Psychological Assessment

Transcription

1 Psychological Assessment Cross-Cultural Examination of Measurement Invariance of the Beck Depression Inventory II Jessica Dere, Carolyn A. Watters, Stephanie Chee-Min Yu, R. Michael Bagby, Andrew G. Ryder, and Kate L. Harkness Online First Publication, October 13, CITATION Dere, J., Watters, C. A., Yu, S. C.-M., Bagby, R. M., Ryder, A. G., & Harkness, K. L. (2014, October 13). Cross-Cultural Examination of Measurement Invariance of the Beck Depression Inventory II. Psychological Assessment. Advance online publication.

2 Psychological Assessment 2014 American Psychological Association 2014, Vol. 26, No. 4, /14/$ Cross-Cultural Examination of Measurement Invariance of the Beck Depression Inventory II Jessica Dere University of Toronto Scarborough Stephanie Chee-Min Yu Queen s University Carolyn A. Watters University of Toronto R. Michael Bagby University of Toronto Andrew G. Ryder Concordia University and Jewish General Hospital, Montreal, Quebec, Canada Kate L. Harkness Queen s University Given substantial rates of major depressive disorder among college and university students, as well as the growing cultural diversity on many campuses, establishing the cross-cultural validity of relevant assessment tools is important. In the current investigation, we examined the Beck Depression Inventory Second Edition (BDI II; Beck, Steer, & Brown, 1996) among Chinese-heritage (n 933) and Europeanheritage (n 933) undergraduates in North America. The investigation integrated 3 distinct lines of inquiry: (a) the literature on cultural variation in depressive symptom reporting between people of Chinese and Western heritage; (b) recent developments regarding the factor structure of the BDI II; and (c) the application of advanced statistical techniques to the issue of cross-cultural measurement invariance. A bifactor model was found to represent the optimal factor structure of the BDI II. Multigroup confirmatory factor analysis showed that the BDI II had strong measurement invariance across both culture and gender. In group comparisons with latent and observed variables, Chinese-heritage students scored higher than European-heritage students on cognitive symptoms of depression. This finding deviates from the commonly held view that those of Chinese heritage somatize depression. These findings hold implications for the study and use of the BDI II, highlight the value of advanced statistical techniques such as multigroup confirmatory factor analysis, and offer methodological lessons for cross-cultural psychopathology research more broadly. Keywords: major depressive disorder, Beck Depression Inventory Second Edition, factor analysis, measurement invariance, cross-cultural Supplemental materials: Major depressive disorder (MDD) is among the most prevalent psychiatric disorders in North America and around the world (Kessler et al., 2003; Patten et al., 2006), and it represents Jessica Dere, Department of Psychology, University of Toronto Scarborough; Carolyn A. Watters, Department of Psychology, University of Toronto; Stephanie Chee-Min Yu, Department of Psychology, Queen s University; R. Michael Bagby, Departments of Psychiatry and Psychology, University of Toronto; Andrew G. Ryder, Department of Psychology and Centre for Clinical Research in Health, Concordia University, and Culture and Mental Health Research Unit and Lady Davis Institute, Jewish General Hospital, Montreal, Quebec, Canada; Kate L. Harkness, Department of Psychology, Queen s University. Stephanie Chee-Min Yu is now at the Department of Social Work and Social Administration, The University of Hong Kong. The first and second authors contributed equally to this work. Preparation of this article was supported by a fellowship granted to Jessica Dere in the Social Aetiology of Mental Illness training program, a Strategic Training Initiative in Health Research funded by the Canadian Institutes of Health Research (CIHR), based at the Centre for Addiction and Mental one of the greatest sources of disease burden internationally (Cuijpers, Beekman, & Reynolds, 2012; Üstün, Ayuso-Mateos, Chatterji, Mathers, & Murray, 2004). 