ASSIMILATION CHOICES AMONG IMMIGRANT FAMILIES: DOES SCHOOL CONTEXT MATTER?

Abstract

This paper explores the relationship between social context, measured in terms of school characteristics, and the assimilation of immigrant adolescents. First, it develops a measure of assimilation based on comparing immigrant adolescents to native peers within the same school. Second, it investigates whether immigrant adolescents’ degree of assimilation varies systematically according to school SES. Third, it explores the role of parental and adolescent behavior in creating such variation. Results show that both Asian and Hispanic immigrant youth are less assimilated to native youths’ substance use and delinquency patterns in lower-SES schools. This association can be explained by parenting behaviors and adolescent friendship choices for Asian youth, but not Hispanic youth.

The recent renewed wave of mass immigration to the United States (U.S.) has sparked attempts to re-think theories of immigrant adaptation and assimilation. Many scholars have argued that the experiences of immigrants currently entering the U.S. differ in fundamental ways from the experiences of those who arrived in the early twentieth century. Immigrants’ settlement patterns upon arrival constitute one such difference. Earlier cohorts of immigrants typically settled in central cities, often in ethnic enclaves (Alba and Nee, 2003), and did not disperse into suburban and less ethnically segregated areas until later generations, whereas today’s immigrants often settle directly in the suburbs (Alba et al., 1999). Moreover, economic and social changes have brought about deterioration of many central cities, meaning that immigrants who do settle in these traditional areas may find themselves in economically isolated, highly segregated neighborhoods (Suarez-Orozco and Suarez-Orozco, 1995; Waldinger, 2001). For immigrants’ children, residential location carries additional importance because it typically determines the schools they attend. Those whose families settle in central cities often attend educationally and socially disadvantaged inner-city schools, while their peers in more affluent suburbs also have the advantage of better-funded suburban schools. Thus, there is a great deal of variation and inequality in the types of school and community contexts that immigrants encounter in the U.S. The implications of such diversity in social contexts for immigrant adaptation are not yet fully understood, but many scholars have suggested that adolescents in immigrant families, who must cope simultaneously with developing independent identities and issues of acculturation, may be particularly influenced by the surrounding environment (Hirschman, 2001; Portes and Rumbaut, 2001; Portes and Zhou, 1993; Zhou, 1997). However, few empirical studies have explored whether social context affects patterns of assimilation among adolescent children of immigrants and their families. This paper explores the relationship between school context and the extent to which immigrant families assimilate.

One prominent attempt to describe contemporary immigration experiences is segmented assimilation theory (Portes and Zhou, 1993), which argues that there are many possible pathways of assimilation for immigrants to follow. While classical assimilation theory assumed that immigrant families would eventually settle among and assimilate into the native middle class (Gordon, 1964), segmented assimilation theory identifies this “traditional” type of assimilation as only one possible assimilation trajectory for contemporary immigrant families. This “traditional” assimilation trajectory entails increasing access to educational and economic opportunities as immigrants become incorporated into the American mainstream. Segmented assimilation theory argues that assimilation does not always entail such benefits: An immigrant family assimilating in an impoverished inner-city area may become incorporated into the urban underclass, leading to stagnant or decreasing educational and economic outcomes. Whether assimilation has positive or negative consequences is thus dependent on the particular native group with whom immigrants assimilate, which in turn will be influenced by the immigrant family’s local residential context. Finally, an immigrant family may choose not to assimilate fully. This third possible assimilation trajectory – sometimes referred to as “selective acculturation” (Gibson 1993) – involves deliberate preservation of the immigrant group’s culture and values, accompanied by forms of assimilation necessary for economic integration (Portes and Zhou, 1993; Rumbaut, 1994; Zhou, 1997). The segmented assimilation perspective suggests that this third path may be the most beneficial for immigrants who settle in disadvantaged contexts, as it may allow them to avoid assimilating into the urban underclass.

A recent study by Xie and Greenman (2005) tested segmented assimilation theory’s implication that the outcomes of assimilation may differ by local context. They found no evidence that assimilation had different consequences in low-poverty neighborhoods than in high-poverty neighborhoods. However, this result should not necessarily be interpreted as a rejection of segmented assimilation theory. Such an interpretation would require an unrealistic assumption that assimilation itself is a given condition, exogenous to both expected consequences and the local context. In this paper, I avoid such an assumption by examining assimilation behaviors, local context, and assimilation consequences jointly. My argument draws on not only segmented assimilation theory, but also the literatures on social context, parenting, and peer effects on adolescent development. Together, these literatures suggest that immigrant parents and their children may actively modify their assimilation behaviors in response to local contexts.

Immigrant families that settle in economically disadvantaged areas are likely aware of the dangers that impoverished communities and schools pose for their children. However, given the modest financial means of many immigrant families upon arrival in the U.S., they may find it difficult to avoid settling in such areas. They may instead make efforts to protect their children from assimilating into the surrounding school and community contexts. Framed in terms of the above discussion, this means that if an immigrant family realizes that full assimilation may entail downward mobility, it may try to follow the third assimilation pathway of only limited or partial assimilation.

These insights suggest an alternative interpretation of segmented assimilation theory. Rather than the consequences of full assimilation differing according to local context, the theory can be construed to imply that assimilation behavior differs by local context. Immigrant families in low-SES contexts may have reason to avoid full assimilation. Immigrant families in higher-SES contexts, by contrast, may not be as concerned about the potentially deleterious effects of the surrounding environment on their children. It follows that immigrant families’ choices about whether and how much to assimilate may depend on the local context.

This paper explores the relationship between social context, measured in terms of school characteristics, and the assimilation of adolescents in immigrant families. (For ease of expression, the term “immigrant adolescent” will henceforth be used to refer to any adolescent in an immigrant family (i.e., who lives with at least one immigrant parent), regardless of the adolescent’s birthplace. Differences between foreign-born and U.S.-born youth will, however, be accounted for in the analysis.) The goals of this research are threefold: First, in contrast to previous literature, I develop a measure of assimilation that is explicitly grounded in the local context by comparing the delinquency and substance use behaviors of immigrant adolescents with those of their native peers within the same school. Second, using this measure, I investigate the hypothesis that immigrant adolescents’ degree of assimilation varies systematically according to school socioeconomic status. Third, I explore the potential role of parental and adolescent behavior in creating such variation. I test the hypothesis that both immigrant parents and adolescents adjust their behavior in response to social context so as to diminish the potential negative effects of assimilating into disadvantaged contexts.

Theoretical Motivations and Hypotheses

Assimilation

Sociological studies of the adaptation and incorporation of immigrants into American society have generally been framed in terms of assimilation. Scholars of the great wave of immigration that ended in the 1920s typically viewed assimilation as a process through which immigrants gradually shed the cultures and customs of their home countries and adopted the language, expressive habits, and eventually the spatial distribution and socioeconomic characteristics of “mainstream” Americans – usually defined as white middle-class Protestants (Gordon, 1964). Modern perspectives on assimilation, such as that expounded by Alba and Nee (1997, 2003), have refined the idea of assimilation to recognize that immigrant groups can also influence mainstream society. In a definition that remains neutral about direction of the influence, Alba and Nee describe assimilation as “the decline, and at its endpoint the disappearance, of an ethnic/racial distinction and the cultural and social differences that express it” (1997, p. 863). The critical aspect of assimilation in this definition, however, is still the decline of ethnic distinction – that is, a process by which two distinct groups become more similar to each other. This idea is at the heart of the concept of assimilation.

Most recent empirical research on assimilation has focused on change over time or across generations in immigrants’ language usage (Alba and Nee, 2003; Portes and Rumbaut, 2001), labor market outcomes (Borjas, 1985; Chiswick, 1978; Schoeni, 1997) residential patterns (Alba et al., 1999; Alba, Logan and Stultz, 2000; Alba and Nee, 2003; Iceland and Scopilliti, 2008), friendship or marriage preferences (Quillian and Campbell, 2003), or on the consequences of such changes for educational, economic, or health outcomes (Greenman and Xie, 2008; Harker, 2001; Harris, 1999; Mouw and Xie, 1999; Portes and Hao, 2002; Rumbaut, 1997; many others). While these investigations are important to our understanding of immigrant adaptation, as studies of assimilation they leave a crucial gap: If assimilation is the decline of differences between groups, how can we understand immigrants’ assimilation without also looking at non-immigrants (henceforth “natives”)? To know if differences between immigrants and natives are declining, it is necessary to compare the two groups. This, however, raises another problem: To which natives, precisely, shall we compare immigrants? Classical assimilation perspectives assumed that middle-class Protestant whites were the natural reference group against which to evaluate immigrants’ assimilation. One valuable contribution of segmented assimilation theory is the recognition that American society is very diverse, and that there are multiple possible native groups with which immigrants may assimilate. Segmented assimilation theory points to the diversity of residential settlement patterns of new immigrant families as one source of variation in the assimilation pathways they experience. There is considerable variation in both the ethnic and socioeconomic makeup of receiving communities and their associated schools, and therefore in the native Americans with whom immigrants will come into contact.

