Testing For Measurement Invariance of Attachment Across Chinese and American Adolescent Samples

Abstract

Adolescent attachment to formal and informal institutions has emerged as a major focus of criminological theories since the publication of Hirschi’s work in 1969. This study attempts to examine the psychometric equivalence of the factorial structure of attachment measures across nations reflecting Western and Eastern cultures. Twelve manifest variables are used tapping the concepts of adolescent attachment to parents, school, and neighborhood. Confirmatory factor analysis is used to conduct invariance test across approximately 3,000 Chinese and U.S. adolescents. Results provide strong support for a three-factor model; the multigroup invariance tests reveal mixed results. While the family attachment measure appears invariant between the two samples, significant differences in the coefficients of the factor loadings are detected in the school attachment and neighborhood attachment measures. The results of regression analyses lend support to the predictive validity of three types of attachment. Finally, the limitations of the study are discussed.

Keywords

measurement invariance test family attachment school attachment neighborhood attachment Chinese adolescents American adolescents

Introduction

Adolescent attachment to formal and informal institutions has emerged as a major focus of criminological theories since the late 1960s. Hirschi (1969), for example, treated attachment as one of the four key elements in capturing the theoretical concept of social bonding. He posits that strong attachment serves as a protective factor that helps insulate juveniles from committing deviant behaviors. Similarly, in their integrated theory, Elliott, Ageton, and Canter (1979) suggested that internal and external control such as family bonding produces a significant and inverse relationship with delinquent behavior (also see Elliott, Huizinga, & Ageton, 1985; Menard & Grotpeter, 2011). Although attachment can be conceptualized and measured in a variety of ways, for example, as emotional and psychological connection between juveniles and social institutions (Hirschi, 1969; Popp & Peguero, 2012; Stewart, 2003) and sensitivity to the opinions of others (Taylor, 2001), the pivotal role of attachment in youth development is well recognized throughout the broader fields of social sciences.

A review of the criminological literature suggests that adolescent attachment is commonly defined as a multidimensional concept, including attachment to family and attachment to school, two of the most important conventional institutions in a modern society (Hirschi, 1969; Schreck, Stewart, & Fisher, 2006; L. Zhang & Messner, 1996). Over the course of last four decades, numerous studies have investigated the hypothesized link between these two forms of attachment and delinquent behavior (e.g., Chen, 2009; Schreck et al., 2006; Ward, Boman, & Shayne, 2012; L. Zhang & Messner, 1996). Recently, the application of attachment has been expanded to the studies of victimization (e.g., Popp & Peguero, 2012; Schreck et al., 2006), and the integration between social bonding and self-control (Taylor, 2001; Ward et al., 2012). Outside the United States, social bonding theory has been tested in European countries including the Netherlands (Junger & Marshall, 1997), France (Hartjen & Priyadarsini, 2003), and in Asian countries such as the Philippines (Shoemaker, 1994) and Turkey (Ozbay & Ozcan, 2008). In addition, two studies using Chinese samples exclusively examined the effect of attachment on juvenile delinquency (see Gu, Lai, & Ye, 2011; L. Zhang & Messner, 1996).

Given the intense research attention to attachment in the criminological literature (e.g., Menard & Grotpeter, 2011; Ozbay & Ozcan, 2008; Popp & Peguero, 2012; L. Zhang & Messner, 1996), it is surprising to find that to date no published study has examined the attachment measurement equivalence¹ from a cross-national standpoint. The purpose of this study is to fill this void by comparing the factorial structure of attachment measures between American and Chinese students. China is a society where social attachment is often identified as the cultural foundation and the primary mechanism of internalization of Chinese values (Cao & Hou, 2001; Curran, 1998). Thus, it is important to ask if the attachment measurement derived from the social bonding theory in the United States measures a similar trait in a sample of Chinese youth. To cross-nationally validate the attachment scale, three latent factors are used to tap into the concept of adolescent attachment to conventional institutions in the modern society: attachment to parents, attachment to school, and attachment to neighborhood. Our data were collected via school-based surveys in the United States and in China totaling more than 3,000 students. Confirmatory factor analysis (CFA) is utilized to assess if the observed items are invariant between the two samples. In addition, regression analyses are used for both samples to further test the respective predictive validity of the attachment measures in accounting for self-reported delinquent behavior, low self-control, and pro-violence attitudes.

To our knowledge, this study is the first of its kind that uses a multisample design to investigate the psychometric equivalence of attachments across nations. This study endeavors to make at least two important contributions to the literature on social bonding. First, as the United States and China are known to have distinctively different cultural and historical traditions (e.g., individualism vs. collectivism) and socioeducational dynamics (e.g., learning process that cultivates innovative ability vs. learning process that rewards mechanical repetitions), attention should be directed to examine if a best-fitting factorial model of attachment derived from the U.S. sample can be found invariant in the Chinese sample. If not, what are the observed items that contribute to the noninvariance between two samples? Making use of the group invariant analysis enables us to identify the items by examining two distinctive components: coefficients of item loadings and factor covariance. Second, the importance of neighborhood dynamics has not been a major part of social attachment in social bonding theory. Although Hirschi (1969) included survey questions about one’s neighborhood, it has rarely been incorporated into research on attachment. Traditionally, neighborhood was not a major factor that attracts the attention of criminological theories focusing on individual behaviors. The resurgence of social disorganization theory and collective efficacy once again rekindles the interest of researchers in studying the impact of neighborhood environment on criminal behavior (e.g., Sampson, Raudenbush, & Earls, 1997). In this study, attachment to neighborhood is included in the analysis as one of the three primary factors.

Literature Review

In Western philosophy, an individual’s attachment is considered a virtual cornerstone of human society. Aristotle famously argued that “man is by nature a social animal” and human interactions and attachments initiate with the first layer of society, attachment to one’s family. The heritage of Western political philosophy features a variety of propositions regarding different ways that individuals can become attached to their society. Attachment to others through belief in God was emphasized by St. Augustine; economically derived and class-oriented attachment was emphasized by Karl Marx; and rational fear-induced commitment to protective human relationships was emphasized by Thomas Hobbs. Early criminological theories accordingly reflect an emphasis upon individual attachment to one’s social environment. For example, based on empirical data Durkheim (1951) observed that there was an inverse association between suicide rates and social integration as related to attachment to three particular institutions—the prevailing religion, “domestic” society, and political society (p. 209). He echoed Aristotle’s observation that humans are social animals and attachment to social institutions is a critical factor in accounting for deviant behaviors. Subsequent theories of delinquency remain focused on the individual’s attachment to institutions as an effective means of curbing deviant behaviors. Anomie, a concept of alienation and detachment from traditional societal norms, is considered to be associated with the acceptance of deviant norms and activities (Merton, 1938). Besides the philosophical discussion on the importance of social attachment as a key to understanding human interactions, systematic conceptualization and operationalization of the concept “attachment” in the United States began with Hirschi’s work some 45 years ago.

Juvenile attachment can be defined as an individual’s affective bonding to an institution in his or her interactions with society (Hirschi, 1969). For adolescents, the principal attachments of interest include the family, the neighborhood wherein he or she interacts with neighbors and plays with friends, and the school where he or she acquires knowledge, learns skills, and associates with the peers (e.g., Hirschi, 1969). Among the advocates of social bonding theory, the extent of attachment is held to be the most important reason for explaining why rule-compliant adolescents refrain from delinquent activities (Hirschi, 1969). More specifically, Hirschi (1969) identified four essential elements of social control and among them attachment to family and attachment to school are at the core of his theory. This phenomenon of personal attachment/bonding refers to a juvenile’s psychological affection for positive role model and sensitivity to prosocial others, such as those coming from one’s family, close friend circle, and school.

The first dimension of attachment included in most studies is family attachment. It is frequently measured with a focus on self-reported relationship with father and mother (Menard & Grotpeter, 2011; Song, Thompson, & Ferrer, 2009). For example, Chen (2009) measured family attachment with three items: whether an adolescent believed his or her mother or father was warm and loving and could be trusted by sharing personal secrets with them. Schreck et al. (2006) measured parental attachment as effective communication between parents and child in terms of frequency and quality of exchanges. Recently, using the second wave of the International Self-Report Delinquency (ISRD-2) data collected from 30 countries, Posick and Rocque (2015) tested the utility of family bonding as a determinant of school youth victimization. Their family bonding variable is comprised of four items, including closeness to mother and father, leisure time with the family, and having dinner with the family. Either from the standpoint of overall assessment of parents/child relationship or the quality of communication, the measure of family attachment attempts to assess if there is an emotional, affectionate, and rather exclusive bonding between adolescents and their parents (Chen, 2009; Hirschi, 1969).

