Introduction

After a long period of neglecting gender in migration and integration research (Pessar & Mahler, 2003; Suarez-Orozco & Qin, 2006), there is now a growing number of quantitative studies on labor market integration, educational success, and other structural outcomes of migrant women as well as the transmission of gender roles (Soehl et al., 2012; Bayrakdar & Guveli, 2021; Röder, 2014). In addition, qualitative studies have focused particularly on issues of gender equality, emancipation, and intersectionality (Billson, 1995; Das Gupta, 1997; Killian, 2006; Amelina & Lutz, 2019). Both perspectives repeatedly refer directly (Billson, 1995; Suarez-Orozco & Qin, 2006) or indirectly (Dahl et al., 2020; Nandi & Platt, 2020) to the keepers-of-the-culture-hypothesis: In migrant families, mothers are considered to be particularly responsible for passing on ethnic identities and cultural practices of the home country, and daughters, in particular, are raised to maintain the culture of origin.Footnote 1

Despite the relative prominence of the keepers-of-the-culture-hypothesis (hereafter referred to as KOTCH), there is at best ambiguous empirical evidence. It is often not only confirmed (Billson, 1995; Casey & Dustmann, 2010; Mondal et al., 2020) but also rejected (Umaña-Taylor et al., 2013; Platt, 2014), and even opposite patterns can be found (Estrada, 1993; Güngör & Bornstein, 2009). One reason for this could be that the KOTCH has only been tested in fragments and is often only addressed as a side result. Many qualitative studies take into account the perspective of the women solely (Billson, 1995; Salih, 2000; Killian, 2006). This is crucial because there is a risk of drawing wrong conclusions when fathers’ contributions and sons’ developments remain unconsidered (Platt & Polavieja, 2016). On the other hand, research often focused on gendered transmission without taking children’s development into account (Sabatier, 2008; Huijnk et al., 2012; Platt & Polavieja, 2016), or referred only to a specific dimension of transmission (Idema and Phalet 2007; Paterno & Gabrielli, 2014).

In this paper, I aim to overcome these deficits and will examine the KOTCH in detail. I draw on the longitudinal CILS4EU data, which provides survey information from adolescents and their parents in four European countries. Particularly, I test whether the ethnic and mainstream identities of second-generation sons and daughters develop differently over time, whether both developments are interrelated differently for both genders, and whether the transmission of identity differs between parent-child dyads. Ethnic and mainstream identification in this context refers to feelings of belonging to the group of origin and the majority society of the host country and the respective values, traditions, and cultural practices of these groups (Schwartz et al., 2006). The data provides useful insights into the identity developments of second-generation migrants, as longitudinal information is available for the age of 14 to 16, a phase in which a confrontation with one’s origin is likely (Phinney, 1993). Contrary to the studies mentioned above, I consider all aspects of the KOTCH within one comprehensive model by using bivariate conditional growth curve models in a multigroup comparison (Preacher et al., 2008). Even though parts of this method have already been used occasionally to study migrants’ identities and acculturation (Martinovic & Verkuyten, 2012; Umaña-Taylor et al., 2013; Schachner et al., 2018), to my knowledge, growth curve models and their manifold applications have not yet been used to analyze migrants’ identity development.

Cultural Identity, Ethnic Socialization, and Cultural Transmission

Cultural identity refers to a “sense of solidarity with the ideals of a given cultural group and to the attitudes, beliefs, and behaviors manifested toward one’s own (and other) cultural groups as a result of this solidarity” (Schwartz et al., 2006: 6). A person’s self-chosen degree of group attachment is an indicator of how strongly that person incorporated the respective values and cultural practices, such as traditions and customs, into his or her own cultural identity (ibid.). Since migrants and their descendants have contact with both the group of origin and the group of the majority society, ethnic and mainstream identity are particularly relevant dimensions within cultural identity (ibid.). I refer to both of them in order to test the KOTCH. They allow to cover a wide range of culture-specific ways of thinking and behaving, as the feelings of belonging to certain cultural groups like the families’ ethnic heritage or the society one lives in are connected to the certain feelings, thoughts, and behaviors shared in these groups (Hogg et al. 1995).