For this research I conceptualize assimilation as the degree of difference between immigrants and natives within the local context. This conceptualization recognizes that opportunities for inter-group interaction are largely dependent on spatial proximity. Presumably, inter-group interaction is necessary for assimilation to occur – that is, for differences between groups to decline. Thus, the appropriate group of natives to which to compare immigrants is one with which they have frequent contact, such as natives who attend the same school. For some immigrants, this native comparison group will be middle-class whites, while for others it may be multi-racial or composed primarily of minority individuals or working- class whites. Regardless, I measure immigrants’ degree of assimilation as the difference between immigrant adolescents and their native counterparts within the same school.

Under this definition, assimilation must be defined with respect to a particular outcome that affords a comparison between immigrants and natives. While there are many possible choices, here I examine differences between immigrant and native adolescents with respect to the risk behaviors of serious delinquency and controlled substance use. These outcomes are appropriate for several reasons. First, previous research has established that they are among the outcomes that concern immigrant parents as their children become “Americanized” (Perreira, Chapman, and Stein, 2006; Portes and Rumbaut, 2001; Zhou and Bankston, 1998). Second, previous research has indicated that recent immigrant children do well relative to natives with respect to both substance use and delinquency, but that their advantage tends to fade over time (Powell, Perreira and Harris, 2009) and in later generations (Bui, 2008; Bui and Thongniramol, 2005; Greenman and Xie, 2008; Harris, 1999; Rumbaut, 1997). This implies that the behaviors of immigrant children converge to those of natives as they experience assimilation, but this proposition has not been explicitly tested. Finally, these outcomes are social behaviors, as adolescents usually engage in them while with others (Haynie and Osgood, 2005). Therefore, I can expect these outcomes to be particularly influenced by adolescents’ peer groups. Such peer-influenced outcomes are especially likely to reveal assimilation in the form of declining differences between groups. I use the term “behavioral assimilation” to refer to these measures of assimilation. Behavior differences between immigrants and natives fall under the rubric of “cultural assimilation” or “acculturation” in Milton Gordon’s (1964) classification of assimilation variables (see Gordon 1964, Table 5). I prefer the term “behavioral assimilation” for this analysis due to its greater specificity and the fact that it emphasizes the focus on a specific behavioral comparison between immigrants and natives within the local context. Readers should be aware, however, that this type of assimilation is only one of many possible forms.

Specific hypotheses: Contextual, parenting, and peer effects

Several bodies of literature inform the hypotheses tested in this research. An extensive literature on the effect of social context on adolescent development suggests that adolescent outcomes differ depending on school and neighborhood characteristics. In turn, social contexts are linked with interpersonal relationships that are crucial to adolescent development, particularly relationships with parents and peers. This paper integrates the arguments of segmented assimilation theory with those of the social context, parenting, and peer effects literatures. Below, I briefly review these literatures and draw on them to derive specific research hypotheses.

While there are various ways to conceptualize social context, the majority of research on contextual effects on youth outcomes has examined neighborhood contexts. For the central question of this paper – does the degree of behavioral similarity between immigrant adolescents and their native peers vary by social context – it is preferable to examine school contexts. Schools provide the majority of adolescents’ opportunities for friendship and social interaction (Ennett and Bauman, 1993; Gaviria and Raphael, 2001), meaning that inter-group interactions that facilitate assimilation take place more frequently within schools than in other social contexts. Furthermore, previous studies of contextual effects on immigrant youth have contended that theoretical arguments developed with respect to neighborhood contexts also apply to school contexts (Pong and Hao, 2007). As a practical matter, schools also have the advantage of being discrete social units with clear boundaries (Gaviria and Raphael, 2001), while neighborhoods lack such boundaries. Schools are therefore the preferred definition of social context for this analysis.

Previous literature on the effects of school context has primarily examined educational outcomes. Studies of contextual influences on adolescent risk behavior have instead tended to use neighborhoods as their definition of context. Because much of the current knowledge about this topic comes from neighborhood-based research, both school and neighborhood effects literature will be reviewed briefly here.

Many empirical studies have assessed the relationship between socially or economically disadvantaged neighborhood or school environments (usually measured as some combination of poverty rates, unemployment, education levels, public assistance rates, and/or prevalence of single-parent families) and adolescent delinquency or violence. Such research has typically found that neighborhood or school disadvantage is associated with higher levels of delinquency and/or violence (Anderson, 2002; Bellair and McNulty, 2005; Bellair, Roscigno, and McNulty, 2003; Leventhal and Brooks-Gunn, 2000; Sampson and Groves 1989). Controlled substance use has been examined less frequently in relation to school and neighborhood contexts, and findings from such studies have been less consistent. While some studies have found that neighborhood disadvantage is related to higher substance use (Rankin and Quane, 2002), other research has found either no relationship (Allison et al., 1999) or that youth in high-SES neighborhoods (Reardon, Brennan, and Buka, 2002; Ennet et al., 1997) or schools (Hoffman, 2006) are actually somewhat more likely to use controlled substances.

Past studies have not always concluded that associations between school or neighborhood characteristics and youth outcomes are causal. For example, Rankin and Quane (2002) find that neighborhood disadvantage is associated with higher delinquency and substance use rates, but that this association is due primarily to the sorting of families with disadvantageous characteristics into poor neighborhoods. Despite the lack of consensus about causality, an extensive literature documents correlations between contextual characteristics and youth outcomes, including risk behaviors (see Leventhal and Brooks-Gunn (2000) for a comprehensive review of the literature on neighborhood effects).

These correlations are a key motivation for the present research. According to modern assimilation perspectives (including both “new” assimilation theory (Alba and Nee 2003) and segmented assimilation theory (Portes and Zhou 1993), immigrants will influence and be influenced by the natives with whom they have contact. Thus, if risky behaviors are more prevalent among native youth in low-SES contexts than in high-SES contexts, the peer groups of immigrant adolescents will include more problematic peers in low-SES contexts. Behavioral assimilation in such contexts would imply a greater level of risky behavior, on average, than behavioral assimilation in high-SES contexts. This implication is an important motivation for segmented assimilation theory, which suggests that immigrant youth who assimilate into disadvantaged school and neighborhood contexts are at risk of adopting “oppositional youth cultures” (Zhou 1997). The theory does not take into account that both immigrant parents and adolescents may foresee the negative consequences of this type of behavioral assimilation and deliberately take steps to avoid it. If so, immigrant adolescents attending disadvantaged schools may assimilate less than those attending more advantaged schools. It follows that differences between immigrant adolescents’ outcomes and those of their native peers will be smaller in high-SES schools than in low-SES schools. The first hypothesis is therefore:

H1 – Behavioral Assimilation: The gap in risk behavior between immigrants and natives will be larger in lower-SES schools than in higher-SES schools

Note that although previous literature indicates that there is likely a stronger relationship between school SES and delinquency than between school SES and substance use, the behavioral assimilation hypothesis applies to behavioral assimilation in terms of substance use as well as delinquency. As long as parents and teenagers observe high levels of at least some risk behaviors in lower-SES schools – whether these include high delinquency rates, teen pregnancy, school dropout, gang activity, etc. – perception of a “riskier” social environment may cause them to take steps to avoid assimilating. Thus, parents’ responses to risk behaviors other than substance use may produce the anticipated effect on substance use, and the behavioral assimilation hypothesis is meaningful regardless of whether this analysis finds higher average substance use in low-SES schools.