1 MDD is projected to Health and the University of Toronto. Preparation of the article was also supported by the Joseph-Armand Bombardier Doctoral Scholarship, awarded to Carolyn A. Watters by the Social Sciences and Humanities Research Council of Canada, and by a New Investigator Award from the CIHR, awarded to Andrew G. Ryder. Correspondence concerning this article should be addressed to Jessica Dere, Department of Psychology, University of Toronto Scarborough, 1265 Military Trail, Toronto, Ontario, Canada M1C 1A4, or to Kate L. Harkness, Department of Psychology, Queen s University, 222 Craine, Kingston, Ontario, Canada K7L 3N6. jdere@utsc.utoronto.ca or harkness@queensu.ca 1 Disease burden broadly refers to a disease s relative impact on morbidity and mortality and is commonly measured in disability-adjusted life years (Üstün et al., 2004). 1

3 2 DERE, WATTERS, YU, BAGBY, RYDER, AND HARKNESS rank first among all sources of disease burden in high-income countries by the year 2030 (Mathers & Loncar, 2006). It typically has its initial onset in late adolescence or young adulthood (Eisenberg, Gollust, Golberstein, & Hefner, 2007). Rates of MDD in this age group, particularly among college and university students, appear to be even higher than in the general population and may be increasing (Bayram & Bilgel 2008; Ibrahim, Kelly, Adams, & Glazebrook, 2013; Young, Fang, & Zisook, 2010). The onset of MDD at this life stage is associated with significant negative consequences across numerous life domains that persist throughout adulthood. Crucial to early identification and intervention efforts in MDD is the existence of well-validated, psychometrically sound assessment tools. The increasing cultural diversity of many undergraduate student bodies (e.g., Horn, Peter, Rooney, & Malizio, 2002) means that assessment tools for MDD must also be cross-culturally valid. This is particularly salient given the significant cultural variation in the presentation of depressive symptoms (e.g., Kirmayer, 2001; Ryder et al., 2008). In the current study we examine one of the most commonly used self-report measures of depressive symptoms the Beck Depression Inventory Second Edition (BDI II; Beck, Steer, & Brown, 1996) among Chinese-heritage and Europeanheritage undergraduate students in North America. 2 With the aim of advancing the understanding of the cultural shaping of depressive symptom reporting, we integrate three distinct lines of inquiry: (a) the literature on cultural variation in depressive symptom reporting between people of Chinese and Western heritage; (b) recent developments regarding the factor structure of the BDI II; and (c) the application of advanced statistical techniques to the issue of cross-cultural measurement equivalence. Supplementing this cross-cultural focus, our analyses also include examination of potential gender effects within each cultural group. To the best of our knowledge, this study is the first to examine measurement invariance of the BDI II on the basis of both cultural and genderwithin-culture group comparisons. Cultural Variations in the Reporting of Depression Symptoms The empirical literature on cultural variations in depression is dominated by comparisons of Chinese and Western samples. It is rooted in a long-standing theoretical and empirical interest in the phenomenon of Chinese somatization, a popular topic in the interdisciplinary field of culture and mental health (see Ryder & Chentsova-Dutton, 2012). The idea that people of Chinese heritage tend to emphasize somatic (e.g., fatigue, sleep difficulties) rather than psychological (e.g., hopelessness, guilt) symptoms of depression has been discussed for several decades (e.g., Kleinman, 1982; Parker, Cheah, & Roy, 2001; Ryder et al., 2008); this pattern has been examined among Chinese-heritage samples in both East Asian and Western contexts (e.g., Parker, Chan, Tully, & Eisenbruch, 2005; Yen, Robins, & Lin, 2000). A relative emphasis on somatic symptoms is contrasted whether explicitly or implicitly with a Western emphasis on psychological symptoms of depression, which has been termed psychologization (e.g., Ryder & Chentsova-Dutton, 2012). Although Chinese somatization has been discussed since the 1970s, direct cross-cultural comparisons between Chinese and Western samples in the presentation of symptoms have only recently been conducted. In reviewing the existing empirical literature, a contrast emerges between clinical and nonclinical samples regarding the extent of support for the idea of Chinese somatization. The only two crossnational studies with clinical samples provide support for Chinese somatization and for Western psychologization (Parker et al., 2001; Ryder et al., 2008). Parker et al. (2001) found that a greater percentage of Malaysian Chinese outpatients than Euro Australian outpatients reported a somatic symptom as their primary presenting complaint. Furthermore, the Chinese group endorsed more somatic symptoms on a symptom questionnaire, whereas the Euro Australians endorsed more psychological symptoms. Ryder et al. (2008) compared depressive symptom reporting between Chinese and Euro Canadian depressed psychiatric outpatients and found greater somatic symptom reporting among the Chinese and greater psychological symptom reporting among the Euro Canadians. In both of these investigations, the cultural difference in psychological symptom reporting was a stronger effect than the difference in somatic symptom reporting. The pattern of cultural variation in depressive symptom reporting is not as consistent in studies using college student and community samples. Broad claims to the effect that Chinese groups always somatize distress and depression are likely overstated (see also Dere et al., 2013; Zhou et al., 2011). Yen et al. (2000) found that Chinese students seeking mental health services endorsed a higher proportion of somatic symptoms of depression, using a Chinese translation of the Center for Epidemiological Studies Depression Scale (CES-D; Radloff, 1977), than did a Chinese student control group. In a separate study comparing Chinese, Chinese American, and Euro American students, however, the Chinese group reported a lower proportion of somatic symptoms on the CES-D than did the other groups; participants in this study were not recruited or differentiated on the basis of help-seeking status. In their interpretation of the findings across their two studies, Yen et al. (2000) suggested that a Chinese emphasis on somatic symptoms might occur only in the context of help seeking. In a comparison of adolescents in Hong Kong and the United States, Stewart et al. (2002) found that, among adolescents diagnosed with MDD, those in Hong Kong endorsed fatigue/loss of energy more frequently and irritability less frequently than did those in the United States. In contrast, among adolescents with diagnostically subthreshold depression, those in Hong Kong endorsed a variety of both somatic (i.e., weight/appetite changes, fatigue/loss of energy, psychomotor changes) and psychological (i.e., worthlessness/guilt, concentration difficulties/indecisiveness, suicidal ideation) symptoms more often than did those in the United States. These latter results in particular are contrary to theoretical expectations about the absence of psychological symptoms among Chinese samples. Though their study was not a cross-cultural comparison, Chang (2007) examined symptom reporting on a Chinese version of the BDI II (C-BDI II) among probably depressed (i.e., C-BDI II score 16) and nondepressed (i.e., C-BDI II score 16) Chinese college students in Taiwan. Groups were compared on the 2 The terms Chinese-heritage and European-heritage are used here to refer to the ethno-cultural background of our participants in broad terms, in recognition of the heterogeneous makeup of the two cultural groups within our sample.

