic identity on intermarriage acceptability (95% CI)

Having established that the identity of the outgroup matters for intermarriages, let us now test the effects of the child’s gender. A simple means test—with 95% confidence interval—suggests that it does (see Fig. 2). Recall, the variable is scaled such that higher values indicate more acceptance of intermarriage. We indeed see treatment effects. Let us focus first on the daughter treatment. Across all four ethnic groups, respondents are statistically less likely to approve of their daughters being in an intermarriage than the generic child.

Figure 2. Means test: gender of child in consideration by ethnic outgroup (95% CI)

But it is not just about daughters versus the control group. The ethnic identity of the outgroup also matters—and support declines significantly as the outgroup is perceived to be farther from the ingroup. While (1) the Hungarian effect for daughters is statistically no different from that of the Moldovans, and (2) the Chinese effect for daughters is statistically no different from that of the Hungarian, there is, however, (3) a difference between support for daughters marrying a Moldovan (more support) versus a Chinese (less support). And as expected, the effects for daughters marrying a Roma are the absolute lowest. This corroborates our argument that willingness to have daughters marry outgroups decreases as the cultural gap widens. It is also consistent with the protective mechanism. Societal gender biases suggest that daughters need more protection, but it is not necessarily just relative to sons. It is also about the ethnic identity of the (potential) sons-in-law. When (potential) sons-in-law are from distant, distinct, and possibly disdained outgroups, there is the innate concern that the daughter will suffer. This worry manifests as less willingness to have the daughter marry into these outgroups.

Conversely, when the focus is on the son treatment, the results are less consistent. As we see in Fig. 2, while the means for sons are always larger than those for the generic child, the differences are not necessarily significant. This suggests that the baseline for child may still be subconsciously male. This is not far-fetched considering the supposedly genderless child (copil) in Romanian is still masculine. And while the effects of the son treatment may not be significantly different from that of the control group, this is only the case when the (potential) daughter-in-law is Moldovan (nominally a coethnic) or Hungarian.

Conversely, when the focus is on the Chinese as the outgroup, support is statistically different from the control group. The difference is also significant when it is about sons marrying a Roma. In both cases, the difference for sons—contrary to daughters—is in the positive direction. This aligns with our expectation about outgroup distance and our theorized mechanisms through which gendered trust operates. When the ethnic outgroup is clearly distant, distinct, and possibly disdained, ensuring the ethnic Romanian family name carries through a son matters substantially.

To better understand the effects of how the child’s gender affects intermarriage approval, we run a series of multivariate regressions. We estimate four models—one per outgroup—using ordered logit with county fixed effects and robust standard errors. The results can be found in Table 3. At first glance, the direction and significance of the coefficients remain consistent across outgroups. The son treatment is always positive, and the daughter treatment is always negative. What is striking, however, is the magnitude of effect with the daughter treatment. When respondents are primed to think about a daughter, they are 40–60% less likely than those in the control group to be comfortable with their child marrying an outgroup.

Table 3. Multivariate regression: Effects of gender treatment on interethnic group trust

Notes: Dependent variable: 0 = strongly disagree that intermarriage is okay; 4 = strongly agree that intermarriage is okay.

^a Models estimated using ordered logit with county fixed effects.

^b Reference category: control (child).

* p ≤ .10.

** p ≤ .05.

*** p ≤ .01.

Considering the gender of respondent

Here we have focused on the gender of the object being entrusted—that is, the what—in interethnic relations. Thus far, we have remained agnostic as to the subject doing the trusting—that is, the who. While the more historic gender gap in political tolerance traditionally favored the idea that women were less tolerant than men, the modern gender gap in politics suggests the opposite: women are perceived to be more liberal (and thus more tolerant) than men. There is some evidence that women are more religious than men, but they are less religiously dogmatic (Schnabel Reference Schnabel2018). Women are also less likely to support the far right (Coffé Reference Coffé2016; Givens Reference Givens2004; Spierings and Zaslov Reference Spierings and Zaslove2015; Reference Spierings and Zaslove2017). Consider that the coefficient for the respondent’s gender in Table 3 is never statistically significant. This suggests that there may be a shift in gender-based social norms for parents.