The remaining hypotheses tested in this research concern mediating factors that may explain the relationship (if any) between school context and the behavioral assimilation of immigrant adolescents. The first potential mediator is the peer group. Adolescence is commonly recognized to be a time when family relationships become less salient and peer relationships take on increasing importance. Violence and delinquency have been frequently studied with respect to peer effects. The extensive literature on this topic has found a consistent and strong correlation between individuals’ delinquency and that of their friends (Haynie and Osgood, 2005). ¹ Moreover, the effect of peers is not limited to close friends: Felson et al. (1994) find that school peers’ average values regarding violence and delinquency have an influence on the behavior of individual boys, even controlling for the boys’ own attitudes. Peer influences have also been studied as potential mediators between disadvantaged social contexts and youth outcomes. Rankin and Quane (2002) and Haynie, Silver, and Teasdale (2006) both find that some of the effect of neighborhood disadvantage on adolescent violence is due to lower-quality peer groups in poor neighborhoods. This suggests that for immigrant families in disadvantaged contexts, limiting children’s assimilation into negative peer groups could be an effective protective strategy. One way to limit assimilation is for adolescents to forge friendships primarily within the immigrant ethnic community, thus limiting friendships with native peers. This strategy is likely to be successful in reducing negative behaviors that are less prevalent among immigrants than natives, including substance use and delinquency; it would not be likely to be helpful (or could even be harmful) in reducing negative behaviors more prevalent among immigrants than among surrounding natives. Inter-ethnic friendship may thus mediate the hypothesized relationship between school context and behavioral assimilation. The second research hypothesis is therefore:

H2 – Inter-ethnic friendship: Inter-ethnic friendship will be less common for immigrant students in lower-SES schools than for immigrant students in higher-SES schools

Another way of avoiding assimilation into problematic peer groups is for immigrant adolescents to limit friendships according to friends’ behavior rather than their ethnicity. Selecting friends who are not as engaged in risk behavior is one way that immigrant adolescents can create a “buffer” between themselves and the surrounding social environment, thus limiting the potential negative influence of social assimilation. However, it would be unrealistic to expect immigrant students’ friends to engage in fewer risk behaviors in schools where those behaviors are more prevalent; instead, the key prediction is that immigrants’ friendship groups are less affected by the school environment than those of natives. Thus, an interaction is implied between being an immigrant and being in a lower-SES school and/or a school with higher levels of student involvement in risk behavior. The third research hypothesis is therefore:

H3 – Peer Behavior Interaction: Attending a lower-SES school or a school with higher average levels of student risk behavior will have a smaller impact on the risk behavior of immigrant students’ friends than on the risk behavior of native students’ friends

Parenting practices are the second proposed mediating factor between school context and behavioral assimilation. Parents are likely aware of the dangers of friendships with deviant peers, and they may take action to protect their children from such influences. For example, Furstenberg et al. (1999) and Jarrett (1997) both found that parents in disadvantaged neighborhood contexts were highly aware of the dangers that such contexts posed for adolescents. These parents’ strategies to protect their children included restricting children’s friendship choices and their freedom of movement in the neighborhood, increasing parental supervision, and encouraging children to participate in activities that take place outside rather than within the neighborhood.

While the above studies were based on native families, there is reason to think that immigrant parents may be even more concerned about contextual dangers. Because many immigrants come from countries in which parents’ and teachers’ authority over the young is stronger and standards for behavior are much less lenient than in the United States (Portes and Rumbaut, 2001), they are likely to have stricter standards for adolescent behavior. While native parents may be no less fearful of extreme outcomes, such as children using hard drugs or being involved in gang violence, they may perceive less serious behaviors – such as drinking alcohol or acting rowdy with friends in public places – as more normal for adolescents. Therefore immigrant parents may a lower tolerance for youth risk behavior and be more proactive in shielding their children. The limited empirical evidence in this area supports this idea. Portes and Rumbaut (2001) found that the perceived permissiveness of U.S. culture made immigrant parents especially fearful of the dangers “Americanization” posed to their children. Perreira et al. (2006) found that Latino immigrant parents were specifically concerned about the potential negative consequences of their children’s friendships with native peers. The parents feared that such friendships increase their children’s risk of drug use, violence, and teenage pregnancy. Parents’ efforts to protect their children from such influences included increasing parental supervision and discouraging children from socializing with native peers.

Many studies have shown that parenting behaviors have an important influence on adolescent outcomes, including risk behavior. Several studies have found that parental supervision is related to lower levels of delinquency (Furstenberg et al., 1999; Haynie and Osgood, 2005; Haynie and South, 2005; Rankin and Quane, 2002; Simons et al., 2005). Friendships with deviant peers have also been shown to be reduced by increased parental monitoring (Knoester, Haynie, and Stephens, 2006; Rankin and Quane, 2002; Simons et al., 2005). If immigrant parents do indeed increase supervision and control in response to disadvantaged social contexts, they may reduce their children’s risk behavior both directly and through reducing their affiliations with problematic peers. Parenting behaviors are therefore potentially important mediators between school disadvantage and behavioral assimilation of immigrant adolescents. The fourth research hypothesis is thus:

H4 – Parental Control: Parental control will be higher for immigrant youth in lower-SES schools than for those in higher-SES schools

If parenting and friendship do indeed serve as mediators between social context and behavioral assimilation, they must explain part or all of the observed relationship between social context and behavioral assimilation. Therefore, assuming the empirical analysis supports the behavioral assimilation hypothesis (H1) by showing a lower degree of behavioral assimilation for immigrant adolescents in lower-SES schools, I will test one final hypothesis:

Data and Methods

This study uses data from Wave 1² of the National Longitudinal Study of Adolescent Health (Add Health) (Harris 2009). Add Health is a school-based survey of adolescents who were in grades 7–12 in 1994–1995. The in-school portion of the survey was administered to all students in the sampled schools who were present on the day of the survey. The in-school questionnaire covered such topics as demographic characteristics, parental education, health status, academic grades, and friendships, and was completed by more than 90,000 adolescents. A smaller “core” sample of Add Health respondents was selected to complete more in-depth interviews at home. Additional topics covered by this portion of the survey include nationality of students and of their parents, language spoken in the home, and many detailed measures of health risk behaviors, family dynamics, and psycho-social adjustment. Add Health is a good data source for this study because not only is its sample large and nationally representative, it also contains over-samples of Chinese, Cubans, and Puerto Ricans. As a result, I have adequate sample sizes of both Asian and Hispanic first- and second-generation adolescents (whom I collectively term “immigrant adolescents,” since they are all adolescents in immigrant families). Unfortunately, I do not have adequate sample sizes of other groups, so I limit my analysis to Asians and Hispanics.

Variables

School SES

Add Health has limited options for measuring school SES. A survey given to school administrators included some potentially useful questions, but high rates of missing responses preclude their use. Instead, I measure school SES by aggregating students’ responses to a question on parental education from the in-school survey. Aggregated responses to this question have been used by past researchers as an indicator of school SES. For example, Pong and Hao (2007) measure school SES using the percent of students’ parents who have a college degree or higher. I take a similar approach in this study. I measure school SES as the percentage of students’ mothers who did not finish high school, which ranges from less than 1% to 44%. The continuous version of this variable is used in all regression models. I also experimented with several categorical versions of this variable. All yielded very similar results to the continuous version.

Adolescent risk behavior

The Add Health in-home survey provides detailed information on participation in risk behaviors, from which I construct my dependent variables. I consider two types of risk behavior: Delinquent behavior and controlled substance use. My measure of delinquent behavior is based on a series of questions asking respondents to report whether or not they have participated in particular undesirable, illegal, or violent activities in the past year. The behaviors asked about range in seriousness from “acting rowdy in a public place” to shooting or stabbing someone. Because the less serious behaviors are not uncommon among adolescents, I focus here on more serious behaviors. I construct a scale measuring the number of such behaviors the adolescent reports having engaged in during the past year. The specific behaviors are: Used or threatened to use a weapon to get something from someone; pulled a knife or gun on someone; shot or stabbed someone; carried a weapon to school; sold illegal drugs; broke into a home or building to steal something; stole something worth more than $50; got into a “serious” physical fight; hurt someone badly enough in a fight to need medical attention; got hurt in a fight badly enough to need medical attention; and took part in a fight of one group against another. The scale ranges from 0 (for respondents who reported no such behaviors) to 11 (for respondents who participated in every behavior). Because these behaviors vary by age, I age-standardize the measure by taking the respondent’s age-specific percentile score on the delinquency scale. The final variable thus potentially ranges from 0 to 100.

I derived the measure of controlled substance use from the self-reported frequency of use of tobacco, alcohol, and marijuana. As expected, use of controlled substances varies highly with age and by substance. Therefore, I age-standardized the three items by calculating the respondent’s age-specific percentile score for each substance. I then combined the information from the three items into a single scale by taking the average percentile score across all three.

Mediating variables

The in-school survey is used to measure friends’ ethnicity and participation in risky behaviors. The survey asked each student to name up to 10 close friends in the same school, making it possible to link each individual’s survey responses with those of his/her friends. It is thus possible to measure friends’ characteristics and behavior directly from their own survey responses. The in-school survey included ordinal scales measuring frequency of fighting and use of alcohol and tobacco. The fighting scale ranges from 0 (never fought in the past year) to 4 (got into 7 or more fights in the past year). I treat the average score on this scale among a student’s friends as an indicator of friends’ delinquency. Unfortunately, the in-school survey did not measure other types of delinquent or violent behavior, so it is not possible to construct a delinquency scale for respondents’ friends that is equivalent to the scale used for the main sample (which is based on the in-home survey). The ordinal scales for alcohol and tobacco use range from 0 (no use over the past year) to 6 (used the substance “nearly every day” in the past year). I create a single substance use scale for each friend by summing the responses to the two questions. I use friends’ average score on this scale as a measure of friends’ substance use. As with delinquency, it is not possible to create an exact equivalent of the substance use measure constructed for the main sample, which also includes marijuana use.