4 CROSS-CULTURAL EXAMINATION OF THE BDI II 3 basis of their scores on somatic and cognitive-affective factors on the C-BDI II, as derived through exploratory factor analysis. On the whole, the Chinese students endorsed a higher proportion of somatic symptoms (e.g., fatigue, changes in appetite) than cognitive-affective symptoms (e.g., worthlessness, pessimism). Relative emphasis on somatic versus cognitive-affective symptoms varied, however, by level of severity. The probably depressed group placed less emphasis on somatic symptoms and more emphasis on cognitive-affective symptoms than did the nondepressed group. Similar to the findings of Stewart et al. (2002), these results suggest that patterns of symptom reporting vary with level of severity and demonstrate that a simple prediction of a somatic symptom emphasis among Chinese-heritage college students may not be warranted. In sum, the studies above suggest that claims regarding Chinese somatization and Western psychologization have to be qualified, as variation in symptom reporting may depend on symptom severity and treatment-seeking status. What is absent from this literature is any discussion of the extent to which these findings have been influenced by potential variation in how the items on symptom measures are interpreted by respondents from different cultural groups. This issue is important, given that proposed explanations for cultural group differences in depressive symptom reporting include cultural variations in response style or response biases (e.g., Lam, Pepper, & Ryabchenko, 2004) and that there is a substantial literature on cross-cultural measurement and methodology (e.g., Chen, 2008; Little, 1997; Milfont & Fischer, 2010; van de Vijver & Leung, 2000). Central to interpreting the above findings, therefore, is establishing cross-cultural measurement equivalence or invariance across Chinese and Western groups. 3 To our knowledge, the current study is the first to undertake this task in college-age samples. The examination of measurement equivalence for an assessment tool first requires that a baseline factor structure can be established. Equivalence or invariance analyses proceeds by determining the best fitting factor structure across the groups of interest. In the case of the BDI II, this proves to be an area of considerable debate. Therefore, before moving to a discussion of measurement equivalence, we provide a summary of recent developments regarding the factor structure of the BDI II. Factor Structure of the BDI II The BDI II (Beck, Steer, & Brown, 1996) is a 21-item selfreport measure of depressive symptoms, designed to correspond with the Diagnostic and Statistical Manual of Mental Disorders (4th ed.; American Psychiatric Association, 1994) diagnostic criteria for MDD. Respondents rate the severity of symptoms based on their experiences over the previous 2 weeks, on a scale ranging from 0 to 3. The response options vary across items, but the scale always progresses from low to high severity. Individual item scores are summed to provide a total score of depression severity. This measure generally shows strong reliability and validity, across a variety of study populations (e.g., Beck, Steer, Ball, & Ranieri, 1996; Storch, Roberti, & Roth, 2004; Wiebe & Penley, 2005). In their original presentation of the BDI II, Beck, Steer, and Brown (1996) suggested that a two-factor structure provided the best fit among both clinical outpatients and college students. However, the two factors differed between the groups; somaticaffective and cognitive factors were found for the clinical sample, whereas cognitive-affective and somatic factors were found for the students (see Dozois, Dobson, & Ahnberg, 1998). A large number of alternative factor structures (models) have since been proposed and have received stronger support, across a variety of samples (e.g., Brouwer, Meijer, & Zevalkink, 2013; Quilty, Zhang, & Bagby, 2010; Vanheule, Desmet, Groenvynck, Rosseel, & Fontaine, 2008). A number of these structures, with a focus on models that have previously been tested among undergraduate students and in Chinese-heritage samples, are displayed in Table 1. Detailed summaries of the studies referenced in Table 1 can be found elsewhere (Brouwer et al., 2013; Quilty et al., 2010; Vanheule et al., 2008); however, it is worthwhile to highlight several important observations regarding the different models. All of the models except for one (Model 1) include some combination of cognitive, somatic, and affective symptom factors. However, there is considerable inconsistency in the item composition of factors across the proposed structures, and many items are included in different symptom domains across different models. The presence of many different factor structures hinders the interpretation and comparability of results across studies. Adding to this issue, Model 1 is made up of three factors (negative attitude, performance difficulty, and somatic elements) that do not easily map onto the cognitive, somatic, or affective factors included in the majority of other models. Model 1 is also the model most commonly used in studies examining the BDI II among Chinese and East Asian groups (e.g., Byrne, Stewart, Kennard, & Lee, 2007; Whisman, Juss, Whiteford, & Gelhorn, 2013; Wu, 2010; Wu & Huang, 2014), making it particularly difficult to interpret the results of these studies within the broader BDI II literature. Notwithstanding the inconsistencies across these studies, there is some consensus emerging supporting a bifactor structure for the BDI II (Al-Turkait & Ohaeri, 2010; Brouwer et al., 2013; Osman, Barrios, Gutierrez, Williams, & Bailey, 2008; Quilty et al., 2010; Ward, 2006). In a bifactor model, items are specified to load onto a general factor made up of all scale items, and they may have an additional loading on a subfactor composed of a subset of items representing a narrower construct. The effects of the general factor are not included in the loadings on the subfactors. Second-order models are similar to bifactor models in that they both represent a hierarchical structure; however, only bifactor models can separate the unique variance of the subfactors from the variance of the general construct. Therefore, a bifactor approach has the advantage of simultaneously capturing a general construct shared by a set of subfactors, as well as the unique effect of each subfactor over and above the general construct (Chen, Hayes, Carver, Laurenceau, & Zhang, 2012). Support for a bifactor model of depression follows a broader trend in psychopathology research, which suggests that bifactor models offer a valuable approach for analyzing clinical questionnaires that reflect a single overarching construct while also being multidimensional (see Brouwer et al., 2013; Reise, 3 The terms measurement equivalence and measurement invariance are generally seen as synonymous. Equivalence is more often used in conceptual discussions of this topic, including in the cultural literature, whereas invariance is favored in statistically focused discussions (Little, 1997). Following these norms, both terms are used in the current paper as appropriate.