While national and ethnic identity is mostly congruent for the majority and of no importance, the distinction between the sense of belonging to the group of origin and the larger host society is a relevant part of the acculturation process for immigrants and their descendants. It affects outcomes, such as life satisfaction, self-esteem, delinquent behavior, or school performance (Lee, 2019; Paterno & Gabrielli, 2014; Berry et al., 2006; Phinney et al., 2001). Ethnic socialization is the part of identity development in which descendants of migrants actively negotiate their belonging to the mainstream society and the ethnic group (Phinney, 1993). It takes place primarily in adolescence (Schwartz et al., 2018) and contributes to the formation of a secure self-concept that helps to locate and orient in the given society (Akerlof & Kranton, 2000). Adolescents do not develop identity in isolation, but within their embedding in various social contexts. Thus, cultural transmission plays a major role. In terms of the ethnic identity development of second-generation migrants, the transmission within families is of particular importance, as the family is mainly responsible for preserving the cultural origin and its inherent cultural knowledge (Schachner et al., 2017). During ethnic socialization, the relationship between adolescents and their parents changes towards more symmetry and less dependence, while youths orient themselves more towards their peers and have more contact with various role-mediating institutions of secondary socialization, such as school (ibid.). However, the formation of ethnic and mainstream identity is more complex, as it depends on more than transmission processes. Especially the second generation is confronted with attributions by members of the majority or the minority that can ascribe or deny membership in one group or the other (Katartzi, 2018). These barriers and attributions affect immigrant groups to varying degrees. For example, non-white or non-Western migrants and those with greater cultural differences compared to Europe or the USA may be more exposed to discrimination (Schwartz et al., 2018), which can weaken identification with the majority society (“rejection-disidentification,” Jasinskaja-Lahti et al., 2009) and strengthen identification with the group of origin (“rejection-identification,” Branscombe et al., 1999). The migration policies of the receiving countries (e.g., multiculturalist vs. assimilationist) may have a comparable influence on identity formation (Phinney et al., 2001). Consequently, ethnic and mainstream identity can be more easily pronounced simultaneously for some groups, e.g., those without visible characteristics, than for other groups, where both identities are negatively correlated (Schwartz et al., 2018).

State of Research

Billson (1995) laid the groundwork for the KOTCH by providing insight into the lives of ethnic minority women in Canada, some of whom are caught between cultures and have to renegotiate affiliations. Within her field studies, she portrays women as the central agents of transmission. Similarly, other qualitative and ethnographic studies explicitly examine migrant or ethnic minority women and highlight their role as keepers of the culture. In these studies, the transmission of identity or individual cultural elements of ethnicity, such as the cuisine of the country of origin (D’Sylva & Beagan, 2011) or gender role attitudes (Talbani & Hasanali, 2000), is identified as an ambivalent action, since cultural knowledge and the tasks of family work offer not only opportunities for empowerment but also risks of reproducing inequalities and oppression. Although such research reveals hidden and ambivalent mechanisms, their explicitly female perspective does not yet tell much about the KOTCH. A one-sided view of women lacks the contribution of fathers and the development of sons as a comparative framework (Platt & Polavieja, 2016). However, in studies that consider both genders, the results are inconsistent, questioning the existence of the keepers-of-the-culture-phenomenon. Using a selection of different studies summarized in Table 1, I show that this is (a) due to the lack of comprehensive testing of all aspects of the hypothesis and (b) due to the lack of comparability of the studies.

Table 1 Systematic review of studies that examine gendered identity development
Full size table

Lack of Comprehensive Testing

To comprehensively test the KOTCH, it is necessary to examine both identity dimensions (EI and MI, Table 1), their developments over time (LT), transmission through parents (TR), and their mutual influence simultaneously and separately by gender (INT) of children (CI) and parents (PI/DY). There are no studies that meet all of these criteria indicating a great need for research.

Most studies do not have the KOTCH as the primary object of analysis but find gender differences as side outcomes. They usually focus only on children’s ethnic (e.g., Martinez & Dukes, 1997; Mondal et al. 2020) or mainstream identity (Paterno & Gabrielli, 2014). Some include both dimensions but do not consider parental identity transmission (e.g., Buriel & Cardoza, 1993; Platt, 2014). If the parent’s identity or ethnic socialization within the family is considered (e.g., Rumbaut, 1994; Schachner et al. 2018), gendered transmission (Sabatier, 2008) or even dyadic perspectives (e.g., Idema & Phalet; 2007; Casey & Dustmann, 2010) are the exceptions. The relationship between both identity dimensions is also rarely tested gender-specific (e.g., Umaña-Taylor et al., 2013). Additionally, longitudinal studies are the exception and miss one or more of the requirements, mentioned above (e.g., Casey & Dustmann, 2010; Schüller, 2015). Accordingly, previous research only ever depicts a more or less extensive part of KOTCH, so that a comprehensive and differentiated assessment of its existence or its differentiated expressions is hardly possible. In contrast, I will comprehensively test the KOTCH in this analysis and consider all previously mentioned criteria.