Similarly, school attachment reflects an adolescent’s affiliation and emotional connection with school where a student learns conventional knowledge. Similar to self-identification with parents, the measure of school attachment generally includes students’ rating of school environment, their performance, and relationship with their teachers. L. Zhang and Messner (1996), for example, used an index to gauge the concept of school attachment, including perceptions of teachers, attitudes toward the school, and self-ratings of academic ability. Popp and Peguero (2012, p. 8) measured school attachment with a particular focus on the relationship between students and teachers and used items such as “student gets along well with teachers,” “the teaching is good,” and “I go to school because my teachers expect me to succeed” (also see Ozbay & Ozcan, 2008).

Interestingly, attachment to one’s neighborhood has been largely overlooked in the social bonding research. Even a cursory review of the literature reveals that almost all studies operationalize attachment as attachment to family and/or attachment to school. However, the revitalization of research on neighborhood, a key context for human interactions, leads us to suggest that attachment to one’s neighborhood is an indispensable component of adolescent development and socialization. Neighborhood-oriented research comes from a different source of intellectual thought; place-based criminology is often traced back to the Chicago School of Social Ecology and the related social disorganization theory. This deeply researched “Chicago School” conceptualization of contextual influence on individual criminal behavior and juvenile delinquency in criminological research derives from the study of human ecology (Park, 1952). The theory was developed from the now-classic studies of urban crime and delinquency conducted by Shaw and McKay in the Chicago metropolitan area over the period from 1900 to the 1930s (Shaw & McKay, 1942). In more recent years, the Broken Windows thesis reignited the tradition of location-oriented research and has provided the key theoretical link between geographic location, resident alienation, and the incidence of disorder (Wilson & Kelling, 1982). Similarly, the concept of community collective efficacy highlighted the importance of neighborhood attachment in the 1990s; this concept was originally developed by Sampson et al., (1997) based on their study of differential crime and delinquency rates across Chicago neighborhoods. The concept in question reflects the “willingness and intentions to intervene on behalf of the neighborhood would be enhanced under conditions of mutual trust and cohesion.” Collective efficacy has been used in a considerable number of subsequent studies (e.g., Burchfield, 2009; Gibson, Zhao, Lovrich, & Gaffney, 2002; Silver & Miller, 2004; Woldoff, 2002). For example, Ren, Cao, Lovrich, and Gaffney (2005) found that collective efficacy (also referred to as collective security) was a robust predictor of citizen confidence in the police (also see Huebner, Schafer, & Bynum, 2004; Lai, Cao, & Zhao, 2010). In this regard, Silver and Miller (2004) noted that “people who feel a strong sense of attachment to their neighborhoods are likely to also feel a greater sense of responsibility for maintaining order within them . . . ” (p. 557). In the research literature on neighborhood attachment, two related dimensions are explored—namely, attitudinal attachment and behavioral attachment. Attitudinal attachment reflects the degree to which residents are satisfied with their neighborhoods as a place to live (with respect to neighbors, amenities, access to transportation, etc.; Burchfield, 2009; Gibson et al., 2002; Woldoff, 2002). Behavioral attachment, in contrast, is related to residents’ willingness to intervene in an event (e.g., loud party) on behalf of neighbors and/or neighborhood quality of life (a reflection of collective efficacy; Silver & Miller, 2004). We believe that attachment to neighborhood can also be applied to juveniles who spend much of their time playing and socializing in their neighborhoods. Similar to other attachments, adolescent attachment to one’s neighborhood is viewed as an integral part of social attachment.

Research Questions

Despite the fact that attachments to family and school have been empirically tested extensively in the United States and other countries, there remains an unanswered question: Do adolescents of different countries perceive attachments similarly or differently? It is reasonable and logical to assume that significant differences may exist, given the cultural traditions and social environments between the United States and China. In this regard, Jiao (2001) pointed out that the Chinese cultural context is rooted in moral order, communitarianism, and collectivism (also see Jiang & Lambert, 2009). In contrast, American culture extols social order maintained through law, private property, and individualism. More specifically, Confucius proposed an “ideal type” society that is governed by broadly shared virtues and a ubiquitous set of the group-based moral principles, the violation of which brings on a deep sense of individual shame (Anderson & Gil, 1998). To learn and internalize the group-based moral principles, Confucius placed particular emphasis upon a learning process oriented toward social attachment starting at an early age within families and later entailing progressively extensive social institutions through adolescence and into one’s adulthood. Chen (2002) observed concerning this process that “people first learn to behave in response to the family’s needs; this helps set the stage for behaving in terms of school’s needs, the work unit’s needs, and the neighborhood’s needs” (p. 49). Consequently, social attachment to one’s family, neighbors, work unit, and school is deemed to be vital to the process of internalizing values of informal self-control to maintain social harmony (Lai et al., 2010; Shaw, 1996). This informal and self-regulated mechanism of unwritten rules of appropriate social conduct continues to serve as a model of human conduct for Chinese people (Chen, 2004).

A review of the literature suggests that the concept of social attachment and its basic three types have a long history in both the United States and China. Thus, the first research question is to assess the three-factor model of attachment in both Chinese and U.S. adolescent samples. Simply put, we ask, does a three-factor model of attachment have adequate fit with the observed data in both samples? The second research question concerns the measurement equivalence of attachment instrument across U.S. and Chinese students. Do factors perceived by American students and Chinese students remain invariant across the two groups? If nonequivalence is detected, which item(s) perceived by the students is (are) significantly noninvariant across the two samples? Finally, as a natural extension of the invariance test, this study is to assess the utility of the attachment measures in predicting juvenile delinquency, low self-control, and violent attitudes across both samples.

Method

Sample and Data Collection

The target population in China and the United States in the current study was seventh to ninth graders, aged 12 to 15. The research site for the Chinese sample is Hangzhou, the capital city of Zhejiang province, about 150 miles southwest of Shanghai. Hangzhou has been a rapidly growing city and according to the city official website (http://eng.hangzhou.gov.cn/), the population of long-term residents (not including the migrant population) in urban districts was 5.52 million based on census estimate in 2008. The city is a vivid reflection of the social and demographic changes in the coastal area in China where the economic boom has been most noticeable. Hangzhou is an ideal site to examine social attachment, given its rapid economic growth as well as a large influx of migrant workers from less prosperous areas in China.

Due to the large student population and complexity of its demographics in Hangzhou, a multistage cluster sampling technique was used for the sample selection. As a result, nine middle schools that were located in the five urban core districts in the city were selected for the study. In each of the selected schools, one class was randomly selected from the seventh to ninth grades, respectively. In collaboration with the Zhejiang Provincial Juvenile Delinquency Institute (ZPJDI), we have gained access to all these nine schools selected for the sample, making a 100% overall school participation rate. The scenario of a perfect participation rate we obtained from these schools in Hangzhou was also found in five ISRD-2 participating nations, including Armenia, Bosnia and Herzegovina, Lithuania, Finland, and Surinam (Marshall, 2010). Data were gathered in late December 2009 and early January 2010 by means of anonymous, self-report questionnaires (paper-and-pencil) administered during one classroom period. The Chinese school year is different from that of the United States; the winter break usually starts in the middle of January before the Chinese Lunar New Year. Researchers worked closely with the trained members of the research staff from ZPJDI for data collection. At least two researchers were present in one classroom to administer the surveys and teachers, and school administrators were asked to leave before the questionnaires were distributed.