5 4 DERE, WATTERS, YU, BAGBY, RYDER, AND HARKNESS Table 1 Overview of Previously Published BDI II Factor Structures Derived and/or Tested in College or University Student Samples, Chinese-Heritage Samples, and Studies Testing Complex BDI II Structures (i.e., Second-Order and Bifactor Models) Item First-order oblique Bifactor 1 a,b 2 a,b 3 4 b 5 b 6 1. Sadness NA C C C A 2. Pessimism NA C C C A C 3. Past failure NA C C C C C 4. Loss of pleasure PD C C S A 5. Guilty feelings NA C C C C C 6. Punishment feelings NA C C C C C 7. Self-dislike NA C C C C C 8. Self-criticism NA C C C C C 9. Suicidal thoughts NA C C C A C 10. Crying NA C C S S 11. Agitation PD C C S S 12. Loss of interest PD C C S A 13. Indecisiveness PD C S C C 14. Worthlessness NA C C C C C 15. Loss of energy S S S S S S 16. Changes in sleep S S S S S S 17. Irritability PD C C S S 18. Changes in appetite S S S S S S 19. Concentration difficulty PD S S S S S 20. Tiredness S S S S S S 21. Loss of interest in sex S C C S S Note. BDI II Beck Depression Inventory II; A affective; C cognitive (or cognitive-affective: Models 2, 3, 4, 6); S somatic (or somatic-elements: Model 1; somatic-affective: Models 2, 3; somatic-vegetative: Model 4); NA negative affectivity; PD performance difficulty. Model 1 Al-Turkait & Ohaeri (2010) a,b ; Byrne et al. (2007; with Item 21 deleted) a ; Whisman et al. (2013) a ; Wu (2010); Wu & Huang (2014). Model 2 Al-Turkait & Ohaeri (2010) a,b ; Osman et al. (2008) b ; Storch et al. (2004); Whisman et al. (2000), modified from Beck et al. (1996; clinical sample). Model 3 Beck et al. (1996; student sample). Model 4 Brouwer et al. (2013) b ; Dozois et al. (1998). Model 5 Beck et al. (2002), as cited in Brouwer et al. (2013) b. Model 6 Brouwer et al. (2013); Quilty et al. (2010); Ward (2006). a Second-order model. b Bifactor model. Bonifay, & Haviland, 2013; Watters, Keefer, Kloosterman, Summerfeldt, & Parker, 2013). Ward (2006) was the first to investigate a bifactor model of the BDI II and proposed a model in which only some of the items were assigned to either a cognitive or a somatic subfactor (Model 6 in Table 1). Of interest, the items that did not load onto either subfactor in Ward s model show the greatest inconsistency in terms of factor assignment across previous studies, as shown in Table 1. Further, when comparing the quality of different models, subsequent studies have found that Ward s (2006) model not only is optimal compared to first and second-order versions of the same model (Al-Turkait & Ohaeri, 2010; Brouwer et al., 2013; Quilty et al., 2010) but is also superior to bifactor versions of other BDI II models (Al-Turkait & Ohaeri, 2010). All of the loadings in Ward s model tend to be strong and significant, whereas this is not the case with other bifactor BDI II models. In other words, Ward s model appears to be the most parsimonious bifactor BDI II structure. However, no bifactor model of the BDI II including Ward s model has been tested for measurement equivalence across either culture or gender. Measurement Equivalence When making comparisons between groups distinguished by such characteristics as cultural background or gender, researchers routinely assume that the instrument(s) or test(s) they are using have measurement equivalence, such that they are measuring the same construct across the groups in the same way (Chen, 2008; Little, 1997). An instrument or test can be said to show equivalence across groups when members of each group assign the same meanings to the test instrument and its constituent items and when respondents who share the same level of the underlying construct obtain the same score regardless of group membership. Such equivalence permits more meaningful group comparisons and readily interpretable results than does a methodological scenario in which measurement invariance is not examined (Meredith, 1993). Nevertheless, the majority of cross-cultural studies do not typically examine measurement equivalence. Insufficiently powered sample sizes and a lack of appropriate statistical methods likely contribute to the failure to examine or establish measure invariance. Ongoing advances in structural equation modeling have led to robust statistical approaches for addressing measurement equivalence, often referred to as tests of measurement invariance (MI). Within the structural equation modeling framework, multigroup confirmatory factor analysis (MG-CFA) is generally accepted as the most common approach (Chen, 2008; Milfont & Fischer, 2010). MG-CFA can also be used to conduct latent group comparisons, testing for group differences in latent means, covariances, and variances. Such comparisons, known as tests of struc-