Lack of Comparability of the Studies

In addition to the different foci on the KOTCH, there are also variations in content and methodology which make it difficult to draw conclusions about the existence of the keepers-of-the-culture-phenomenon. A major difference within the literature is the selection and measurement of identity concepts. These range from universal measures intended to capture cultural identity in its entirety to specific cultural aspects that test only a particular part of the KOTCH. The latter includes gender roles (Idema & Phalet, 2007), the division of labor tied to them (Platt & Polavieja, 2016), and language use (Güngör & Bornstein, 2009). Universal measurement is either based on various multidimensional scales (e.g., Martinez & Dukes, 1997), the sense of belonging to an ethnic group or host society (e.g., Schneider et al., 2012), or on respondents’ self-labelling (e.g., Estrada, 1993), such as “American,” “Mexican American,” “Chicano,” or “Hispanic.” The use of labels allows an effective testing of gendered transmission by comparing the label choices of the children with those of the parents. However, the exact extent of transmission, the mutual relations between cultural identities, and their individual developments cannot be traced in detail. Most studies that use labels as indicators and thereby support the KOTCH in their respective tested parts were conducted in the USA (e.g., de Snyder et al., 1982; Buriel & Cardoza, 1993). This complicates comparisons with results that rely on scales (tending to affirm the KOTCH, e.g., Sabatier, 2008; Huijnk et al., 2012) or used feelings of belonging as measurement (tending to reject the KOTCH, e.g., Schneider et al., 2012; Schüller, 2015). These two operationalizations have been used primarily in Europe, where the countries of origin, the reception context, and historical migration paths differ substantially compared to North America. No clear patterns for or against the KOTCH can be identified between both regions, finding both affirmation and rejection in each (Dion & Dion, 2004; Umaña-Taylor et al., 2013; Casey & Dustmann, 2010; Platt, 2014). Moreover, even within the relatively homogeneous group of Mexican migrants and their descendants in the USA, the results are inconclusive (e.g., Buriel & Cardoza, 1993; Umaña-Taylor et al., 2013).

Depending on context and region, ethnic identities blur more or less with other identity-related categories. In the USA, these are primarily race (Martinez & Dukes, 1997) or pan-ethnic identities of non-immigrant minorities (like Hispanics, Rumbaut, 1994). Within an immigrant group, the interplay of these categories may also vary by community size, phenotypic visibility and the associated discrimination, and other contextual factors (Rumbaut, 1994; Schwartz et al., 2006). In addition to the origin-perception constellation, results of the KOTCH also may differ by generational status. As Umaña-Taylor et al. (2013) point out, the transmission processes between the 1st and 2nd generation and the 2nd and 3rd generation follow different patterns. The scarce research on the 2.5th generation also shows that their integration experiences differ from those of families with two foreign-born parents (Kalmijn, 2010; Emonds & van Tubergen, 2015). This also holds for the KOTCH, which is largely confirmed when the 2.5th generation is involved (e.g., Obućina & Saarela, 2020). Contrary, no consistent pattern exists for the 2nd generation. The interplay with other identity constructs, such as gender or religiosity, as well as other group affiliations (local or within a migrant community) is also a question that needs to be addressed (Jacobson, 1997; Schneider et al., 2012), but is difficult to disentangle empirically. Whether or not the KOTCH is confirmed may also depend on the respondents’ age. Ethnic socialization proceeds differently at different ages, as does parental influence (Phinney, 1993; Schachner et al., 2017). However, a pattern as to when gender-specific transmission begins or fades cannot be deduced from the existing literature.

Accordingly, an examination of the existing literature reveals, on the one hand, the need for research that is explicitly and extensively dedicated to the keepers-of-the culture-phenomenon. On the other hand, it seems that gender-specific identity development and transmission is not necessarily an automatically operating process within migrant families, as there is evidence both for and against the keepers-of-the-culture-phenomenon. It depends on macro-structural conditions such as the reception context, can be entangled with other identity constructs, and is difficult to capture empirically.

Theoretical Framework and Hypotheses

The KOTCH states that in migrant families, mothers are particularly responsible for passing on ethnic identities and cultural practices of the home country and that daughters are raised to maintain these (Suarez-Orozco & Qin, 2006). In its extended form, it further implies that women identify less with mainstream society and that the associated sense of belonging is more likely to be passed on by fathers. The hypothesis is supported by two main arguments. Within both, the development of cultural identities (ethnic and mainstream) is inseparably linked to norms of gender, family, and sexuality. They serve to legitimize gendered power relations, to regulate gendered access to resources and society, and are simultaneously a central part of the cultures with which one can identify (Billson, 1995; Bartkowski, 2003; Samuel, 2010). The identification and behavioral orientation towards certain groups, in turn, can reproduce these patterns.

Access to the Public Sphere and the Reproduction of Gendered Responsibilities

Most societies are male-dominated, even though the extent varies greatly between cultures. The direct and indirect control over women operates in two complementary ways (Bourdieu, 2001; Wischermann & Mueller, 2004). First, through the regulation of access to the public sphere and the power associated with it. Second, through a normative framework of values and expectations that binds women to the private sphere, which involves responsibility for raising children, keeping traditions, and doing domestic work. In consequence, women spend more time on parenting, and socializing children. In the context of migration, however, this explanation falls short.