The U.S. data came from students in 11 public and 4 private schools in five cities from three geographically and socioeconomically diverse regions (Northeast, Midwest, and Southwest).² The U.S. version of ISRD-2 was sponsored by the National Institute of Justice (NIJ) as a part of an international collaborative effort to repeat the ISRD in more than 30 countries.³ The U.S. study followed the ISRD-2 city-based sampling protocol as much as possible, but some modifications from the standardized design had to be made. The initial sample design for the United States was a city-based, purposeful sampling plan, requiring the selection of one large city, one medium-sized city, and three small towns. Researchers selected the initial research sites based on geographic location representing different levels of socioeconomic development (a large city in the Southwest [Texas], a medium city in the Midwest [Illinois], and a cluster of small towns in the Northeast [Massachusetts and New Hampshire] and logistic considerations, that is, presumed access to schools and available research assistance). Because of refusal to participate by some of the originally selected school districts, the medium-sized city and small towns were replaced by closely matched and comparable selections in the same geographic area and a revised sampling plan was developed. As a result, one large city, one medium-sized city, and a sample from a cluster of small townships were included in the sample selection.⁴ Paper-and-pencil surveys in the classroom settings were carried out in the fall of 2006 and spring of 2007. Questionnaires were filled out by 96% of the Chinese sample and by 64% of the U.S. sample, resulting in 1,043 and 2,401 usable surveys, respectively.⁵

Survey Instrument and Measures

The ISRD-2 survey instrument was utilized for both Chinese and American samples. The survey instrument taps a wide range of delinquent behaviors from serious crimes to minor delinquent or risky behaviors. In addition, many items were included in the ISRD-2 to gauge the extent of social attachment, school and neighborhood contexts, self-control, and attitudes toward violence among participants. The validity and reliability of the ISRD core questionnaire (e.g., self-reported offenses) have been examined and found to be quite satisfactory (Bruinsma, 1994; Marshall & Webb, 1990, 1994; S. Zhang, Benson, & Deng, 2000). Similarly, attention has been paid to the study on psychometric efficiency of the attitudinal measures of the ISRD instrument. For example, Marshall and Enzmann (2012) examined the measurement reliability, validity, and dimensionality of the concept of self-control and its application cross-nationally. The English instrument of ISRD-2 was translated into Mandarin by the first three authors who are Chinese natives and criminal justice faculty in the U.S. universities. They have had an extensive experience with translation from English to Chinese or vice versa and are heavily involved in survey-based research projects in China. Then, the Chinese version of the survey instrument was pretested among the 16 Chinese exchange students in the university where the first two authors teach to make the questionnaire better fit the Chinese social, cultural, and language contexts.

In total, 12 manifest items were used to measure three components of social attachment; similar measures tapping into these three key components of social attachment have been reported in the relevant literature (e.g., Lu, Yu, Ren, & Marshall, 2013; Posick & Rocque, 2015; H. Zhang, Zhao, Zhao, & Ren, 2014). Family attachment was captured by the three items measuring closeness to mother and father and leisure time with the family. The respondents were first asked to rate the following two items on a 4-point Likert-type scale ranging from 1 = unable to get along to 4 = get along very well: “P1: How do you usually get along with the man you live with (father, stepfather . . . )?” “P2: How do you usually get along with the woman you live with (mother, stepmother . . .)?” The third question of family attachment concerns leisure time with the family by asking “P3: How often do you and your parents (or the adults you live with) do something together, such as going to the movies, going for a walk or hike, visiting relatives, attending a sporting event, and things like that?” This item structures on a 6-point scale, including almost never (1), once a year (2), few times a year (3), once a month (4), once a week (5), and more than once a week (6).⁶ The concept of school attachment was gauged by four items on a 4-point Likert-type scale (1 = not at all true, 2 = not true, 3 = true, 4 = very true): “S1: If I had to move, I would miss my school,” “S2: Teachers do notice when I am doing well and let me know,” “S3: I like my school,” and “S4: There are other activities in school besides lessons.” Neighborhood attachment was measured by five items led by the questions: “N1: My neighbors notice when I am misbehaving and let me know,” “N2: I like my neighborhood,” “N3: People around here are willing to help their neighbors,” “N4: This is a close-knit neighborhood,” and “N5: People in this neighborhood can be trusted.” Response categories range from 1 (not at all true) to 4 (very true). Ambiguous answers or no answers to the 12 social attachment items were treated as missing values for both samples and removed from the original datasets. Listwise deletion resulted in 56 and 307 missing cases for the Chinese and the U.S. samples, respectively. The results of the t tests on the demographic variables (e.g., age, gender, grade, intact family) show no significant differences between the valid sample and the missing cases. This suggests that the missing values for the study variables are random in nature. The values of Cronbach’s α for three types of attachment are reported in Table 2.

Analytical Procedure

CFA is used to test for equivalence of the factorial structure of attachment items across the Chinese and the U.S. middle school students. CFA enables the researchers to assess whether a given factor structure has an adequate fit with the observed data by postulating patterns of relations a priori and testing this hypothesized structure statistically (Ang, Chong, Huan, Quek, & Yeo, 2008). We specifically followed the analytical procedures of multigroup invariance test across independent samples outlined by Byrne (2012). This type of test is suitable for applications involving two or more samples where the central attention is whether the components of the measurement model operate invariantly across particular groups of interest (Byrne, 2012).

According to Byrne (2012), testing the factorial equivalence of attachment measures involves the following steps. First, the manifest/observed variables are analyzed for each group separately and the purpose is to establish a best-fitting baseline model for each group based on the theoretical construct and the satisfactory goodness-of-fit statistics. Second, both groups are combined to produce a basic measurement model with no equality constraints imposed on parameters. The same number of factors and the factor loading pattern derived from the baseline model will be estimated again simultaneously in the multigroup model. Next, factor loadings are constrained to be equal across groups. The full constrained model will be compared against the basic measurement model. Significant outcome between two models suggests the noninvariance across groups. In seeking evidence of noninvariance, the factor loadings of the items appearing to be the most divergent across groups will be identified.⁷ Then, the equality constraint imposed on this item previously will be relaxed/removed. This partial invariance model will be compared against the basic measurement model to check if difference in model fit between these two models is statistically significant. A significant diagnosis suggests the necessity of freeing additional parameters. The purpose of this procedure is to yield a partial invariance model that is not statistically different from the basic measurement model. By repeating this procedure, the items whose factor loadings are significantly noninvariant across groups will be successfully identified and freely estimated; all items on attachment, except for the identified items, are operating equivalently across two groups of students. Finally, factor covariances are constrained to be equal across groups.

In this study, analyses are conducted by using Mplus version 6. As analysis of the 12 manifest variables revealed evidence of slight nonnormality, all analyses were conducted by using a Mplus estimation option (MLM) for maximum likelihood estimation with standard errors and a mean-adjusted chi-square test statistic that are robust to nonnormality (Muthén & Muthén, 1998-2012). A variety of absolute and relative (or incremental) indices were consulted to assess model fit. The absolute fit index includes chi-square (χ²) statistics, where χ² is the likelihood ratio statistic used to test whether a given model provides an acceptable fit to relevant observed data. The criteria for χ²/df ratio range from as low as 2.0 (Barrett, 2007; Tabachnick & Fidell, 2007), to 3 (Kline, 2005), to 4 (Hu & Bentler, 1999), and to 5 (Wheaton, Muthen, Alwin, & Summers, 1977). At the center of the argument regarding which model fit statistic is the best indicator is that chi-square values are sensitive to sample size, and its use leads to the rejection of virtually any model derived from a large sample (Bentler, 2007; Miles & Shevlin, 2007). Generally, the accepted rule is that χ²/df needs to be below 5 for a large sample.⁸ For the purpose of comparison between models, a chi-square difference test is needed to examine whether difference in model fit between partial invariance models and the basic measurement model is statistically significant. To determine this information, given that analyses were based on MLM estimation, requires the Satorra–Bentler scaled chi-square difference test, where the usual normal-theory chi-square statistic is divided by a scaling correction to better approximate chi-square under nonnormality (Byrne, 2012; Satorra & Bentler, 2010).⁹ Another absolute fit indicator, “one of the most informative fit indices” according to Diamantopoulos and Siguaw (2000, p. 85), is the root means square error of approximation (RMSEA), which takes the error of population approximation and degrees of freedom into account and characterizes the lack of fit of the hypothesized model to the population covariance matrix. In more recent studies, the cutoff points of RMSEA have been reduced to be below 0.06 (Hu & Bentler, 1999; also see Steiger, 2007) to constitute a good fit.