6 CROSS-CULTURAL EXAMINATION OF THE BDI II 5 tural invariance, provide the advantage of controlling for measurement error (Chen, 2008; Little, 1997; Meredith, 1993). Four levels of measurement equivalence are generally discussed, with each level defined by a more restrictive set of requirements (e.g., Milfont & Fischer, 2010). The first is functional equivalence, which refers to whether or not the construct of interest exists in each group under study; this level of equivalence cannot in fact be tested statistically and is therefore not a focus of the current study. 4 The next level is termed configural or structural equivalence; it is met when a construct is made up of the same number of factors, with the same items associated with each factor, in each group. If this level is not met, the assessment tool is not measuring the same construct across groups. When the factor loadings of all items are also equal across the groups, the next level of equivalence, metric or factor loading equivalence, is met. Equivalence at this level is required for meaningful comparison of predictive relationships across groups. Finally, scalar or intercept equivalence is met when individual items show the same point of origin (i.e., intercept) across the groups; this level of equivalence is necessary for comparing group means (Chen, 2008; Little, 1997). Chen (2008) demonstrated the potential consequences of assuming rather than testing for MI in cross-cultural research. In a series of simulation studies, Chen (2008) found that a lack of metric invariance could lead to artificial interaction effects between two constructs. Furthermore, a lack of metric and scalar invariance can lead to spurious group differences or to a failure to uncover actual group differences. These findings suggest that failing to test for MI likely contributes to the inconsistent pattern of results that can be seen in various areas of the cross-cultural literature, including studies on cultural variation in depressive symptom reporting. Consideration of measurement issues is necessary in order to advance the field, by helping to tease apart actual cultural variations from those driven by measurement artifacts. Testing for MI is relevant when considering the need for psychometrically sound assessment instruments that are appropriate for the identification of depression among culturally diverse populations, such as those found on many North American college and university campuses. Measurement Invariance Analyses of the BDI II To date, four studies have examined MI with the BDI II. Wu and colleagues investigated MI across gender among Taiwanese adolescents and college students, using the Chinese version of the BDI II (Wu, 2010; Wu & Huang, 2014). Byrne et al. (2007) conducted MI testing of the BDI II among adolescents in Hong Kong and the United States, using the Chinese version of the BDI II in the Hong Kong sample. Whisman et al. (2013) conducted the only other MI study of the BDI II with a cross-cultural focus; they examined the BDI II across gender and across ethnocultural groups (White, Black, Asian, and Latino) among U.S. college student samples but did not examine gender within each ethno-cultural group. All of these studies examined either a first-order or a secondorder version of the three-factor structure that consists of negative affect, performance difficulty, and somatic elements. A bifactor model was not considered in any of these studies. Nevertheless, all four studies showed strong invariance (i.e., scalar or intercept invariance; Meredith, 1993) when using a common criteria of a change in confirmatory fit index of more than Taken together, an important next step is to integrate the recent focus on bifactor models, MI testing, and group comparisons in depressive symptom reporting into a single cross-cultural study using the BDI II. The