The values and traditions the migrants brought with them can serve as orientation in the new environment and at the same time be perceived as particularly threatened (Das Gupta, 1997). When migrants want to transmit their culture to the next generation, they take care that their roots are not lost (Das Gupta, 1997; Renzaho et al., 2011). Many migrants come from countries where, compared to the country of residence, a more traditional image of women prevails, where women’s access to the public sphere is limited, and duties of domestic care and cultural preservation in the private sphere are more strongly expected (Röder, 2014). Although migrants develop more egalitarian gender role attitudes over time (Röder & Mühlau, 2014), these differences may be a reason why the transmission of ethnic identity from first- to second-generation migrants flows mainly from mothers to daughters. This pattern is supported by religious beliefs, which are more pronounced among migrants compared to mainstream societies and which are more conservative concerning gender equality (Röder, 2014). Furthermore, the pattern seems embedded in homolinear transmission preferences (Obućina & Saarela, 2020; Platt & Polavieya, 2016). This means that transmission is more strongly from fathers to sons and from mothers to daughters. If the mother already has the role of the keeper of the culture and is more strongly associated with the country of origin, these preferences could also determine the stronger transmission to the daughter. Especially when parents stem from former colonial regions, another normative obligation resonates: Postcolonial nationalist movements oftentimes reinforced women’s role as keepers of the culture, thus interweaving gender inequality with a cultural heritage that must be defended against Western or colonialist values, even after migration (Coomaraswamy, 2002).

The strong interweaving of gender with religious and ethnic identity is reflected in the realities of families’ lives. As boys “do not have to conform to the ‘ideal’ ethnic subject” (Suarez-Orozco & Qin, 2006: 171), they are less controlled than daughters across “nearly every ethnic background as well as across different historical periods” (ibid., see also for a review). In consequence, daughters often participate less in activities outside the family, like meeting peers or attending school programs (ibid.; Soergel, 2017). These restrictions are also linked to cultural norms of sexuality and family so that interactions with the opposite sex are particularly tightly under surveillance (Carol, 2016). This may also lead to concrete expectations regarding the choice of partner. While sons are more likely to be allowed to enter interethnic relationships and marriages, daughters are often expected to marry a man from their own culture (Das Gupta, 1997; Lucassen & Laarman, 2009). These patterns are a consequence, source, and indicator of the expectations that daughters act as keepers of the culture (Samuel, 2010).

Daughters who try to emancipate themselves from these expectations can become “caught between cultures” (Talbani & Hasanali, 2000; Dahl et al., 2020), which not only leads to stress and conflicts but often also results in a compromise-based double burden (Noh et al., 1992; Billson, 1995). They move closer to mainstream society and become more independent by having their place in the public sphere (for example, by participating in the labor market), but at the same time, they still have to be responsible for cultural preservation within their ethnic group, including domestic work, raising children, and keeping traditions (Billson, 1995). Thus, emancipation is not only gradual but initially rather one-sided towards mainstream identity. This pattern is partly supported by 1st-generation mothers, who also can get caught between cultures. It would be too short-sighted to claim that they pass on the self-experienced inequality unfiltered. Mothers are aware of their opportunity to change power relations through migration and teach more egalitarian values to their children (Billson, 1995). However, their desire to pass on ethnic knowledge and to bring the culture of origin closer to the children is partly opposed to these emancipation efforts. The main problem is that preserving the culture to which these women are committed to some degree often involves preserving traditions that keep women in their lower positions (ibid.).

Anticipated Gendered Experience of Ethnicity

While daughters may get caught between cultures because they want to abandon traditions but are not allowed to, sons may get caught in between cultures because the mainstream society sanctions them if they adhere to their culture of origin. Mainstream society evaluates ethnic identities gender-specific (Suarez-Orozco & Qin, 2006; Güngör & Bornstein, 2009). While female ethnicity is perceived as something exotic that women should be proud of, male ethnicity is often perceived as something threatening and regressive (Hughes et al., 2006) with the consequence of facing more ethnic discrimination (Güngör & Bornstein, 2009). Parents may anticipate this gendered perception and its consequences more or less consciously (Thomas & Speight, 1999; Hughes et al., 2006). They may raise sons to be more sensitive to discrimination, externalize their ethnic identity less, and adapt more towards the mainstream society. Otherwise, sons may cope with perceived rejection expressed by the majority by strengthening their feeling of ethnic belonging (Branscombe et al., 1999) and weakening their identification with the mainstream society (Jasinskaja-Lahti et al., 2009). Sons compared to daughters should therefore find it more difficult to develop dual identities (Killian, 2006).

The two main arguments presented here support the KOTCH. To examine its empirical truth in detail, this paper differentiates the hypothesis into several sub-hypotheses. From a dyadic perspective, they imply that the father-son dyad and the mother-daughter dyad should be most different from each other:

H1: (a) Girls’ ethnic identity is higher and develops stronger over time, while (b) boys’ mainstream identity is higher and develops stronger over time.

H2: Boys are less able to combine both ethnic and mainstream identities than girls.

H3: Concerning the dyad-specific transmission of identities, (a) mothers have a stronger influence on their children’s identity and (b) daughter’s identities are more strongly influenced by their parents.