Among incremental fit indices or approximate fit indices are a group of measures that do not rely upon the chi-square in its raw form, but rather compare the chi-square value with the null or independence model (Hooper, Coughlan, & Mullen, 2008). These indices attempt to answer the question “How well is my model doing, compared with the worst model that there is?” (Miles & Shevlin, 2007, p. 870). Two of these indices are commonly used in CFA. The comparative fit index (CFI) is the most commonly used and it assesses “the fit of a user-specified solution in relation to a more restricted, nested baseline model” in which the “covariances among all input indicators are fixed to zero” positing no relationship among variables (Brown, 2006, p. 84). The CFI statistic ranges from 0 to 1.00, with values greater than 0.95, indicating a reasonably good fit between the hypothesized model and the empirical data (Hu & Bentler, 1999). The Tucker–Lewis Index (TLI) or nonnormed fit index is the other measure suggested by Jöreskog and Sörbom (1989) for assessing a model’s overall fit; it is based on a ratio of the squared sum of discrepancies to the observed variances. A TLI value around 0.95 or larger indicates a good fit (Hu & Bentler, 1999; Loehlin, 1992; Schumacker & Lomax, 2004).¹⁰

The final step in the analysis involves the ordinary least squares (OLS) regressions to test the predictive capacity of three types of attachment in both the Chinese and the U.S. samples. Relevant research has shown that social attachment is a vital predictor in delinquent behavior (e.g., Y. Zhang, Day, & Cao, 2012), low self-control (e.g., Kort-Butler, Tyler, & Melander, 2011), and attitudes toward violence (e.g., Harris, 2009). In the current study, these outcome variables are all measured as multiple-item scales (see Appendix B for the measurement).

Findings

Sample Characteristics

The demographic characteristics of two samples are reported in Table 1.¹¹ In both samples, a little more than half of the participants were male (52.5% in the Chinese sample and 52.3% in the U.S. sample) and the vast majority of the subjects fell within the age range of 12 to 15 (92.3% in China and 95.7% in the United States). The average age is almost identical for both samples (13.99 for the Chinese sample and 13.95 for the U.S. sample). While the distribution of the class grades was relatively even in the Chinese sample (seventh grade = 31.8%, eighth grade = 32.2%, and ninth grade = 35.9%), ninth graders accounted for half of the respondents (49.8%) and about one fourth were in Grades 7 (25.2%) and 8 (25%) in the U.S. sample. The vast majority of the students came from intact family (84.2% in the Chinese sample and 73.8% in the U.S. sample).¹²

Table 1.

Comparison of Demographics Between the Chinese and U.S. Samples.

Variables	Chinese sample (n = 1,043)		U.S. sample (n = 2,401)
Variables	Valid n (%)	M (SD)	Valid n (%)	M (SD)
Gender
Female (0)	494 (47.5)		1,144 (47.7)
Male (1)	547 (52.5)		1,253 (52.3)
Age
12	86 (8.3)	13.99 (1.10)	285 (11.9)	13.95 (1.11)
13	286 (27.5)		532 (22.2)
14	306 (29.4)		719 (30.0)
15	282 (27.1)		756 (31.6)
16 +	81 (7.8)		103 (4.3)
Grade
Seventh	332 (31.8)		606 (25.2)
Eighth	337 (32.3)		600 (25.0)
Ninth	374 (35.9)		1,195 (49.8)
Intact family
Nonintact family (0)	164 (15.8)		626 (26.2)
Intact family (1)	874 (84.2)		1,765 (73.8)

As to the descriptive statistics of the 12 observed items, in both samples, the 3 items for family attachment hold the highest mean values, followed by the school attachment and neighborhood attachment items. On average, the Chinese adolescents in the sample gave higher ratings for 11 of 12 items than their American counterparts. The only item that American students rated higher is the fourth item for school attachment. That is, there are other activities in school besides lessons (see Table 2). To examine the mean differences between the two samples, t tests were conducted. In spite of the significant differences in the mean ratings in all the 12 items except for N2 across two samples, it is difficult to determine whether or not such differences result from true mean differences or from differences in the functioning of the measures across cultures. This warrants the psychometric invariance test across two samples.

Table 2.

Descriptive Statistics for the 12 Manifest Attachment Variables.

Items	Minimum	Maximum	Chinese sample		U.S. sample
Items	Minimum	Maximum	M	SD	M	SD
Family/parental attachment
P1: How do you usually get along with the man you live with (father, stepfather . . . )?***	1.00	4.00	3.623	0.606	3.523	0.748
P2: How do you usually get along with the woman you live with (mother, stepmother . . .)?***	1.00	4.00	3.711	0.524	3.612	0.673
P3: How often do you and your parents (or the adults you live with) do something together?***	1.00	6.00	4.834	1.392	4.568	1.445
			Cronbach’s α = .612		Cronbach’s α = .515
School attachment
S1: If I had to move, I would miss my school.***	1.00	4.00	3.419	0.847	3.247	0.959
S2: Teachers do notice when I am doing well and let me know.***	1.00	4.00	3.247	0.846	3.029	0.929
S3: I like my school.***	1.00	4.00	3.270	0.830	2.927	0.988
S4: There are other activities in school besides lessons.***	1.00	4.00	3.349	0.930	3.472	0.899
			Cronbach’s α = .750		Cronbach’s α = .635
Neighborhood attachment
N1: My neighbors notice when I am misbehaving and let me know.***	1.00	4.00	2.726	1.146	1.830	1.020
N2: I like my neighborhood.	1.00	4.00	3.127	0.988	3.121	1.002
N3: People around here are willing to help their neighbors.***	1.00	4.00	3.255	0.936	3.089	0.963
N4: This is a close-knit neighborhood.***	1.00	4.00	3.153	0.970	2.618	1.067
N5: People in this neighborhood can be trusted.***	1.00	4.00	3.248	0.901	2.962	1.020
			Cronbach’s α = .898		Cronbach’s α = .770

p < .05. **p < .01. ***p < .001.

Preliminary Single-Group Analyses

The first step of the invariance test is to estimate a best model with no constraints imposed for each sample. Therefore, a best-fitting baseline model was established separately for both Chinese and American adolescent samples to examine whether the three-factor measurement model of attachment fits the observed data for each sample. The baseline models are presented schematically in Figure 1. It is important to note that the CFA practices allow error terms of observed items to correlate within the same latent factor to produce a better model fit when the relationship between the two items is close (Albright & Park, 2009; Byrne, 2012). Upon reviewing the modification indices and expected parameter change statistics for both the Chinese and the U.S. models, the residual covariance between the first two neighborhood attachment items (i.e., N1 and N2) was the most seriously misfitting parameters. Thus, a pair of error terms between N1 and N2 was allowed to be correlated for both the Chinese and American models (see Figure 1). Estimation of this respecified model, for each adolescent group, yielded more fit statistics that were significantly improved from the previous model without the pair.¹³

Figure 1.

Baseline models of attachment structure for the Chinese and U.S. samples.

All the manifest variables are significantly loaded onto their respective targeted factors (i.e., family attachment, school attachment, and neighborhood attachment), indicating a solid link between the relevant theoretical concepts and the observed variables.¹⁴ The covariance values among the three factors were all small, well below the recommended cutoff criteria of .85 (e.g., Gomez, Burns, Walsh, & Hafetz, 2005), indicating the relative independence of three factors. Overall, the goodness-of-fit indices for both models meet all the criteria of CFA, suggesting that the theoretical three-factor model fits the empirical data well. For example, in the Chinese model, the chi-square/degree of freedom ratio is 2.460, well below the conventional limit of 5 for a large sample. The CFI and TLI are 0.981 and 0.976, respectively; both are above the recommended level of 0.95. Finally, the RMSEA is only 0.037, well below the conventional cutoff point of 0.06. Likewise, the overall fit of the model for the U.S. sample was acceptable, with the chi-square/degree of freedom ratio being 3.743, CFI and TLI being 0.978 and 0.971, respectively, and the RMSEA being 0.034. Assessment of the goodness-of-fit indices suggested that the three-factor measurement model is a good-fitting model for both the Chinese and the U.S. samples.

Testing for Invariance Across Two Samples

The best-fitting baseline models provide the foundation against which we test a series of increasingly stringent CFA models related to attachment factorial structure across two samples. A total number of 8 multigroup CFA models were conducted by using the combined Chinese and American data simultaneously, including (a) one basic measurement model with no constraints, (b) one invariance model with full constraints, and (c) six partial invariance models. A summary of all the goodness-of-fit results, in addition to the ΔMLMχ² values, is reported in Table 3. As noted previously, a basic measurement model (Model 1) was first established with no equality constraints imposed on any of the parameters. Based on the model fit indices, this model provides a good fit to the data (MLMχ²/df = 3.076, CFI = 0.979, TLI = 0.973, RMSEA = 0.035). Subsequently, all free factor loadings were constrained to be equal across both samples in Model 2. Although the fit of this multigroup model was considered reasonable, the significant difference detected by the chi-square difference test between the constrained model (Model 2) and the unconstrained basic model (Model 1) indicated that the factor loadings were not equivalent across both samples (ΔMLMχ² =138.226, Δdf = 9, p < .001). Thus, we expect to detect some evidence of noninvariance related to the factor loadings of observed variables across the two samples. An examination of the modification index (MI) values suggested that the constraint on the item N1 be released to improve overall model fit and to reduce nonequivalence between two models. This item N1 is a measure of neighborhood attachment; specifically, the item relates to the neighbors taking responsibility of notifying the respondent when he or she is misbehaving. Accordingly, Model 3 was conducted to remove the constraint on N1 while keeping the constraints on all other items. The results show that there is a stronger relationship between this item and neighborhood attachment for the Chinese sample compared with the U.S. sample (factor loadings: 1 vs. 0.50). However, the MLM chi-square difference test between Model 3 and Model 1 showed a significant difference, suggesting noninvariance across both samples (ΔMLMχ² = 57.762, Δdf = 8, p < .001).