I also examine the ethnic composition of friendship groups by constructing a measure of the propensity of an adolescent to make friends within his/her ethnic group. Opportunities for inter-ethnic friendship are heavily influenced by relative group size (Zeng and Xie 2008). That is, the more co-ethnic peers available, the higher the likelihood of having co-ethnic friends. Here I wish to focus on friendship choices, which are controlled by the individual, rather than school ethnic composition, which is not under individual control. Therefore, I use a measure of friendship composition that is purged of opportunity structure (school ethnic composition) and instead reflects immigrant students’ degree of preference for co-ethnic friends. ³ This measure, which I refer to as F, is the difference between the proportion of co-ethnics found among the adolescent’s friends and the proportion of co-ethnics in the adolescent’s school. If there is no ethnic preference in friendship, we would expect these two proportions to be equal. A negative value indicates a tendency to choose co-ethnics as friends, a value of zero indicates that the respondent has no ethnic preference in friendship, and a positive value indicates that the respondent tends to choose friends outside his/her ethnic group.

The in-home survey contains series of questions designed to measure parental supervision of and control over adolescents, which I use to operationalize parental monitoring. One series of questions asks whether a parent is home at certain times of day, including before school, after school, at dinnertime, and at bedtime. Because the after school and early evening hours are the times adolescents are thought to be most prone to engaging in risky behaviors, I construct a measure of very low parental supervision at these times of day. The measure is a binary variable coded 1 if a parent is “almost never” or “never” present after school and a parent is present during dinner less than three evenings a week. Several other measures of parental monitoring were also constructed based on these questions (i.e., measures including supervision at other times of day), but because all provided similar results, only one is presented in the tables.

Another series of questions measures parental efforts to control adolescent behavior. These questions, which are answered by the adolescent, consist of seven items measuring whether or not the parent allows the adolescent to make his/her “own decisions” about curfews, friendship choices, what to wear, how much television to watch, which programs to watch, bedtime, and what to eat. I construct two measures of parental control from these questions. Because the item about friendship choices is the most directly relevant to my research questions, I construct a binary indicator equaling 1 if the adolescent makes his/her own decisions about friends, 0 otherwise. The second measure is a count of the number of “own decisions” an adolescent is allowed to make. Finally, a parent was interviewed whenever possible during the in-home survey.⁴ Although parents were not asked about supervision or specific rules for their children, they were asked how often they “make decisions with” the reference child. I create a binary indicator of parental decision participation equaling 1 if the parent reports “always” or “often” making decisions with the adolescent, 0 otherwise. For brevity, I refer to this group of supervision and decision-making items as parental control measures.

Control Variables

Add Health provides an array of information on family background and demographic characteristics. The following variables are used as controls in the analysis: Age (included as a series of single-year-of-age dummy variables measuring Wave 1 age (omitted category= under age 14)); gender (omitted category=male); family structure (3 categories: single-parent family; step-parent or “other family”; and 2 biological parent family (omitted)); logged family income (imputed for those with missing parent interviews); parent interview missing (dummy variable indicating whether the parent interview was missing (omitted=not missing)); parental education (continuous variable giving years of education of the single parent if a single-parent family, average education of the parents if a 2-parent family); immigrant generation and length of stay (3 categories: first-generation with less than 5 years in the U.S. (omitted), first generation with greater than 5 years in the U.S., and second generation); school racial composition (3 variables giving student body percent black, Hispanic, and Asian (percent white omitted due to multicollinearity)); speaks English at home (dummy variable indicating whether English is the primary language spoken at home (omitted= foreign language spoken at home); and specific Hispanic or Asian ethnicity (Hispanic omitted category is Mexican, Asian omitted category is Chinese).

Methods

Due to the approach I take to defining and measuring behavioral assimilation, the primary dependent variable in my analysis is the difference in risk behavior between an immigrant adolescent and non-immigrant adolescents in the same school. In addition to measuring the respondent’s own delinquency or substance use, it is therefore necessary to measure average school levels of the same behavior. A potential problem arises from the relatively small sample sizes in many schools for the in-home survey, from which the measures of risk behavior are taken. Because the behaviors in question are strongly influenced by sex and age, it is preferable to calculate a school average of risk behavior net of the sex and age composition of the respondents sampled in that school, which is largely a function of chance. Therefore, instead of simply taking the average level of behavior, I regress risk behavior on sex, age, and a series of school dummy variables for non-immigrant adolescents. I use a modeling strategy in which i adolescents are clustered within j schools. In the following models, the superscript 0 refers to non-immigrants, while the superscript 1 refers to immigrants. The equation used is thus ${Y^0}_{ij}=B_{0j}^0+{B_1^0}^{\prime }{A^0}_{ij}+{{\epsilon }^0}_{ij}$, where Y⁰_{i j} is natives’ risk behavior, A⁰_ij is a vector of variables including age and sex, and $B_{0j}^0$ is the school-specific intercept term measuring average behavior differences across schools. I then apply the coefficients from the resulting regression model to the sample of immigrant students and generate a predicted value for each student based on his/her age, gender, and school, ${\widehat{Y}}_{ij}^0$. Under the assumption that school differences in risk behavior among native youth do not depend on sex or age, this predicted value is equivalent to a measure of the average behavior of natives of the same age and sex as the immigrant adolescent within a particular school.

I then define my dependent variable as the difference between the immigrant adolescent’s level of risk behavior and the school average level of risk behavior, $Y_{ij}^1-{\widehat{Y}}_{ij}^0$, which will be referred to as d (for difference). This two-step estimation strategy, in which differences between immigrants and non-immigrants are incorporated into the first-step calculation of the dependent variable, allows me to later restrict my sample to immigrants without losing the ability to model differences between immigrants and non-immigrants. This is preferable because effects for covariates included in the models may differ for immigrants and natives, particularly those having to do with assimilation. Restricting the model to immigrants ensures that the results are accurate for the main group of interest, immigrant adolescents⁵.

I run all analyses separately for Hispanics and Asians. Specific models are explained more fully when they are presented in the “results” section. Due to the high data requirements for making within-school comparisons, I do not have a sufficient sample sizes to examine specific national origin groups separately. I include national origin as an additive control in my analytical models. In all statistical analyses, I use appropriate weights to account for stratified sampling, non-proportionate non-response, and non-proportionate attrition. Observations without sample weights were dropped from the analysis. In regression analyses, I also appropriately correct standard errors for Add Health’s complex survey design, including its use of clustering and stratification. The total sample sizes are 1,451 Hispanic immigrant youth and 810 Asian immigrant youth. Due to small numbers of missing values, analytic sample sizes vary somewhat for models with different dependent variables. Sample sizes for each specific model are included in the tables. The sample is spread across 96 schools, all of which contain at least one sampled Asian or Hispanic immigrant adolescent (33 schools were dropped from the analysis due to containing no immigrants). Of these 96 schools, 16 schools contain no Hispanic immigrants and 25 schools contain no Asian immigrants. The average number of sampled Hispanic immigrants per school is 17.6, while the average number of sampled Asian immigrants per school is 9.7.

Results

Descriptive Results

Appendix A gives descriptions and averages by race/immigration status of all variables used in the analysis. Past work has revealed that risk behaviors should be lower for immigrants than natives, and indeed Appendix A shows that both immigrant groups have somewhat lower substance use rates than natives. Although Hispanics have higher average delinquency rates than natives, this is most likely due to differences in socioeconomic status: Appendix A also reveals that Hispanic immigrants attend lower-SES schools and come from families with lower incomes and lower parental education than natives. The same is not true for Asian immigrant youth, who come from families with incomes and parental education similar to those of natives. Appendix B presents descriptive statistics showing the relationships between school SES, risk behavior, and “traditional” assimilation measures that have been used in previous literature (generation, length of stay for first-generation immigrants, number of immigrant parents, and home language use). With one exception (length of stay for Hispanics), the average school percentage of mothers lacking a high school diploma is higher for the less assimilated, but in most cases the difference is small (1–2 percentage points). Risk behavior differences are generally consistent with previous literature in showing lower levels of substance use and delinquency for first-generation (Bui, 2008; Bui and Thongniramol, 2005; Greenman and Xie, 2008; Harris, 1999; Powell, Perreira and Harris, 2009) and less assimilated youth (Greenman and Xie, 2008), although there are exceptions for both Asians (for number of immigrant parents) and Hispanics (for length of stay). The interrelationships among traditional measures of assimilation, school SES, and risk behavior indicated by Appendix B point to the importance of relying on multivariate models, which can estimate the effects of these different factors simultaneously, to test the paper’s main hypotheses.