Data, Methods, and Analyses

Data

I use data from three waves of the Children of Immigrants Longitudinal Survey in Four European Countries (CILS4EU, Kalter et al., 2017). The children of immigrants and natives living in England, Germany, Sweden, and the Netherlands were randomly selected via a multi-stage procedure with an oversampling of schools with a high proportion of migrants. In addition to the interviews with mostly 14-year-old adolescents of an entire school class, the study contains information from teachers, parents, and social networks. For the analyses, the Dutch sample had to be excluded due to missing information on ethnic identity in the third wave. Furthermore, I consider only adolescents who belong to the second or 1.75th migrant generation. The sample includes only youth with both parents not born in the country of residence. The youth also were either no more than 6 years old when they migrated with their parents or are born in the country of residence.

Operationalization

Measuring cultural identity is a major challenge in quantitative research. Theoretical contributions and qualitative studies remark that ethnic identities can often only be inadequately captured in surveys because they are multidimensional, time-changing, and cannot always be categorized clearly (Burton et al., 2010; Leszczensky & Gräbs Santiago, 2015). The measurement in the CILS4EU data can meet many of the requirements as it includes multiple time points and two independent dimensions of cultural identity (ethnic and mainstream). Respondents were first asked about their identity of the country of residence before they were allowed to openly state their affiliation to other ethnic or national groups below a small pre-selection. This allowed respondents to self-categorize, thus avoiding external attributions that force respondents into certain categories. If they assigned themselves to another group, they could indicate their identification with this ethnic group on a 4-point scale, as they had already done with their identification with the mainstream society. While these identities were asked for the children in three consecutive years, the identification of the parent, as well as all other measurements at the parental level, is only available in the first wave. I use this information to estimate the transmission of identities from the participating parent to his or her children. The participating parent was not chosen at random but was the one most likely to be responsible for the child’s school matters.Footnote 2 For the respective other parent, data on identification are not available.

The parent-child dyads for the multigroup comparison include the four combinations father-son, father-daughter, mother-son, and mother-daughter. Parental education was reduced to two categories: low if the parent has a degree below upper secondary education and high if the parent has at least upper secondary education. I further control for whether respondents have a Muslim background by dichotomizing the religious affiliation of the surveyed parent (Muslim vs. other/none), as Islam reflects the main boundary in Western European integration debates (Foner & Alba, 2008). In the context of a “cultural racism” (ibid.), the identity development of Muslim migrants is affected by a particularly critical perception from mainstream society (Jasinskaja-Lahti et al., 2009; Statham, 2016; Yazdiha, 2019). Furthermore, Muslim migrants hold on to values and traditions of the origin culture longer compared to other groups (Röder & Mühlau, 2014) and their religious and ethnic identity are strongly intertwined (Martinovic & Verkuyten, 2012; Phalet et al., 2018). Controlling for countries of residence (England, Germany, and Sweden) serves as a proxy for macro-structural differences in policy, reception climate, and selective immigration from different countries of origin.

Methods

To test the hypotheses, two identity developments (ethnic and mainstream) across time and the parental influence on them must be compared between four dyads simultaneously. Bivariate, conditional latent growth curve models in a multigroup comparison, a special type of structural equation modeling, allow such analyses (Preacher et al., 2008; Curran et al., 2010). Latent growth curve models are an advantageous and powerful way of carrying out longitudinal analyses. They focus on the analysis of developments by considering the measurement points of a variable as imperfect representatives of a latent start value (intercept) at the beginning of the observation and a latent change value (slope) over time (see Fig. 1a). They are “imperfect representatives” in the sense that this procedure, unlike other methods, takes errors of measurement into account for each point in time (e1 to e3). Another special feature is that the correlation between intercept and slope can be taken into account to control for ceiling effects and, due to the variance-covariance based estimation, this method handles missing values (MCAR and MAR) more effectively via full information maximum likelihood estimations (Enders & Bandalos, 2001). The level of the latent intercept at time t1 is estimated by fixing its influence on all observed variables to 1, while the influence of the latent slope is linearly extrapolated from 0 to represent a linear change over time. This method can take into account two developments simultaneously and in conjunction with each other (see Fig. 1b).

Fig. 1
figure 1

Stepwise development of a complex latent growth curve model. Note: y1–y3 and z1–z3, manifest variables; e1–e6, measurement errors; Intercept and Slope, latent variables; X1–X4, explanatory variables; arrows, direct effects; double arrows, covariances; 0–2, fixed values to represent linear models

Full size image

The bivariate growth curve model includes the development of ethnic and mainstream identity. The latent intercepts and slopes indicate the average start and change parameters for the identity developments under observation. However, they are not constant but have a variance that can be explained by other variables. Such explanatory variables are foremost the parental identities as well as the control variables mentioned before. Growth curve models including such variables are called “conditional” (see Fig. 1c).

In the present analysis, not only two developments (mainstream and ethnic identity) and their determinants have to be estimated simultaneously but also gender-specific characteristics of identity developments as well as dyad-specific transmissions. Structural equation modeling allows comparisons of complete models across such dyadic groups with all parameters at the same time. In these multigroup comparisons, I first estimate the four dyadic models independently and then gradually equate the different types of parameters between the groups. If such a fixation leads to a significantly worse representation of the empirical data by the more restricted model compared to the model with free estimations, gender-specific developments or a dyad-specific transmission are revealed.