Table 3.

Tests for Invariance of Perceived Social Attachment Across U.S. and Chinese Samples: Summary of Model Fit and χ² Difference Test Statistics.

Model	MLMχ²	df	MLMχ²/df	CFI	TLI	RMSEA
M1: Basic measurement model with no equality constraints imposed	307.623	100	3.076	0.979	0.973	0.035
M2: All factor loadings constrained	474.201	109	4.350	0.964	0.956	0.044
M3: All factor loadings constrained except for N1	381.212	108	3.530	0.973	0.967	0.038
M4: All factor loadings constrained except for N1 and N5	351.547	107	3.285	0.976	0.970	0.036
M5: All factor loadings constrained except for N1, N5, and N4	340.601	106	3.213	0.977	0.971	0.036
M6: All factor loadings constrained except for N1, N5, N4, and S2	325.670	105	3.102	0.978	0.972	0.035
M7: All factor loadings constrained except for N1, N5, N4, S2, and S4	313.275	104	3.012	0.979	0.973	0.034
M8: All factor loadings constrained except for N1, N5, N4, S2, and S4;	316.808	107	2.961	0.979	0.974	0.034
Factor covariance constrained
Model comparison
M2 vs. M1, ΔMLMχ² = 138.226, Δ df = 9, p < .001
M3 vs. M1, ΔMLMχ² = 57.762, Δ df = 8, p < .001
M4 vs. M1, ΔMLMχ² = 33.788, Δ df = 7 p < .001
M5 vs. M1, ΔMLMχ² = 24.715, Δ df = 6, p < .001
M6 vs. M1, ΔMLMχ² = 13.468, Δ df = 5, p > .05
M7 vs. M1, ΔMLMχ² = 4.111, Δ df = 4, p > .05 (ns)
M8 vs. M7, ΔMLMχ² = 7.815, Δ df = 3, p > .05 (ns)

Note. All MLMχ² are statistically significant at p < .0000. MLM = Mplus Estimation Option; CFI = Comparative Fit Index; TLI = Tucker–Lewis Index; RMSEA = Root Mean Square Error Approximation; ns = not significant.

Similar procedures were repeated until the insignificance between the partial invariance model and Model 1 was reached. As a result, four additional models were estimated and the constraints for four items were removed step by step. In Model 4, the nonequivalent item (N5) tapped into neighborhood attachment; specifically, the item is about the trustworthiness of the neighbors. The factor loading of this item in neighborhood attachment is stronger in the U.S. sample than that in the Chinese sample (1.28 vs. 1.06). In Model 5, the nonequivalent item identified was N4 (i.e., this is a close-knit neighborhood). There is a stronger relationship between this item and neighborhood attachment for the Chinese sample in comparison with the U.S. sample (1.28 vs. 1.14). In Models 6 and 7, the constraints for two items of school attachment were removed. These two items are S2 (teachers do notice when I am doing well and let me know) and S4 (there are other activities in school besides lessons). The estimate of factor loadings suggested a stronger relationship between both items and school attachment for the Chinese sample than the U.S. sample (S2: 0.90 vs. 0.66; S4: 0.90 vs. 0.66). The release of both constraints improved model fit, and the chi-square difference test indicates invariance between the partially invariance model (M7) and the unconstrained basic measurement model (M1; ΔMLMχ² = 4.111, Δdf = 4, p > .05).

Finally, the factor covariances of three factors were constrained in addition to the factor loadings of five items to test for the invariance of structural parameters. Goodness-of-fit results of Model 8 were MLMχ²/df = 2.961, CFI = 0.979, TLI = 0.974, and RMSEA = 0.034. Comparison with Model 7 yielded a corrected difference value that was not statistically significant (ΔMLMχ² = 7.815, Δdf = 3, p > .05). This information conveys the notion that, despite the presence of five noninvariant factor loadings, the factor covariances remain equivalent across the Chinese and U.S. adolescents.

Predictive Validity of the Three-Factor Model

OLS regression analyses were conducted to assess the predictive power of three-factor attachment on four-outcome scales: (a) last-year versatility/prevalence, (b) lifetime versatility/prevalence, (c) low self-control, and (d) pro-violence attitude (for descriptive statistics of the scales and the demographic variables included in the OLS regressions, see Appendix A and Table 1, respectively). Prior to running the regression analysis, a correlation matrix was examined for each sample separately. No bivariate correlations among the independent variables were high enough to raise the concern for multicollinearity. A test of variance inflation factor (VIF) was run for each sample, and no score was found to be higher than 1.47 and 1.55 for the Chinese and the U.S. samples, respectively. The results of regression analyses for the Chinese sample are shown in Table 4. All four equations are statistically significant, and the R squared scores range from a low of .053 to a high of .118. All three types of attachment were found to be inversely associated with the four outcome variables, in line with the expected direction. Among the demographic variables, gender is the only predictor that is statistically significant across the board, except for low self-control. Males are more likely than their female counterparts to commit delinquent acts, have low levels of self-control, and have pro-violence attitudes. Similarly, compared with the reference group Grade 7, Grade 9 is a statistically significant predictor of low self-control and pro-violence attitudes.

Table 4.

Ordinary Least Squares Regression Results for the Chinese Sample.

Predictor	Last-year versatility		Lifetime versatility		Low self-control		Pro-violence attitude
Predictor	b (SE)	β	b (SE)	β	b (SE)	β	b (SE)	β
Attachment variables
Family attachment	−.120**	−.097	−.264***	−.131	−.334***	−.118	−.113**	−.088
Family attachment	(0.041)		(0.066)		(0.090)		(0.041)
School attachment	−.113***	−.125	−.204***	−.136	−.202**	−.095	−.137***	−.142
School attachment	(0.032)		(0.052)		(0.073)		(0.033)
Neighborhood attachment	−.053**	−.093	−.096**	−.102	−.248***	−.184	−.065**	−.108
Neighborhood attachment	(0.020)		(0.033)		(0.046)		(0.021)
Demographical variables
Gender	.445**	.094	1.101***	.141	.096	.009	.563***	.112
Gender	(0.150)		(0.241)		(0.339)		(0.155)
Grade 8	−.164	−.032	.009	.001	.360	.030	.287	.053
Grade 8	(0.187)		(0.301)		(0.421)		(0.192)
Grade 9	−.246	−.050	−.020	−.003	1.557***	.133	.507*	.096
Grade 9	(0.189)		(0.305)		(0.428)		(0.196)
Intact family	−.186	−.028	−.419	−.038	−.851	−.054	−.427	−.060
	(0.215)		(0.346)		(0.482)		(0.221)
	F = 8.616***		F = 14.096***		F = 19.376***		F = 14.980***
	N = 962		N = 962		N = 960		N = 971
	R² = .053		R² = .087		R² = .118		R² = .092

p < .05. **p < .01. ***p < .001.

The results of the regression analyses for the U.S. sample are reported in Table 5. Similar to the results derived from the Chinese sample, four regression models are all statistically significant. The values of R squared measures are all above .10, indicating higher levels of predictive power of the attachments in the U.S. sample. While family attachment and school attachment are both significant predictors of four outcome variables, neighborhood attachment in the U.S. sample has no utility in explaining last-year or lifetime versatility. At the same time, neighborhood attachment is negatively associated with two attitudinal measures―low self-control and violent attitudes. With respect to the demographic variables, gender remains an important predictor, whereas intact family was inversely related to the four outcome variables.

Table 5.

Ordinary Least Squares Regression Results for the U.S. Sample.