Table 1 presents descriptive statistics on the relationship between risk behaviors, school SES, and race/immigration status. For descriptive purposes, school SES has been coded into a 3-category variable where the top category (“high” SES) includes schools whose percentage of mothers who did not graduate from high school is at the 25^th percentile or below, the “moderate” category includes schools between the 25^th and 75^th percentiles, and the “low” category includes schools at or above the 75^th percentile. There are 24 low-SES schools, 48 moderate-SES schools, and 24 high-SES schools. In order to ease interpretation, the information is also presented graphically in Figure 1 (substance use) and Figure 2 (delinquency). For substance use, being in a lower-SES school is related to a more favorable outcome for both Asian and Hispanic immigrant adolescents. While immigrant students in high-SES schools have substance use levels very close to the median, substance use levels among those in low-SES schools are only at the 38^th percentile (Asians) or 45^th percentile (Hispanics). This pattern does not hold among either white or black natives, for whom there is no consistent relationship between school SES and substance use, but is similar among the small group of “other” natives. This lack of a clear relationship for natives is consistent with previous literature on contextual effects on youth substance use.

Table 1.

Means of Risk Behavior Variables by Race/Immigration Status and School SES

	Substance Use				Delinquent Behaviors
		School SES				School SES
	N	Low	Moderate	High	N	Low	Moderate	High
Race/Immigration Status
Asian immigrant	805	37.9 **, +++	43.0 *	50.5	810	40.0 ++	47.2	48.1
Hispanic immigrant	1448	45.0 *	47.2	49.2	1451	51.7	56.1	51.4
White native	5252	52.6	51.6 *	53.5	5294	53.4 ***, ++	48.1	47.5
Black native	2101	47.7 +	45.1	46.7	2121	60.7 **, ++	57.0	55.3
Other native	934	52.1 **	51.7 **	57.7	942	61.8 **, ++	52.2	51.4

Statistically different from high SES at:

^*p< .10

^**p< .05

^***p<.01

Statistically different from moderate SES at:

⁺p< 0.1

⁺⁺p< 0.05

⁺⁺⁺p<.01

Figure 1.

Substance Use by School SES

Figure 2.

Delinquency by School SES

For Asian immigrant youth, delinquency is significantly lower in low-SES schools than in high-SES schools. For Hispanic youth there is no obvious pattern. For all three groups of native youth, by contrast, delinquency levels appear to increase as school SES goes down. The results for both substance use and delinquency suggest a different relationship between school SES and risk behavior for immigrants than for natives. While suggestive, these results are not definitive due to the possible confounding influence of family-level SES and assimilation factors, which are likely related to school SES. I therefore turn to regression models to test the paper’s main hypotheses.

Multivariate Results

Table 2 presents results from models predicting d, the measure of behavioral assimilation. Due to the way the dependent variable was constructed, correctly interpreting the regression results in Table 2 requires an examination of the constant from the model including only school SES and no covariates (first panel). Recall that the dependent variable, d, is defined as the difference between an immigrant’s risk behavior and the mean risk behavior for natives in the same school, adjusting for age and gender. A negative average value of d thus indicates that immigrants engage in lower levels of risk behavior than natives in the same school. The school SES coefficient tells us whether immigrants are engaging in less risk behavior, relative to natives, as the proportion of students’ mothers without a high school diploma increases. A negative coefficient would indicate that immigrants compare more favorably to natives in lower-SES schools. However, the sign of the coefficient alone cannot inform us about the central research question of this paper – whether the behavior of immigrant and native youth is more or less similar in lower-SES schools. To make this interpretation, we need to know how the risk behaviors of immigrant and native youth compare in the highest-SES schools. If there is no difference in high-SES schools, or if immigrants have lower risk behavior than natives in these schools, then a negative coefficient on school SES indicates an increasing behavior gap as school SES declines. Thus for each model, it is necessary to first examine the sign and significance of the constant term. If the constant term is negative or not significantly different from 0, it is valid to interpret a negative coefficient on school SES as evidence of an increasing behavior gap.⁶

Table 2.

The Effect of School SES on Immigrant Adolescents’ Behavioral Assimilation

	Asians				Hispanics
	Substance Use		Delinquent Behaviors		Substance Use		Delinquent Behaviors
	coef	(se)	coef	(se)	coef	(se)	coef	(se)
Model 1 - No Controls:
School SES	−0.26	(0.14) *	−0.45	(0.17) **	−0.05	(0.10)	0.00	(0.19)
Intercept	−2.58	(3.06)	0.51	(3.05)	−1.52	(2.24)	2.37	(4.46)
Model 2 - With Controls:
School SES	−0.31	(0.15) **	−0.95	(0.28) ***	−0.39	(0.17) **	0.17	(0.30)
Age 14	3.43	(2.97)	14.25	(7.92) *	4.74	(2.26) **	0.17	(5.21)
Age 15	−0.92	(2.93)	7.65	(8.24)	−1.76	(2.76)	−1.29	(5.01)
Age 16	−5.12	(3.47)	7.01	(8.43)	1.52	(2.22)	−1.68	(5.89)
Age 17	−3.62	(3.48)	5.11	(7.45)	0.54	(2.31)	0.22	(6.44)
Age 18	−2.19	(3.38)	4.49	(7.68)	−3.91	(2.45)	−5.71	(5.80)
Age 19+	−2.45	(4.04)	3.03	(8.65)	−3.83	(2.52)	−7.37	(6.26)
Female	−0.72	(2.15)	5.05	(2.30) **	0.08	(1.23)	3.30	(2.68)
Single-parent family	−0.95	(1.53)	2.40	(5.00)	3.18	(1.74) *	6.06	(2.91) **
Step-parent family	−2.59	(1.42) *	−6.53	(3.28) **	0.83	(1.77)	9.00	(3.19) ***
Family income	0.47	(1.13)	0.01	(2.85)	0.23	(0.43)	−0.38	(0.70)
Parent interview missing	1.58	(2.24)	−1.05	(3.30)	1.82	(2.31)	−2.66	(2.27)
Average parental education	0.19	(0.25)	−0.99	(0.38) **	0.18	(0.19)	0.71	(0.48)
First generation, LOS>5 years	3.73	(1.17) ***	0.66	(4.29)	2.75	(2.06)	−2.39	(3.76)
Second generation	3.38	(1.28) ***	−1.96	(5.21)	6.70	(1.94) ***	5.71	(4.28)
School % black	8.82	(4.89) *	−4.01	(7.71)	8.60	(4.72) *	−0.46	(14.77)
School % Hispanic	11.33	(7.69)	28.09	(13.61) **	16.62	(7.09) **	−5.33	(9.33)
School % Asian	18.44	(6.68) ***	32.84	(12.03) ***	3.05	(7.47)	2.09	(17.91)
Speaks English at home	7.25	(2.09) ***	3.06	(4.08)	3.23	(1.25) **	−3.16	(2.53)
Cuban	--	--	--	--	−0.34	(2.07)	−3.98	(3.72)
Puerto Rican	--	--	--	--	−2.00	(2.25)	−2.05	(3.48)
Central/South American	--	--	--	--	−2.77	(1.70)	−4.45	(3.13)
Other Hispanic	--	--	--	--	−5.39	(2.15) **	−11.18	(3.78) ***
Filipino	−0.33	(2.44)	−1.99	(3.01)	--	--	--	--
Japanese	−5.27	(2.79) *	−2.40	(7.35)	--	--	--	--
Indian	3.13	(4.89)	12.41	(7.64)	--	--	--	--
Korean	0.11	(1.94)	4.06	(4.13)	--	--	--	--
Vietnamese	0.20	(2.50)	−3.61	(3.41)	--	--	--	--
Other Asian	0.61	(2.51)	12.25	(3.86) ***	--	--	--	--
Intercept	−21.97	(10.55) **	−1.89	(31.07)	−12.39	(6.14) **	−4.16731	12.53823
Analytic sample size	805		810		1448		1451

Statistical Significance:

^*p< .10

^**p< .05

^***p<.01

Note: Omitted categories for the independent variables are as follows: Age 13, male, 2-biological-parent family, parent interview not missing, first generation with length of stay less than 5 years, speaks non-English language Mexican ethnicity (for Hispanics models), Chinese ethnicity (for Asian models)

H1 - Behavioral Assimilation Hypothesis

H1 (the behavioral assimilation hypothesis) is tested in Models 1 and 2 in Table 2. Model 1 includes only school SES without any covariates. The intercept terms in the first panel of Table 2, for both outcomes and both ethnic groups, are not significantly different from 0. We may therefore interpret negative school SES coefficients as indicators of increasing behavior gaps. Model 2 adds an extensive set of control variables, including family-level socioeconomic and assimilation indicators. Because of the possibility that the bivariate relationship between school SES and behavioral assimilation could be driven by either family SES (i.e., immigrant youth from higher-SES families may be more likely to go to high-SES schools and also be more assimilated) or family assimilation level (i.e., youth from families who have been in the U.S. longer may be more likely to go to higher-SES schools), Model 2, rather than Model 1, provides the key test of the behavioral assimilation hypothesis. For substance use, the school SES coefficient for Asian youth is only marginally significant in Model 1, but becomes significant at the .05 level following the addition of the control variables in Model 2. The coefficient of −.31 in Model 2 indicates that for every 1 percentage point increase in the school’s percent of students’ mothers who did not graduate from high school, the difference between Asian immigrants’ and natives’ substance use scores increases by about a third of a percentile point. For Hispanic youth the results are similar: While there is no significant relationship between school SES and substance use in Model 1, Model 2 reveals that the substance use gap is larger in schools with a greater percentage of mothers who did not finish high school.