I fix the parameters between the four dyads in the following order (Preacher et al., 2008):

  1. (1)

    Means: The mean values of the latent variables, to test gender-specific identity development.

  2. (2)

    Covariances: The correlations between slopes and intercepts as well as between manifest variables.

  3. (3)

    Regression coefficients: All regression coefficients, to examine the dyad-specific influence of parental identities on children’s identity development.

  4. (4)

    Intercepts: The mean values of the parental identity and the confounding variables.

Robustness Checks

Due to unequal categorizations between the survey countries, the information divides the countries of origin only into rough regions. These are Eastern Europe, Southern Europe, Africa, South Asia, Other Asia, and Other. The latter sums up Latin America and the Caribbean, North America, Other Europe, and Unknown Origin. Taking these groups into account produced the same results in the analysis and provided no further insights. To avoid unnecessary complexity, the models are set up without the information of origin. Furthermore, I repeated the analyses only for those families in which parents hold traditional gender role attitudes.Footnote 3 With the restriction that due to the lower number of cases, the father-son dyad could no longer be taken into account, no gender differences in development and transmission were found for the remaining three dyads. Additionally, I also estimated equal models with another measurement of ethnic identity: “How important is it for you personally to maintain the customs and traditions of this group?” The results did not differ from the identity developments with the previous measurement, which suggests that the results are sufficiently robust. Approximately 50% of the youth in the sample did not participate in all three waves. To control for selective panel attrition, I repeated the analyses only for those for whom both identity measures are available in all three waves. There appears to be a slightly selective dropout of those with lower mainstream identity starting level in Sweden, which translates into a higher country effect on the respective Intercept. However, all other coefficients and the multigroup comparison are equal.

Results

Sample Description

Table 2 plots the mean values of ethnic and mainstream identity separately by gender for different characteristics and gives the correlations with the respective parental identities. The mean values of ethnic identity remain constantly high across time and do not differ between genders. This pattern is independent of Muslim family affiliation, parental education, and country of residence. However, for girls, there is a slightly increasing correlation with parental identity over time, especially if the family is Muslim and lives in Sweden. Despite relatively small sample size in England, there is a particularly negative correlation of parental ethnic identity with boys’ ethnic identity.

Table 2 Means and correlations by children’s gender
Full size table

Compared to ethnic identity, mainstream identity is less developed in both gender but increases over time. Boys start with lower scores but develop more strongly and catch up with the girls over time. Subgroups show a similar trend, although differences between Muslim and non-Muslim families and parental education must be taken into account. There are also notable differences between countries of residence. In Germany, the mainstream identity of the second generation is particularly low across both genders compared to England and Sweden. The correlation with parental identity is higher for mainstream identity than for ethnic identity. However, clear transmission patterns exist only among girls. For boys, in some subgroups, as well as in the overall group in year 3, we do not find clear transmission of mainstream identity.

Development of the Bivariate Conditional Growth Curve Model

The bivariate results indicated that although there are differences in ethnic identity levels between subgroups, they do not develop over time and are not influenced by external factors. However, mainstream identity seems to increase over time. Accordingly, I first examine separately for both identity development whether the consideration of variable slopes in the growth curve models offers any advantage at all in estimating development. Table 3 illustrates that the inclusion of such a random slope in the univariate ethnic identity model (M2) adds unnecessary complexity as it does not significantly improve the model fit (p(ΔChi2): 0.121) compared to the random-intercept only model (M1). Furthermore, the goodness of fit indicated by the CFI and RMSEA remains very good without considering the random slope in the estimation of the ethnic identity across the three consecutive measurements. However, the inclusion of the random slope (M4) improves the fit for modeling mainstream identity development (p(ΔChi2): 0.000), thus becoming essential for a good fit to the given mainstream identity data.

Table 3 Latent growth curve model fits
Full size table

According to these results, the bivariate unconditional growth curve model (M5), in which both identities are taken into account at the same time, is set up without a random slope for the ethnic identity estimation. The model fit is still excellent with a CFI of 0.998 and a RMSEA of 0.012. With the stepwise addition of the time-invariant variables (M6 to M9), this fit does not become significantly worse, which is why the model configuration M9 is the starting point for the multigroup analysis. Table 5 in the appendix contains the detailed results of this bivariate, conditional, latent growth curve model, which does not yet distinguish between dyads.

Results of the Multigroup Comparison

Models M1a to M1d in Table 4 depict the model characteristics when each dyad is estimated separately. The number of dyads in which fathers provided information (M1a and M1b) is substantially lower. Because of this, the very good fits in M1a and M1b should be interpreted cautiously as they can be susceptible to distortion (Chen, 2007). A closer look at the detailed results at least did not reveal any problems, for example, regarding implausible estimates. For the two subgroups in which mothers are represented, the theoretical model reflects the empirical data very well (M1c and M1d). Next, I capture all dyads at once, whereby all parameters are still freely estimated (M2). This unconstrained estimation also produces very good fits and is the starting point for the stepwise parameter fixing across the four dyads.