Predictor	Last-year versatility		Lifetime versatility		Low self-control		Pro-violence attitude
Predictor	b (SE)	β	b (SE)	β	b (SE)	β	b (SE)	β
Attachment variables
Family attachment	−.969**	−.212	−1.379***	−.219	−.969***	−.250	−.351***	−.198
Family attachment	(0.106)		(0.145)		(0.092)		(0.041)
School attachment	−.487***	−.131	−.600***	−.117	−.348***	−.110	−.088**	−.061
School attachment	(0.088)		(0.121)		(0.076)		(0.034)
Neighborhood attachment	−.018	−.007	−.149	−.041	−.106*	−.047	−.097***	−.094
Neighborhood attachment	(0.062)		(0.084)		(0.054)		(0.024)
Demographical variables
Gender	2.868***	.151	3.997***	.153	1.553***	.096	1.714***	.230
Gender	(0.423)		(0.580)		(0.367)		(0.164)
Grade 8	.514	.023	.689	.023	.978	.052	.433	.050
Grade 8	(0.608)		(0.834)		(0.530)		(0.235)
Grade 9	.521	.050	.301	.011	.162	.010	−.127	−.017
Grade 9	(0.521)		(0.715)		(0.453)		(0.202)
Intact family	−1.571**	−.063	−2.573**	−.075	−1.906***	−.090	−.873***	−.091
	(0.552)		(0.758)		(0.483)		(0.212)
	F = 8.990***		F = 34.781***		F = 36.167***		F = 41.725***
	N = 2,041		N = 2,042		N = 2,029		N = 2,016
	R² = .105		R² = .113		R² = .123		R² = .138

p < .05. **p < .01. ***p < .001.

Discussion and Conclusion

A primary feature of this study concerns the concept of attachment perceived by American and Chinese adolescents. As the first study that cross-nationally validates the attachment measures, formulation of the model derived from a consensus of findings based on a review of the attachment literature. The comparison of attachment among adolescents between the two countries has its significance in many ways. Of primary interest here, however, is to answer the questions: Can the factorial measurement of attachment theoretically and empirically derived from the United States hold true in China? If so, do two measurement models demonstrate invariance across the U.S. and Chinese samples? If not, what are the items that are noninvariant across two samples?

Results from the single-group CFA showed that data from both samples fit the three-factor baseline model adequately, providing strong support for a three-factor model for the 12-item attachment measure across samples. The three-factor models primarily derived from social control theory are similar in the factor structure across the two samples. Multigroup CFA demonstrated invariance of the factor structure, factor loadings, and factor covariance except for five nonequivalent items. Although partial measurement invariance was observed across samples, the consistency in goodness-of-fit indices between the unconstrained basic measurement model and the partially constrained models suggests that the partial invariance is acceptable (Ang et al., 2008; Cheung & Rensvold, 2002).

The results of the multigroup CFA analysis suggest that there is more invariance than noninvariance in the factorial measurement of attachment. More specifically, 7 out of a total of 12 manifest variables were invariant between the Chinese and the U.S. samples. This indicates that psychometric properties of these items can find their place in comparative studies, making the comparative study of criminological theories a worthwhile endeavor. We have a strong reason to believe that studies on the cross-national applicability of criminological theories, including examination of the validity and reliability of their measurements, are the key to the future advancement of the discipline.

The latent factor that shows consistent invariance across the U.S. and Chinese samples is family attachment. Attachment to family has been the most fundamental element of human social network and the most frequently measured attachment in the literature (e.g., Chen, 2009; Popp & Peguero, 2012; L. Zhang & Messner, 1996). As early as more than 2,000 years ago, Aristotle famously noted that family is the nucleus of social structure in any human society. That observation has not changed much with the eclipse of time. Despite different social environments, cultural heritages, and political systems, family attachment measured by a three-item latent variable remained invariant between the U.S. and Chinese samples. This finding indicates that American students and Chinese students in our samples tend to respond to our measures of family attachment in a similar fashion. The three items tapping family attachment touch the core elements of this construct—namely, the self-reported identification with father and mother (e.g., Menard & Grotpeter, 2011; Wu, Lake, & Cao, 2013; L. Zhang & Messner, 1996). Without the roles of parents, family loses its fundamental meaning, and a review of the literature reveals that relationship with father and mother reflects the essence of family attachment.

One of our research inquiries was to identify the items that manifest significant noninvariance across two samples in the factorial structure. The factors that show the most obvious noninvariance concern attachment to one’s neighborhood. The coefficients of three items were freely estimated to obtain an invariant model. Two of three coefficients indicate a stronger relationship with the targeted factor “Neighborhood Attachment” for the Chinese sample compared with that of the U.S. sample.¹⁵ More specifically, the two items are as follows: “My neighbors notice when I am misbehaving and let me know” and “This is a close-knit neighborhood.” At the same time, another two items detected as the major sources of noninvariance across two samples concern the students’ interactions with the teachers and the extracurricular activities in school. The coefficients of both factor loadings suggested a stronger attachment between the Chinese students and their school teachers and environments.

The nonequivalent items discussed above could in part be explained from a cultural standpoint, the emphasis on neighborhood cohesion and education in the Chinese society. Neighborhood in China is not only a place to live but also a place for intimate interactions with neighbors and for participation in civic activities such as morning exercise or area maintenance and upkeep. In urban areas, a neighborhood in a typical Chinese city is usually comprised of multilevel apartment complexes with high population density. Neighborhood committees are omnipresent and largely operated by local retired volunteers who gather information and organize neighborhood activities (Lai et al., 2010). Crime rates are extremely low and morning/evening group exercises (e.g., dancing and Taiji practice) usually attract a large crowd of participants, providing a convenient channel for them to keep abreast of both positive and negative incidents alike happening in their neighborhoods.¹⁶ For adolescents growing up in such environment, their attitudinal and behavioral attachments to their neighborhoods are usually strong and lasting.

An intriguing issue associated with this study concerns the utility of three-factor-model attachment in predicting juvenile delinquency, low self-control, and belief in subculture of violence commonly identified in the literature. The results of the OLS regression analyses clearly show evidence that the three factors are not only good measures of attachment but also significant predictors of the four dependent variables. Two observations deserve to be highlighted here. First, the three-factor model has its solid predictive validity in the important behavioral and attitudinal measures. Family attachment, school attachment, and neighborhood attachment all had independent and significant effects on the four outcome variables in both samples, except for the no effects of neighborhood attachment on versatility measures in the U.S. sample. Based on the standardized coefficients, school attachment was the strongest predictor among three attachments in the Chinese sample. In comparison, family attachment is the most robust predictor in the U.S. sample. Second, the overall predictive power of the three attachments in the U.S. sample was stronger than that in the Chinese sample, in that the R squared measures in the four equations for the U.S. sample are all larger than those in the Chinese sample.

Findings from the present study need to be considered in light of several limitations. First, the literature on the relationship between attachment and juvenile delinquency often includes a scale of attachment to delinquent friends (e.g., Brick, Taylor, & Esbensen, 2009; Leiber, Nalla, & Farnsworth, 1998; Ozbay & Ozcan, 2008). These studies found a positive association between attachment to delinquent friends, deviant behavior and unfavorable attitudes toward the legal authority (e.g., Leiber et al., 1998). In the current study, attachment to delinquent friends was not included due to the unavailability of such items in the survey instrument. Future research needs to include the measure of attachment to delinquent peers as part of social attachment of an adolescent. The second limitation concerns the breadth and range of the measures of three attachment factors. Although the measures are largely derived from the relevant theory and extant literature, the number of items and content coverage are rather limited and do not cover a wide range of behaviors/attitudes. Subsequent studies should incorporate additional items into the analysis of social attachment measures. Finally, as the Chinese data were collected in a large capital city located in the coastal province of more than 60 million population, the findings are rather limited to the coastal area where the economic boom has been the most noticeable.¹⁷ With China having the world’s largest population of school-aged adolescents, students living in inland provinces may perceive attachment differently from their counterparts in the coastal areas. Future research needs to include a sample of Chinese students from different geographic regions in the invariance testing.

Footnotes

Appendix

Appendix B

Measurement of the Dependent Variables.