Results for delinquency are presented in the second panel of Table 2. For Asians, the coefficient of school SES in Model 1 is negative and statistically significant at the .05 level. After the addition of control variables in Model 2, its magnitude more than doubles, growing from −.45 to −.95. Model 2 thus reveals a sizeable relationship between school SES and the delinquency gap for Asians: After controlling for family background and school ethnic composition, for every 1 percentage point increase in the school’s percent of students’ mothers who did not graduate from high school the delinquency gap between Asian immigrant and native youth increases by almost a full percentile point. For Hispanics, by contrast, there is no significant relationship between school SES and the delinquency gap in either of the models. Overall, the findings support the behavioral assimilation hypothesis for both groups when the outcome is substance use, but only for Asians when the outcome is delinquency⁷.

H2: Inter-ethnic friendship hypothesis

What explains this relationship between school SES and immigrant-native differences in risk behavior? H2 and H3 proposed immigrant adolescents’ friendship choices as a possible explanation, while H4 proposed parenting behaviors. Table 3 presents results from the tests of these three hypotheses. The first panel shows results testing H2, the inter-ethnic friendship hypothesis. The coefficients in this panel are from OLS regression models in which F, the propensity for inter-ethnic friendship, is the dependent variable. The unadjusted model shows no significant relationship between school SES and propensity for inter-ethnic friendship for Asians, and an only marginally significant relationship for Hispanics. For Asian youth, but not Hispanic youth, the adjusted model shows a small but statistically significant relationship in the expected direction – lower school SES is related to a lower likelihood of inter-ethnic friendship. The analysis therefore provides support for the inter-ethnic friendship hypothesis for Asian immigrant youth, although the effect size is small. For Hispanic youth, the inter-ethnic friendship hypothesis is not supported.

Table 3.

The Effect of School SES and School Peer Behavior on Friendship and Parental Supervision

	Asians				Hispanics
	Unadjusted		Adjusted		Unadjusted		Adjusted
(Dependent Variables in Italics):	coef	(se)	coef	(se)	coef	(se)	coef	(se)
Friendship
Model 1: Inter-ethnic friendshipa
School SES	0.00	(0.01)	−0.01	(0.00) ***	−0.004	(0.002) *	0.00	(0.00)
Model 2: Friends’ average fightingb
Immigrant Adolescent	0.02	(0.17)	0.14	(0.20)	−0.24	(0.15) *	−0.43	(0.17) **
School SES	−0.002	0.002	−0.004	0.002 **	−0.002	0.002	−0.004	0.002 **
School peer fighting	0.67	(0.08) ***	0.47	(0.10) ***	0.67	(0.08) ***	0.49	(0.10) ***
School SES* Immigrant	−0.01	(0.00) *	−0.01	(0.01)	−0.01	(0.00)	0.00	(0.00)
Peer fighting* immigrant	−0.05	(0.25)	−0.15	(0.29)	0.51	(0.19) ***	0.41	(0.19) **
Model 3: Friends’ average substance useb
Immigrant Adolescent	0.99	(0.62)	0.86	(0.76)	1.21	(0.37) ***	−0.37	(0.45)
School SES	−0.006	0.005	−0.007	0.007	−0.01	(0.01)	−0.01	(0.01)
School peer substance use	0.95	(0.03) ***	0.62	(0.06) ***	0.95	(0.03) ***	0.63	(0.06) ***
School SES* Immigrant	−0.04	(0.02) *	−0.05	(0.03) **	−0.01	(0.01)	0.01	(0.01)
Peer substance use* immigrant	−0.42	(0.18) **	−0.39	(0.21) *	−0.51	(0.14) ***	−0.48	(0.14) ***
Parenting
Model 4: Low supervisionc
School SES	−0.05	(0.03) *	−0.11	(0.05) **	0.02	(0.02)	0.08	(0.03) **
Model 5: No friendship rulesd
School SES	−0.04	(0.01) ***	−0.04	(0.02) *	−0.02	(0.01) ***	−0.03	(0.02)
Model 6: # own decisionse
School SES	−0.03	(0.01) **	−0.02	(0.02)	−0.02	(0.01) **	−0.03	(0.01) **
Model 7: Parent makes decisions with youthf
School SES	0.00	(0.02)	−0.07	(0.04) *	0.02	(0.01)	−0.01	(0.02)

Statistical Significance:

^*p< .10

^**p< .05

^***p<.01

^aOLS regression. Sample includes only immigrant adolescents. Asian N=670, Hispanic N=1079

^bOLS regression. Sample includes both immigrant adolescents of the specified ethnic group and native within the same school. Asian model N=7688, Hispanic model N=8099

^cLogistic regression. Sample includes only immigrant adolescents. Asian N=835, Hispanic N=1473

^dLogistic regression. Sample includes only immigrant adolescents. Asian N=834, Hispanic N=1470

^eOLS regression. Sample includes only immigrant adolescents. Asian N=833, Hispanic N=1464

^fLogistic regression. Sample includes only immigrant adolescents. Asian N=512, Hispanic N=1221

Note: All models control for the same covariates included in Table 2, Model 2

H3- Peer behavior interaction hypothesis

The next two panels of Table 3 test H3, the peer behavior interaction hypothesis, which states that risky behaviors of immigrant students’ friends will be less influenced by school social context than risky behaviors of native students’ friends. For Asians, the sample in these two analyses includes only native youth and youth in Asian immigrant families, while for Hispanics the sample includes only native youth and youth in Hispanic immigrant families. Thus the “immigrant” interaction terms give differences specifically between either Asian or Hispanic immigrant youth and natives. Table 3’s second panel shows results from models in which the average level of fighting among students’ friends is the dependent variable. The coefficients for school peer fighting show that it is positively related to friends’ fighting, as expected. The “main effects” of school SES are statistically insignificant and close to 0, indicating that there is no relationship between school SES and friends’ fighting for native youth (net of average fighting levels in the school). The negative coefficient on the immigrant-school SES interaction term in the unadjusted model for Asians, however, indicates that being in a lower-SES school is related to lower levels of fighting among friends for Asian immigrant youth, but it is only marginally significant. However, after adjusting for covariates there are no remaining significant interactions for Asian youth. By contrast, the interaction between being a Hispanic immigrant and school peer fighting is positive and significant in both the unadjusted and adjusted models. This indicates that while higher levels of school peer fighting are related to higher fighting among friends for non-immigrant youth, this effect is actually stronger for Hispanic immigrant adolescents.

The models shown in the third panel of Table 3 treat friends’ average substance use as the dependent variable. In both unadjusted models, the positive and significant “main” effect of school peer substance use for natives is .95, indicating that higher substance use among students in general predicts higher substance use among natives’ friends. For both immigrant groups, the significant negative interaction terms between being an immigrant and school peer substance use show that this effect is only about half as strong (.95−.42 = .53 for Asians, .95−.51 = .44 for Hispanics) for immigrants. These results are also apparent in the adjusted models, although the effect for Asians drops to the .1 significance level. The interaction term between being an immigrant and school SES is also negative and significant for Asians in the adjusted model, indicating that although school SES has no relationship with friends’ substance use for natives (net of school peer behavior), Asian immigrants’ friends have lower average substance use in lower-SES schools. The peer behavior interaction hypothesis is thus supported for both Asian and Hispanic immigrant youth when the outcome is friends’ substance use, and partially supported for Asian youth (before controlling for covariates) when the outcome is friends’ fighting behavior, though with marginal statistical significance. On the other hand, the results for friends’ fighting behavior among Hispanic immigrant youth were exactly opposite of those predicted by the peer behavior interaction hypothesis.

H4 - Parental control hypothesis

The bottom half of Table 3 presents four models testing the parental control hypothesis (H4). The number of decisions the youth is allowed to make for him/herself is modeled using OLS regression, while the other three parenting measures are modeled using logit regression. Both the unadjusted and adjusted results for Asians show that as the percentage of students’ mothers without a high school diploma rises, the odds of youth having low afternoon/evening supervision levels or lacking rules about friendships decline. While the unadjusted model indicates that Asian students in lower-SES schools are allowed to make fewer of their own decisions, this result disappears after addition of control variables. The parenting results for Asians thus provide limited support for the parental control hypothesis. For Hispanics, the results do not clearly support the parental control hypothesis: While being in a lower-SES school is related to Hispanic immigrant adolescents being allowed to make fewer “own decisions,” after adjusting for covariates it is also related to a higher probability of low parental supervision.