Table 4 Conditional latent growth curve model: dyad multigroup fits and comparisons
Full size table

First, I equate latent means across all four groups in M3. They describe the gender-specific developments of identity as intercepts and slopes. Fixing these parameters does significantly decrease the fit (M3), which implies that the initial levels and the development of ethnic and mainstream identity do not differ between sons and daughters. The result does not support hypothesis 1. Next, I equate the individual covariances across the four groups (M4). This concerns the covariance between the intercepts of ethnic and mainstream identity, between the intercept and slope of mainstream identity, and between the two parental identity responses. Again, this restriction does not lead to a significant deterioration of the fit, so this simplification can be retained. It highlights that the relationship between the development of ethnic identity and mainstream identity is independent of gender. Both in the comparison of mothers and fathers as well as in the comparison of daughters and sons. The results speak against hypothesis 2, in which I assumed a more negatively related development for sons compared to daughters. In the third step, I fix all regression coefficients between the four dyads (M5). Of particular interest is the influence of parental identity on children’s identity development. Here too, there is no deterioration in the model fit compared to the previous model, so that hypotheses 3a and 3b are also not supported. The parental influence does not emanate specifically from the mother and directs not particularly strongly towards daughters, but is of equal size in all four groups. Finally, the intercepts of the manifest variables are fixed. Here, for the first time, a significantly worse fit emerges, which on closer examination is due to different scores for parental mainstream identity and different proportions of belonging to Muslim families between fathers and mothers. In M6b, I fixed these two parameters across two maternal and paternal dyads separately, while parental education and ethnic identity remain equated between all four groups.

Figure 2 shows the results of the multigroup comparison after fixing the parameters as described in M6b. For reasons of clarity, I only present significant values. Table 6 in the appendix gives an overview over all coefficients and standard errors. The results illustrate once again that, contrary to the KOTCH, the developments between sons and daughters are equal. For both genders, ethnic identification is relatively pronounced from the beginning (3.869***) and does not change over time, while identification with the majority society is lower at the start of the observation (2.655***) but increases over the following years (0.385*** per year). The negative covariance between Intercept and Slope on mainstream identity development indicates ceiling effects, whereby young people with high starting values make fewer gains over time. The negative covariance between the two starting values of the cultural identities shows that a higher basic level of ethnic identity is accompanied by a lower level of mainstream identity and vice versa. The fact that this incompatibility does not only exist among the sons speaks against the KOTCH (H2). The strength of this correlation is about the same as for parents. Fathers and mothers transmit their identities to both daughters and sons to the same extent, although fathers report a more pronounced mainstream identity. This also contradicts the KOTCH (H3). The transmission of mainstream identity is twice as large as that of ethnic identity (0.251*** to 0.105***), regardless of the dyad constellation. The ethnic identity of the parents is more pronounced compared to the children, suggesting that there is intergenerational dissimilation of ethnicity. However, children have a lower mainstream identity than their parents at the beginning and only catch up over time, which can be interpreted as intragenerational assimilation.

Fig. 2
figure 2

Results of the multigroup comparison, only significant and unstandardized coefficients. Notes: (a) coefficients for father [mother], both significant at p < 0.001; latent growth curve model with FIML estimation. Source: CILS4EU

Full size image

While a high level of education goes along with a lower ethnic orientation for the parents, belonging to Islam means greater ethnic identification with a simultaneously lower identification with the mainstream society. Compared to the parents, however, living in a Muslim family has far less effect on children’s mainstream identification, indicating that intergenerational adaptation is more extensive in these families. There are also differences between the countries of residence. Compared to Germany, mainstream identity is significantly higher among children and their parents in England and Sweden. However, the place of residence does not affect the ethnic identity of the parents, although there are further differences among the children. The ethnic identity of the children is more pronounced in England and somewhat lower in Sweden compared to Germany. The place of residence also correlates with the level of parental education and the chances that the family belongs to Islam.

Discussion

This paper reviewed the keepers-of-the-culture-hypothesis (KOTCH), which migration research frequently uses to explain gendered acculturation (Billson, 1995; Suarez-Orozco & Qin, 2006). It states that mothers are primarily responsible for ethnic socialization and that this is mainly directed towards the daughters. Mainstream identification, on the other hand, is supposed to be more pronounced in fathers and sons, whereby it should be more difficult for the latter to reconcile both identities. However, the three sub-hypotheses tested in this context had to be rejected, indicating that subsequent gendered outcomes in acculturation are not grounded in gendered ethnic and mainstream identification and transmission during youth.