Variables	Measurement items
Last-year versatility Cronbach’s α = .684 (China) Cronbach’s α = .780 (United States)	Vandalism, shoplifting, burglary, bicycle theft, car theft, car break, computer hacking, pick pocketing/snatching, carrying a weapon, robbery, group fight, assault, drug dealing, and hard drug (Ecstasy/speed, LSD[Lysergic acid diethylamide] /heroin, cocaine) use.
	0 = no, 1 = yes.
Lifetime versatility Cronbach’s α = .702 (China) Cronbach’s α = .829 (United States)	Vandalism, shoplifting, burglary, bicycle theft, car theft, car break, computer hacking, pick pocketing/snatching, carrying a weapon, robbery, group fight, assault, drug dealing, hard drug (Ecstasy/speed, LSD/heroin, cocaine) use.
	0 = no, 1 = yes. Both last-year versatility and lifetime versatility are computed based on a formula suggested by the ISRD-2 Steering Committee and Methodology Working Group. For the specific formula, see Footnote 5 in the NIJ Final Technical Report at https://www.ncjrs.gov/pdffiles1/nij/grants/238299.pdf
Low self-control Cronbach’s α = .852 (China) Cronbach’s α = .882 (United States)	I act on the spur of the moment without stopping to think. I do whatever brings me pleasure here and now, even at the cost of some distant goal. I’m more concerned with what happens to me in the short run than in the long run. I like to test myself every now and then by doing something a little risky. Sometimes I will take a risk just for the fun of it. Excitement and adventure are more important to me than security. I try to look out for myself first, even if it means making things difficult for other people. If things I do upset people, it’s their problem not mine. I will try to get the things I want even when I know it’s causing problems for other people. I lose my temper pretty easily. When I am really angry, other people better stay away from me. When I have a serious disagreement with someone, it’s usually hard for me to talk calmly about it without getting upset. 1 = fully disagree, 2 = somewhat disagree, 3 = somewhat agree, 4 = fully agree. Low self-control is an additive scale based on the 12 items above.
Pro-violence attitude Cronbach’s α = .719 (China) Cronbach’s α = .787 (United States)	A bit of violence is part of the fun. One needs to make use of force to be respected. If somebody attacks me, I will hit him or her back. Without violence everything would be much more boring. It is completely normal that boys want to prove themselves in physical fights with others.
	1 = fully disagree, 2 = somewhat disagree, 3 = somewhat agree, 4 = fully agree. Pro-violence attitude is an additive scale based on the five items above.

Note. ISRD-2 = International Self-Report Delinquency-2nd Wave; NIJ = National Institute of Justice.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The US portion of the ISRD-2 was supported by the National Institute of Justice, U.S. Department of Justice (#2006IJCX0045). The project Principal Investigator (PI) and Co-PI are Drs. Ni “Phil” He and Ineke Haen Marshall from Northeastern University. Points of view or opinions contained in this article are those of the authors and do not necessarily represent the official positions or policies of the funding agency.

Notes

References

Albright

J. J.

Park

H. M.

(2009, April). Confirmatory factor analysis using Amos, LISREL, Mplus, SAS/STAT CALIS (Working paper). The University Information Technology Services Center for Statistical and Mathematical Computing, Indiana University. Retrieved from http://www.indiana.edu/~statmath/stat/all/cfa/index.html

Anderson

A. F.

Gil

V. E.

(1998). China’s modernization and the decline of communitarianism. Journal of Contemporary Criminal Justice, 14, 248-261.

Ang

R. P.

Chong

W. H.

Huan

V. S.

Quek

C. L.

Yeo

L. S.

(2008). Teacher-student relationship inventory: Testing for invariance across upper elementary and junior high samples. Journal of Psychoeducational Assessment, 26, 339-349.

Armstrong

Katz

(2010). Further evidence on the discriminant validity of perceptual incivilities measures. Justice Quarterly, 27, 280-304.

Barrett

(2007). Structural equation modeling: Adjudging model fit. Personality and Individual Differences, 42, 815-824.

Bentler

P. M.

(2007). On tests and indices for evaluating structural models. Personality and Individual Differences, 42, 825-829.

Brick

B. T.

Taylor

T. J.

Esbensen

(2009). Juvenile attitudes towards the police: The importance of subcultural involvement and community ties. Journal of Criminal Justice, 37, 488-495.

Brown

T. A.

(2006). Confirmatory factor analysis for applied research. New York, NY: Guilford Press.

Bruinsma

G. I. N.

(1994, November). The test-retest reliability of the ISRD self-report instrument. Paper presented at the Annual Conference of the American Society of Criminology, Miami, FL.

10.

Burchfield

K. B.

(2009). Attachment as a source of informal social control in urban neighborhoods. Journal of Criminal Justice, 37, 45-54.

11.

Byrne

B. M.

(2012). Structural equation modeling with Mplus: Basic concepts, applications, and programming. New York, NY: Routledge.

12.

Cao

Hou

(2001). A comparison of confidence in the police in China and in the United States. Journal of Criminal Justice, 29, 87-99.

13.

Chen

(2002). Social control in China: Applications of labeling theory and the reintegrative shaming theory. International Journal of Offender Therapy and Comparative Criminology, 46, 45-63.

14.

Chen

(2004). Social and legal control in China: A comprehensive perspective. International Journal of Offender Therapy and Comparative Criminology, 48, 523-536.

15.

Chen

(2009). The link between juvenile offending and victimization: The influence of risky lifestyles, social bonding, and individual characteristics. Youth Violence and Juvenile Justice, 7, 119-135.

16.

Cheung

G. W.

Rensvold

R. B.

(2002). Evaluating goodness-of-fit indexes for testing measurement invariance. Structural Equation Modeling, 9, 233-255.

17.

Curran

D. J.

(1998). Economic reform, the floating population, and crime: The transformation of social control in China. Journal of Contemporary Criminal Justice, 14, 262-280.

18.

Diamantopoulos

Siguaw

J. A.

(2000). Introducing LISREL. London, England: Sage.

19.

Durkheim

(1951). Suicide: A study in sociology ( Spaulding

J. A.

Glencoe

S. G.

, Trans.). Glencoe, IL: Free Press.

20.

Elliott

D. S.

Ageton

S. S.

Canter

R. J.

(1979). An integrated perspective on delinquent behavior. Journal of Research on Crime & Delinquency, 16, 3-27.

21.

Elliott

D. S.

Huizinga

Ageton

S. S.

(1985). Explaining delinquency and drug use. Newbury Park, CA: Sage.

22.

Gibson

C. L.

Zhao

Lovrich

N. P.

Gaffney

M. J.

(2002). Social integration, individual perceptions of collective efficacy, and fear of crime in three cities. Justice Quarterly, 19, 537-564.

23.

Gomez

Burns

G. L.

Walsh

J. A.

Hafetz

(2005). A multitrait-multisource factor analytic approach to the construct validity of ADHD and ODD rating scales with Malaysian children. Journal of Abnormal Child Psychology, 33, 241-254.

24.

Greenberger

Chen

Beam

Whang

Dong

(2000). The perceived social contexts of adolescents’ misconduct: A comparative study of youths in three cultures. Journal of Research on Adolescence, 10, 365-388.

25.

Lai

(2011). A cross-cultural study of student problem behaviors in middle school. School Psychology International, 32, 20-34.

26.

Harris

J. O.

(2009). Relationship of factors pertaining to youth attitudes toward school, self -esteem, and violence (Doctoral dissertation). Ann Arbor, MI: Tennessee State University.

27.

Hartjen

A. C.

Priyadarsini

(2003). Gender, peers, and delinquency: A study of boys and girls in rural France. Youth & Society, 34, 387-414.

28.

Hirschi

(1969). Causes of delinquency. Berkeley: University of California Press.

29.

Hooper

Coughlan

Mullen

M. R.

(2008). Structural equation modeling: Guidelines for determining model fit. Electronic Journal of Business Research Methods, 6, 53-60.

30.

L. T.

Bentler

P. M.

(1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new Alternatives. Structural Equation Modeling, 6(1), 1-55.

31.

Huebner

B. M.

Schafer

J. A.

Bynum

T. S.

(2004). African American and white perceptions of police services: Within- and between-group variation. Journal of Criminal Justice, 32, 123-135.

32.

Jaccard

Wan

C. K.

(1996). LISREL approaches to interaction effects in multiple regression. Thousand Oaks, CA: Sage.

33.

Jessor

Turbin

M. S.

Costa

F. M.

Dong

Zhang

Wang

(2003). Adolescent problem behavior in China and the United States: A cross-national study of psychosocial protective factors. Journal of Research on Adolescence, 13, 329-360.

34.

Jiang

Lambert

(2009). Views of formal and informal crime control and their correlates in China. International Criminal Justice Review, 19, 5-24.

35.

Jiao

A. Y.

(2001). Police and culture: A comparison between China and the United States. Police Quarterly, 4, 156-185.

36.