H5 – Mediation hypothesis

Analyses testing the behavioral assimilation hypothesis (H1) revealed significant relationships between school SES and the substance use gap for both ethnic groups, and between school SES and the delinquency gap for Asian youth. Models 3 and 4 of the behavior gaps, presented in Table 4, now test whether these relationships can be attributed to friendship or parenting factors, respectively (the mediation hypothesis). Table 4, Model 3 adds the friendship variables to the variables included in Table 2, Model 2.⁸ There is no significant effect of inter-ethnic friendship on the substance use gap. For both Asians and Hispanics, the positive effects of friends’ behavior (average substance use) on the substance use gap show that immigrant students whose friends engage in more substance use compare less favorably with their native peers than do immigrant students whose friends have lower levels of substance use. For Asian students, after taking these friendship factors into account, the effect of school SES is considerably smaller than in Table 2, Model 2 and no longer statistically significant. This suggests that the tendency of Asian immigrant students in low-SES schools to choose friends who engage in lower levels of substance use is an important part of the explanation for the larger immigrant-native substance use gap in low-SES schools. The same is not true for Hispanic students: The school SES coefficient changes very little between Table 2, Model 2 and Table 4, Model 3.

Table 4.

Parenting and friendship as mediators between school SES and adolescent’s behavioral assimilation

	Asians				Hispanics
	Substance Use		Delinquent Behaviors		Substance Use		Delinquent Behaviors
	coef	(se)	coef	(se)	coef	(se)	coef	(se)
Model 3 - Adding Friendship variables to Table 2, Model 2
School SES	−0.14	(0.11)	−0.90	(0.25) ***	−0.43	(0.19) **	0.13	(0.29)
Has no friends	10.03	(2.20) ***	2.82	(3.86)	3.19	(1.72) *	4.02	(3.13)
Friends’ average behavior	3.60	(0.51) ***	5.13	(3.43)	1.82	(0.37) ***	5.90	(1.96) ***
Inter-ethnic friendship	−0.28	(0.36)	0.62	(0.71)	0.23	(0.40)	1.22	(1.02)
Analytic sample size	805		810		1446		1449
Model 4 - Adding Parenting to Table 2, Model 2
School SES	−0.16	(0.16)	−0.74	(0.44) *	−0.53	(0.17) ***	0.07	(0.29)
Low supervision	7.16	(2.76) **	8.22	(9.14)	3.34	(2.36)	2.90	(5.15)
No friendship rules	−1.63	(2.31)	−0.56	(6.37)	2.52	(1.56)	−2.96	(3.79)
# own decisions	0.38	(0.62)	−0.61	(1.08)	0.03	(0.46)	0.68	(0.97)
Parent makes decisions with youth	−3.11	(2.80)	0.52	(4.97)	−3.99	(1.39) ***	−4.90	(2.71) *
Analytic sample size	486		491		1193		1197

Statistical Significance:

^*p< .10

^**p< .05

^***p<.01

In addition to the variables shown, Models 3–4 control for age, gender, specific Asian or Hispanic ethnicity, “parent interview missing”, family income, average parental education, family structure, length of stay in U.S. for foreign-born immigrants, immigrant generation, whether English spoken at home, and school % black, school % Asian, and school % Hispanic, and whether adolescent has 1 or 2 immigrant parents.

Full results are available from author upon request.

While friendship factors are part of the substance use story for Asian students, they do not appear to have a similar effect on delinquency. Neither inter-ethnic friendship nor friends’ behavior (average fighting) has any effect on the difference in delinquency between Asian immigrant and native youth. Correspondingly, the addition of these variables does not explain the relationship between school SES and the delinquency gap.

Table 4, Model 4 adds the parenting variables to those included in Table 2, Model 2. For Asians, low parental supervision was the only parenting variable significantly related to school SES at the .05 level after adjusting for controls (Table 3). Therefore, it is most likely candidate to explain the relationship between school SES and the behavior gaps. Table 4, Model 4 shows that low parental supervision is also the only parenting measure that is significantly related to the substance use gap for Asians. Asian immigrant students who lack parental supervision in the afternoon and early evening hours compare less favorably to their native peers than students who are supervised at these times of day. More importantly for the mediation hypothesis (H5), the effect of school SES is reduced by about half compared to Table 2, Model 2. The larger substance use gap for Asian youth in low-SES schools can thus be attributed in part to the higher levels of parental supervision for students in low-SES schools. There is no similar pattern among Hispanics, however: The school SES coefficient actually becomes somewhat larger after the addition of the parenting variables to the substance use model. That parenting does not contribute to the larger substance use gap in low-SES schools for Hispanics is perhaps not surprising, given the lack of a consistent relationship between school SES and parenting shown in Table 3.

Turning to the results for delinquency, we see that parenting behaviors are more successful than friendship factors at explaining the relationship between school SES and the delinquency gap for Asians. Although the school SES coefficient remains significant at the .1 level, it drops in magnitude from −.95 to −.74 between Table 2, Model 2 and Table 4, Model 4. Thus, it appears that parenting factors explain at least some of the relationship between school SES and the delinquency gap. Overall, the results for Asian immigrant youth support the mediation hypothesis: Friendship and parenting factors do appear to explain, at least in part, the relationships between school SES and the behavior gaps. The results for Hispanic immigrant youth do not support the mediation hypothesis.⁹

Discussion and Conclusion

This paper explored the relationship between social context and patterns of assimilation among Asian and Hispanic immigrant youth. Although the degree of difference between immigrants and natives is at the core of the concept of assimilation, most previous work on immigrant assimilation has failed to make specific comparisons between immigrants and natives. The question of which group of natives form the appropriate comparison group for a given group of immigrants has also been under-studied in the literature. By contrast, this paper conceptualized assimilation as behavioral similarity between immigrants and natives within the local context. This behaviorally-based approach to defining and measuring assimilation, in contrast with other frequently-used measures of assimilation such as immigrant generation and length of stay, allows me to develop insights into aspects of the assimilation process that can potentially be controlled by immigrant families. I argue that immigrant families are active and deliberate participants in shaping their own assimilation trajectories.

The analysis confirms findings from previous investigations of the relationship between immigrant assimilation and the risk behaviors of substance use and delinquency, but the research strategy and results also differ in several important ways. Like previous research, including other studies based on Add Health data, the current analysis shows that immigrant children (especially those born abroad) tend to have lower rates of both substance use and delinquency than natives (Bui, 2008; Bui and Thongniramol, 2005; Greenman and Xie, 2008; Harris, 1999; Rumbaut, 1997). However, these findings have typically been based on comparisons between immigrant children and the average outcomes of all native children, despite large differences in the geographic dispersion of and schools attended by children of different immigrant generations. I argue that it is more informative to compare immigrant and native adolescents within the same school. Using this strategy, this analysis demonstrates that the previously documented immigrant advantage in risk behavior is not constant across schools, but rather is larger in low-SES schools than high-SES schools. It also sheds light on potential mechanisms linking school SES to differing levels of substance use and delinquency among immigrant youth. Previous research has examined whether differences between immigrant and native families in parenting-related factors contribute to generational differences in youth delinquency (Bui, 2008), but sources of variation in such mechanisms among immigrant families have not been examined. This analysis explored school context as a potential influence on both parenting behaviors and youth friendship choices among immigrant families.

The paper’s main hypothesis stated that immigrant adolescents will be less assimilated in social contexts in which behavioral assimilation is more likely to have negative consequences. I tested this hypothesis by examining the relationship between school SES and the degree of behavioral similarity between immigrant and native youth within the same schools. The results for Asian youth, and to a lesser extent for Hispanic youth, support this hypothesis. While in the highest-SES schools there are few differences in risk behavior between native youth and either immigrant group, in lower-SES schools Asian immigrant youth tend to engage in less delinquency than their native peers, and both Asian and Hispanic immigrant youth tend to engage in less substance use.

These results, particularly for Asians, are consistent with a scenario in which immigrant families anticipate the consequences of assimilation and therefore avoid assimilating in disadvantageous social contexts, where these consequences are more likely to be negative. However, this leaves us with an unsatisfying “black box” connecting social context to behavioral assimilation. For the above scenario to be plausible, it is necessary to demonstrate concrete intermediary factors, under the control of immigrant families, that would allow adolescents in disadvantaged contexts to avoid being drawn into risky behaviors. Two such mediating variables were proposed: Parenting behaviors and adolescent friendship choices. Both of these factors contributed to the relationship between school SES and behavioral assimilation for Asian immigrant youth. Asian immigrant adolescents in low-SES schools tend to choose friends with better-than-average risk behavior profiles, and their parents tend to supervise them more closely. For Hispanic youth, however, neither parenting nor friendship helped to explain the relationship between school SES and the substance use gap.