First, daughters do not have a more developed ethnic identity and sons do not have a more developed mainstream identity. Instead, the multigroup comparisons between both genders show similar developments across both dimensions. Ethnic identity is already relatively high at the beginning and does not change over time, while the mainstream identity has a lower level but increases every year. Secondly, the extent to which the two identities are incompatible is also equal for both genders and relatively similar to parents, who also do not differ along gender. Thirdly, the transmission of parental identities between the four parent-child dyads is of the same extent. That is, fathers and mothers transmit ethnic and mainstream identity equally to sons and daughters. Independent of the dyad, the transmission of the mainstream identity is twice as large as the transmission of the ethnic identity. The comparison between parents and children also reveals intergenerational dissimilation of ethnic identification and intragenerational assimilation of mainstream identification for second-generation youth. The models further indicate that children from Muslim families differ in their development and that there are significant country differences.

The latter highlights the importance of reception contexts (Phinney et al., 2001), which include migration policy and attitudes towards migration. The results on mainstream identity fit into the migration policy histories of the respective countries: Until the time of the survey, Sweden and Great Britain had long been characterized as liberal and multiculturalist (Geddes & Scholten, 2016). Compared to Germany, which did not see itself as a country of immigration, there was, for example, easier access to citizenship for migrant parents and their descendants (ibid.). The higher mainstream identities of both generations in Sweden and England may reflect these inclusive policies. The particularly large effect in England may be due to migrants from the Commonwealth who were more likely to identify as British, possibly even before they migrated. The lower ethnic identity of Swedish youth compared to Germany, on the other hand, does not fit into this pattern. Possible reasons, such as the different size and origin of migrant communities, differences in perceived discrimination, or effects of the school system, need to be explored in further research as with the present model, specific policy strategies, and the reception climate within countries cannot be differentiated. Also, the influence of the policy changes in the observed countries towards more assimilationist strategies a few years before and during the observation period (Joppke & Morawska, 2003) and their short- or long-term effects cannot be disentangled in this cross-sectional perspective.

Accordingly, it should be discussed why the expected gender-specific patterns did not show up, and what this means for the keepers-of-the-culture-narrative. Various reasons can be named, why the KOTCH is sometimes confirmed (Rumbaut, 1994; Casey & Dustmann, 2010; Mondal et al., 2020) and sometimes not (Huijnk et al., 2012; Umaña-Taylor et al., 2013; Platt, 2014). On the one hand, ethnic identities of different migrant groups overlap to varying degrees with other constructs of social identity such as regional identity (Vathi, 2015), racial identity (Martinez & Dukes, 1997), or religious identity (Platt, 2014). Together with the perception and reaction of the majority society, they may contribute to gender-specific transmission logics or be the driver behind gender-specific development. Further research is needed to differentiate these constructs from a gender perspective. On the other hand, it is conceivable that the gender-specific transmission and development becomes apparent with a delay. This does not only mean that these differences appear in later adolescence (Güngör & Bornstein, 2009), for which the data does not provide more information. Gendered transmission may only emerge through certain life course events, such as marriage or starting a family. Especially in the context of parenthood, one has to decide who will pass on which values and traditions to the next generation. Future research needs to clarify whether such life events can trigger the keepers-of-the-culture-phenomenon.

There are further methodological limitations. First, only one parent was interviewed and only at one time point. Since the selection of the parent was not random, certain dyads might not be representative for migrant families. In families where fathers do not have child-rearing responsibilities, they are unlikely to have completed the questionnaire. These fathers have less transmission potential due to the traditional distribution of roles and self-select out from the analysis, which, in turn, could partly suppress the keepers-of-the-culture-effect. In this paper, I attempted to address this issue in a robustness check in which analyses were conducted only for those families exhibiting traditional gender roles. This did not yield any different results. The second limitation concerns the operationalization. The advantage of taking all cultural practices and values into account through one measurement is a disadvantage when it comes to more accurately determining which specific values and practices are transmitted in a gender-specific way and which are not.Footnote 4 Both limitations prevent the testing of whether parents have complementary roles and transmit different aspects of identity (Sabatier, 2008). Furthermore, the sons and daughters in this study do not come from the same families. To retest the hypotheses, a longitudinal dataset that makes opposite-sex siblings comparable within families would be desirable.

Conclusion

The fact that women in migrant families, as keepers of the culture, pass on cultural practices and associated values seems to be anything but a typical, self-evident process that can be taken for granted. On the one hand, the keepers-of-the-culture-role may be more of a parental or familial expectation. Its practical implementation may be blocked by the empowerment of the next generation or by the influence of the majority society. Future studies taking intergenerational conflicts in ethnic socialization into account could further investigate this (see Dahl et al., 2020). On the other hand, various studies show how mothers use their opportunity in the country of residence to challenge patriarchal structures of the country of origin by raising children in a more egalitarian way and integrating their daughters more strongly into the majority society (Billson, 1995; Vathi, 2015). Due to their key educational role, first-generation mothers might therefore pass on ethnic identity and the related values and practices, but they might choose not to do so, for example, if this goes hand in hand with disadvantage for them (Billson, 1995; Idema & Phalet, 2007). In the future, when referring to gender differences in ethnic socialization, this empowered role-making perspective should be taken into account. The findings suggest that in academic, public, and political discourses, women in immigrant families should initially be characterized as “directors of transmission” rather than “keepers of the culture.”