This non-effect persists even if we consider the effect of the respondent’s gender interacted with parental status. In Fig. 3, we consider the numbers for Moldovans (expected to be the most acceptable for intermarriage) and Romas (expected to be the least acceptable). When we look at Moldovans, the only significant difference occurs when considering parental status rather than the respondent’s gender. Even so, this difference disappears when we consider intermarriage with the Romas. In short, when it comes to the most discriminated ethnic outgroup in Romania, the gender and parental status of the respondent has no significant effect.

Figure 3. Means test: gender of respondent conditioned on parental status (95% CI)

We further explore this non-effect in Fig. 4. We look at the differences in approval gap between female and male respondents for daughters versus sons, focusing again on Moldovans versus Romas. The results show a striking pattern. When it comes to Moldovans, the daughter–son gap is much larger for female respondents than for their male counterparts. While the vast majority of Moldovans are ethnic Romanians and speak a similar vernacular, the legal mechanism may still be a concern—perhaps even more so for an outgroup that is “most similar” to the ingroup.

Figure 4. Means test: daughter versus son difference subject to respondent gender (95% CI)

Conversely, since Romas are subject to extreme discrimination and often associated with poverty, this is more likely to generate a protection element. Indeed, the daughter–son gap is much larger for male respondents. Simply put, male respondents are less inclined to agree to their daughters marrying a Roma—whether it is motivated by instrumental or identity-based reasons.Footnote ⁴ This finding, however, highlights not just differences between female and male respondents but also reflects the importance of the ethnic identity of the outgroup, which shapes mothers’ and fathers’ concerns for a child’s intermarriage differently.

Considering the ethnicity of respondent

From Table 3, we also see that the respondent’s ethnicity matters for outgroup trust. This is not surprising given that our questions tapped into interethnic relations generally and marriages to an outgroup specifically. The measure is dichotomous: it takes on a value of 1 if the respondent is ethnic Hungarian. About 2.5% of the respondents in the survey identified as Hungarian; these individuals opted to take the survey in the Hungarian language as well. Note that Hungarian society—like its Romanian counterpart—is also quite patriarchal. In the Hungarian language, for example, when a man marries, he “takes” on a wife; in contrast, when a woman marries, she “enters the house” of the man. Likewise, when we look at World Values Survey, Hungarian respondents are also inclined to agree that men make better political leaders (31.4%).

At first glance, it seems ethnic Hungarians are less likely than ethnic Romanians to be supportive of intermarriage. The coefficient for Moldovans (β = −2.36; SE = .47)—a proxy for coethnic Romanians—is negative. Likewise, the coefficients for Chinese (β = −1.95; SE = .58) and Romas (β = −2.63; SE = .81) are negative. Not surprisingly, the coefficient for Hungarians is positive (2.39; SE = 1.10).

The Romanian–Hungarian difference may not be about ethnic identity per se. As previously discussed, language can structure gendered thoughts and preferences (Liu et al. Reference Liu, Shair-Rosenfield, Vance and Csata2018). To isolate the effects of language, we contrast individuals who speak a heavily gendered language (Romanian) from their conationals who speak a genderless one (Hungarian). To consider this possibility, we examine the interaction between ethnicity and gender of the child. Specifically, we look at the differences of getting (1) a daughter treatment versus the control and (2) a son treatment versus the control given the respondent’s ethnic identity. If the gendered structure of language can exacerbate gender attitudes, we should see significant effects for ethnic Romanians for daughters but not for their Hungarian counterparts. We focus on the Chinese and Romas given their undoubted outgroup status for both ethnic Romanians and Hungarians.

As per theoretical expectations, Romanians are significantly less accepting of intermarriage when primed explicitly with “daughter” compared to those in the control group (see Fig. 5). We see this for marriage to a Chinese (−.41) and a Roma (−.27). Conversely, there was no significant difference between the responses for “daughter” versus “child” for Hungarians. We find an inverse but similar effect where respondents are more receptive to intermarriage when asked about a “son” rather than a “child.” But again, this finding only occurs for Romanians. On the one hand, it is possible that these results are byproducts of an underpowered Hungarian analysis—as evidenced by the larger standard errors. On the other hand, the point estimates for the Hungarians are not simply larger. In fact, when it comes to the daughter versus child difference for the Chinese, the effect is positive for Hungarians but negative for Romanians. While this calls for more analysis in future work, it suggests that differences in the gendered structure of languages can shape whether an ethnic group is more or less accepting of intermarriage.

Figure 5. Means test: gender of child in consideration subject to respondent’s ethnicity (95% CI)