Why would parenting behaviors or friendship choices have stronger effects on risk behavior assimilation for Asians than for Hispanics? The theoretical perspectives discussed earlier give us no reason to suppose that Hispanic families would be any less concerned about the deleterious effects of assimilation on their children than Asian families. On the contrary, past work has shown that Hispanic immigrant parents are often concerned about the influence of American peers on their children’s behavior (Perreira et al., 2006; Portes and Rumbaut, 2001). It is more likely that the explanation lies in other differences between Asian and Hispanic immigrants.

One potentially important difference may lie in the resources Asian and Hispanic families can muster to effectively guide children’s assimilation. As Portes and Rumbaut (2001, p. 105) point out, “Parental setting of rules is not the same as enforcing them since external factors can prevent effective guidance of children.” While there is wide variation by ethnicity among both Hispanics and Asians, Asian communities may in general offer more support to parents seeking to delay their children’s acculturation or assimilation into native peer groups. The relatively high proportion of high-SES immigrants within many Asian ethnic groups leads to greater availability of community cultural institutions to assist parents in maintaining their children’s connections to the culture of origin. For example, Zhou and Kim (2006) describe the prevalence of language schools, which are accessible to working-class and professional immigrants alike, in both Chinese and Korean communities. On the other hand, Portes and Rumbaut (2001, Chapter 5) found that of all the immigrant groups they surveyed, Mexicans and Nicaraguans were among the least likely to report that coethnics in the neighborhood help each other and are supportive of each other. Such lack of community support puts families at risk of “dissonant acculturation”, in which differences between parents’ and children’s pace of acculturation results in intergenerational communication and relationship difficulties and erodes parents’ ability to guide their children (Portes and Rumbaut, 2001). Thus, Asian families may, on average, have more means than Hispanic families to influence their children’s assimilation, and their efforts to do so may be more likely to meet with success.

Another difference between Asians and Hispanics is the history of immigration: because large-scale immigration from most Asian countries did not commence until recently, the third generation for most Asian groups is small, sometimes almost nonexistent. For Hispanics (particularly Mexicans), by contrast, there has been a long-standing, steady flow of immigration to the United States for many generations, leading to a sizeable population of third-plus generation Hispanics. Thus, there is a substantial same-race native peer group in schools for Hispanic immigrants, but not for Asian immigrants. Indeed, of the sampled Add Health respondents, there are approximately equal numbers of first-generation and third-generation Hispanics, while there are nearly three times as many first-generation as third-generation Asians. Adolescents’ tendency to choose friends of the same race, coupled with the lower availability of same-race native peers for Asian immigrants, may make it easier for Asian than for Hispanic parents to discourage their children’ assimilation into native peer groups¹⁰.

Overall, the analysis is more successful at illuminating assimilation patterns among Asian than Hispanic immigrants. While the above factors may explain why parenting and friendship factors operate differently for the two groups, the paper’s inability to explain why there was a significant relationship for Hispanics between school SES and substance use, but not delinquency, is an additional limitation. Both gender and religious participation were explored as factors that may relate differently to substance use and delinquency for Hispanics, but neither appears to play a role in this finding (results available upon request). Another limitation of the analysis is the inability, due to sample size, to conduct separate analyses for different Asian and Hispanic ethnic groups. There is a great deal of diversity among Asian and Hispanic ethnic groups, and the processes reflected in this paper may not apply equally to all subgroups. It is especially important to note that the findings for Hispanics are driven primarily by Mexicans, as they make up a large majority of the Add Health Hispanic immigrant sample. Finally, the analysis would be stronger if it were based on longitudinal data. The ideal design would be to follow a sample of immigrant youth over time as families move between schools. However, due to several key variables being measured by only by the cross-sectional in-school portion of Add Health, a longitudinal design was not possible for this current analysis.

Finally, the issue of selectivity into neighborhoods and schools deserves comment. As discussed above, the current study is based on cross-sectional observational data. In absence of being able to randomly assign respondents to neighborhoods or schools, there is the danger that the results are biased by respondents’ ability to choose their social contexts. Like most other studies of contextual effects based on non-experimental data, this study makes the implicit assumption that such selection occurs only on the basis of observable characteristics, which are then controlled in the regression models. However, this assumption is not testable and may not be realistic. This potential endogeneity of social context would be particularly problematic if immigrant parents’ residential choices are more affected than those of native parents by how well their children are doing. For example, immigrant parents who suspect problematic behavior among their children may make a greater effort to move to a better neighborhood, even in the absence of having greater financial means to do so. This would imply the selection of immigrant children with higher levels of problem behaviors into better schools, which could contribute to their lower behavioral advantage in high-SES schools. While the present study lacks the necessary data to address such possibilities, the potential endogeneity of school SES is an interesting avenue for future exploration.

In conclusion, I emphasize the importance of recognizing immigrant families as active agents in shaping their assimilation pathways. Modern theoretical perspectives highlight the importance of social context for immigrant assimilation. Segmented assimilation theory, in particular, argues that the effects of assimilation on socioeconomic, health, and other outcomes depend on social context. The results of this study affirm that this emphasis on context is justified, but also suggest a significant extension: That social context may not only moderate the effects of assimilation, but may also affect whether and how much immigrant families choose to assimilate in the first place. Future studies of immigrant families should therefore take care to recognize that patterns of assimilation reflect, at least in part, immigrant families’ deliberate adaptations to the surrounding context.

Appendix A: Variable Descriptions and Means

Variable	Variable Description	Mean for Asian immigrants	Mean for Hispanic immigrants	Mean for natives (all races)
Independent Variable
School SES	Percentage of students’ mothers without HS diploma	16.8	22.7	14.1
Parenting variables
Low parental supervision	Binary - Parent usually not home afternoon/evening	0.09	0.07	0.07
No rules about friends	Binary - makes own decisions about friends	0.78	0.72	0.87
Total number of own decisions	Total number of own decisions allowed to make (of 7)	5.10	4.74	5.20
Parent makes decision with	Parent always/often makes decisions with child	0.73	0.73	0.72
Friend Characteristics
Propensity for inter-ethnic friendship	Propensity for choosing inter-ethnic friends, net of school ethnic composition	−0.30	−0.25	−0.08
Friends’ fighting	Average friends’ fighting score from in-school survey	0.52	0.66	0.68
Friends’ substance use	Average friends’ alcohol/tobacco use from in-school survey	1.61	2.05	2.49
Outcomes
Delinquency	Percentile score on age-standardized delinquency scale	45.7	53.2	50.9
Substance Use	Percentile score on age-standardized scale of use of alchohol, tobacco, and marijuana	43.1	46.3	51.4
Relative to non-immigrants
Delinquency	Difference between R’s delinquency and that of average school peers (age & sex adjusted)	−7.1	2.1	0.0
Substance Use	Difference between R’s substance use and that of average school peers (age & sex adjusted)	−7.0	−2.5	0.0
Control Variables
Age	Respondent’s age at Wave 1 interview	16.2	16.2	15.9
Gender	Binary: 1=Female	0.48	0.50	0.50
Parent interview missing	No parent interview	0.33	0.17	0.10
Family Income	Log of family income, imputed for those with missing parent interview	10.4	9.7	10.4
Average parental education	Average of parental education in 2-parent family, co-resident parent’s education in single-parent family	13.6	10.9	13.4
Single parent family	Binary: 1=single parent family, 0 otherwise	0.17	0.27	0.33
Stepparent family	Binary: 1=stepparent family, 0 otherwise	0.08	0.14	0.14
Length of stay for foreign-born	Binary: 1=Foreign-born, > 5 years in U.S.	0.46	0.28	0.00
Immigrant generation	Binary: 1=Born in U.S.	0.40	0.62	1.00
Speaks English at home	Binary: 1=Speaks English at home	0.53	0.36	0.99
School % black	School % Black	0.18	0.15	0.15
School % Asian	School % Asian	0.18	0.07	0.03
School % Hispanic	School % Hispanic	0.26	0.44	0.11

Appendix B: Averages of school SES and outcomes by “traditional” assimilation measures

	Asians			Hispanics
	School SES1	Substance Use	Delinquency	School SES1	Substance Use	Delinquency
First generation	18.1%	41.2	45.4	23.7%	42.1	47.1
Second generation	14.1%	46.2	46.2	21.4%	48.6	57.3
LOS <=5 years	19.9%	38.5	42.2	22.4%	40.4	49.7
LOS> 5 years	17.6%	42.0	46.3	24.1%	42.7	46.2
2 immigrant parents	16.7%	43.1	46.5	23.4%	44.5	50.1
1 immigrant parent	15.5%	43.5	41.6	19.7%	49.9	60.9
Speaks non-English lang at home	17.5%	38.5	43.0	23.6%	44.0	52.0
Speaks English at home	15.7%	47.2	48.1	20.1%	49.8	55.7

¹Percent of students’ mothers who lack a high school diploma