Jöreskog

K. G.

Sörbom

(1989). LISREL 7. Analysis of linear structural relationships by maximum likelihood, instrumental variables, and least squares methods. Mooresville, IN: Scientific Software.

37.

Junger

Marshall

I. H.

(1997). The interethnic generalizability of social control theory: An empirical test. Journal of Research in Crime & Delinquency, 34, 79-112.

38.

Kim

Ferree

(1981). Standardization in causal analysis. Sociological Methods & Research, 10, 504-523.

39.

Kline

R. B.

(2005). Principles and practice of structural equation modeling (2nd ed.). New York, NY: Guilford Press.

40.

Kline

R. B.

(2011). Principles and practice of structural equation modeling (3rd ed.). New York, NY: Guilford Press.

41.

Kort-Butler

L. A.

Tyler

K. A.

Melander

L. A.

(2011). Childhood maltreatment, parental monitoring, and self-control among homeless young adults: Consequences for negative social outcomes. Criminal Justice and Behavior, 38, 1244-1264.

42.

Lai

Cao

Zhao

(2010). The impact of political entity on confidence in legal authorities: A comparison between China and Taiwan. Journal of Criminal Justice, 38, 934-941.

43.

Leiber

Nalla

Farnsworth

(1998). Explaining juveniles’ attitudes toward the police. Justice Quarterly, 15, 151-171.

44.

Loehlin

(1992). Latent variables models: An introduction to factor, path, and structural analysis (2nd ed.). Englewood Cliffs, NJ: Lawrence Erlbaum.

45.

Ren

Marshall

I. H.

(2013). Exploring the utility of self-control theory for risky behavior and minor delinquency among Chinese adolescents. Journal of Contemporary Criminal Justice, 29, 32-52.

46.

Marshall

I. H.

(2010). “Pourquoi pas?” versus “absolutely not!” Cross-national differences in access to schools and pupils for survey research. European Journal on Criminal Policy and Research, 16, 89-109.

47.

Marshall

I. H.

Enzmann

(2012). The generalizability of self-control theory. In Junger-Tas

Marshall

I. H.

Enzmann

Killias

Steketee

Gruszczyńska

(Eds.), The many faces of youth crime: Contrasting theoretical perspectives on juvenile delinquency across countries and cultures (pp. 285-325, E1-E7). New York, NY: Springer.

48.

Marshall

I. H.

Webb

V. J.

(1990, December). Omaha ISRD pilot study. Paper presented at the meeting of international self-report delinquency project, The Hague, The Netherlands.

49.

Marshall

I. H.

Webb

V. J.

(1994). Self-reported delinquency in a mid-western American city. In Junger-Tas

Terlouw

Klein

M. W.

(Eds.), Delinquent behavior among young people in the western world: First results of the international self-report delinquency study (pp. 319-342). Amsterdam, The Netherlands: Kugler.

50.

Menard

Grotpeter

(2011). Peer influence, social bonding, physical and relational aggression: Perpetration and victimization in an elementary school sample. Victims & Offenders, 6, 181-206.

51.

Merton

R. K.

(1938). Social structure and Anomie. American Sociological Review, 3, 672-682.

52.

Miles

Shevlin

(2007). A time and a place for incremental fit indices. Personality and Individual Differences, 42, 869-874.

53.

Muthén

L. K.

Muthén

B. O.

(1998-2012). Mplus user’s guide (7th ed.). Los Angeles, CA: Author.

54.

Ozbay

Ozcan

Y. Z.

(2008). A test of Hirschi’s social bonding theory: A comparison of male and female delinquency. International Journal of Offender Therapy and Comparative Criminology, 52, 134-157.

55.

Park

R. E.

(1952). Human communities: The city and human ecology. Glencoe, IL: Free Press.

56.

Podsakokk

P. M.

MacKenzie

S. B.

Lee

Podsakoff

N. P.

(2003). Common method biases in behavioral research: A critical review of the literature and recommended remedies. Journal of Applied Psychology, 88, 879-903.

57.

Popp

A. M.

Peguero

A. A.

(2012). Social bonds and the role of school-based victimization. Journal of Interpersonal Violence, 27, 1-23.

58.

Posick

Rocque

(2015). Family matters: A cross-national examination of family bonding and victimization. European Journal of Criminology, 12, 51-69.

59.

Ren

Cao

Lovrich

Gaffney

(2005). Linking confidence in the police with the performance of the police: Community policing can make a difference. Journal of Criminal Justice, 33, 55-66.

60.

Sampson

R. J.

Raudenbush

S. W.

Earls

(1997). Neighborhood and violent crime: A multilevel study of collective efficacy. Sciences, 277, 918-924.

61.

Satorra

Bentler

P. M.

(2010). Ensuring positiveness of the scaled difference chi-square test statistic. Psychometrika, 75, 243-248.

62.

Schreck

C. J.

Stewart

E. A.

Fisher

B. S.

(2006). Self-control, victimization, and their influence on risky lifestyles: A longitudinal analysis using panel data. Journal of Quantitative Criminology, 22, 319-340.

63.

Schumacker

R. E.

Lomax

R. G.

(2004). A beginner’s guide to structural equation modeling (2nd ed.). Mahwah, NJ: Lawrence Erlbaum.

64.

Shaw

C. R.

McKay

H. D.

(1942). Juvenile delinquency and urban areas. Chicago, IL: University of Chicago Press.

65.

Shaw

V. N.

(1996). Social control in China: A study of Chinese work units. Westport, CT: Praeger.

66.

Shoemaker

J. D.

(1994). Male-female delinquency in the Philippines: A comparative analysis. Youth & Society, 25, 299-329.

67.

Silver

Miller

L. L.

(2004). Sources of informal social control in Chicago neighborhoods. Criminology, 42, 551-583.

68.

Song

H. R.

Thompson

R. A.

Ferrer

(2009). Attachment and self-evaluation in Chinese adolescents: Age and gender differences. Journal of Adolescence, 32, 1267-1286.

69.

Steiger

J. H.

(2007). Understanding the limitations of global fit assessment in structural equation modeling. Personality and Individual Differences, 42, 893-898.

70.

Stewart

(2003). School social bonds, school climate, and school misbehavior: A multilevel analysis. Justice Quarterly, 20, 575-613.

71.

Stone

E. F.

Hollenbeck

J. R

(1989). Clarifying some controversial issues surrounding statistical procedures for detecting moderate variables: Empirical evidence and related matters. Journal of Applied Psychology, 74, 3-10.

72.

Tabachnick

B. G.

Fidell

L. S.

(2007). Using multivariate statistics (5th ed.). New York, NY: Allyn & Bacon.

73.

Taylor

(2001). The relationship between social and self-control: Tracing Hirschi’s criminological career. Theoretical Criminology, 5, 369-388.

74.

Ward

J. T.

Boman

J. H. I. V.

Shayne

(2012). Hirschi’s redefined self-control: Assessing the implications of the merger between social- and self-control theories. Crime & Delinquency. Advance online publication. doi:10.1177/0011128712466939

75.

Wheaton

Muthen

Alwin

D. F.

Summers

G. F.

(1977). Assessing reliability and stability in penal models. Sociological Methodology, 8, 84-163.

76.

Wilson

J. Q.

Kelling

(1982). Broken windows. Atlantic Monthly, 211, 29-38.

77.

Woldoff

R. A.

(2002). The effects of local stressors on neighborhood attachment. Social Forces, 81, 87-116.

78.

Worrall

J. L.

(2006). The discriminant validity of perceptual incivility measures. Justice Quarterly, 23, 360-383.

79.

Lake

Cao

(2013). Race, social bonds, and juvenile attitudes toward the police. Justice Quarterly. Advance online publication. doi:10.1080/07418825.2013.778325

80.

Zhang

Zhao

Ren

(2014). Social attachment and juvenile attitudes toward the police in China: Bridging Eastern and Western wisdom. Journal of Research in Crime & Delinquency, 51, 703-734.

81.

Zhang

(2008). Juvenile delinquency and justice in contemporary China: A critical review of the literature over 15 years. Law, Crime, and Social Change, 50, 149-160.

82.

Zhang

Messner

S. F.

(1996). School attachment and official delinquency status in the People’s Republic of China. Sociological Forum, 11, 285-303.

83.

Zhang

Benson

Deng

(2000). A test-retest reliability assessment of the international self-report delinquency instrument. Journal of Criminal Justice, 28, 283-295.

84.

Zhang

Day

Cao

(2012). A partial test of Agnew’s general theory of crime and delinquency. Crime & Delinquency, 58, 856-878.