Current Investigation In the current investigation, we examine depressive symptom reporting on the BDI II between Chinese-heritage (CH) and European-heritage (EH) student groups. First, we test a number of potential baseline models including bifactor models and then conduct an examination of configural, metric, and scalar levels of MI across both culture and gender within culture, prior to making group comparisons. This is the first study to investigate a bifactor model in a Chinese-heritage sample and also the first to conduct MI on a bifactor model of the BDI II across both culture and gender. Our objectives were to (a) determine an optimal factor structure for the BDI II among CH and EH students in Canada; (b) examine the MI of the BDI II in these two groups, across culture and gender; and (c) contingent upon the MI results, examine cultural and gender variation in depressive symptom reporting in these two groups using the BDI II. Method Participants and Procedure Participants were drawn from an archival database of students who completed a large questionnaire battery while enrolled in an introductory psychology class at a Canadian university in the province of Ontario; the university has an undergraduate enrollment of approximately 16,000 students. The battery is administered on an annual basis, and the current data were collected between the years 2005 and The battery consisted of a prescreening inventory and included a variety of self-report psychological measures. These measures included the BDI II and a demographics questionnaire containing a set of basic questions regarding age, gender, and ethno-cultural background. All participants provided written informed consent and received course credit for their participation; ethical approval was received from the university institutional review board. To be included in the current study sample, participants must have provided complete data concerning their gender and ethnocultural identity, as well as complete data on the BDI II. Only students who indicated their ethno-cultural group membership as 4 Although the current study did not examine the functional level of equivalence, previous research suggests that the depression symptom content of Western instruments (such as the BDI II) is comprehensible, endorsed, and even spontaneously reported as presenting problems by at least some respondents in both Chinese and Euro Canadian cultural contexts (e.g., Ryder et al., 2008). 5 Wu and Huang (2014) and Wu (2010) reported that several BDI II items showed a lack of invariance at the scalar level, based on a significant change in the Satorra Bentler chi-square statistic ( S-B 2 ). However, others have suggested that S-B 2 should be used only for descriptive purposes in large sample sizes (such as those found in Wu, 2010, and Wu & Huang, 2014), due to the chi-square statistic s sensitivity to sample size (Cheung & Rensvold, 2002). Using the criteria recommended by Cheung and Rensvold (2002) described in the analysis section, these studies support strong invariance.

7 6 DERE, WATTERS, YU, BAGBY, RYDER, AND HARKNESS either White or East Asian were included in the sample. Students who self-identified as White were included in the European-heritage group. Among those students who selfidentified as East Asian, only those of Chinese heritage were included in the final sample. This inclusion criterion was established in order to achieve a more culturally homogeneous sample; also, the students of Chinese heritage represented a significant majority within the East Asian group. Inclusion in the Chineseheritage group was based on participants response to an openended question inquiring about the country of origin of their family. Students who self-identified as East Asian and who referenced China, Taiwan, and/or Hong Kong in their response to this question were included in the CH group. As the number of EH students was substantially larger than the number of CH students, EH participants were randomly selected to match the CH sample on gender, stratified by year of data collection. After implementation of the inclusion criteria described above and the subsequent matching procedure, the final sample consisted of 933 CH (mean age years, SD 1.25, range 16 to 36 years) and 933 EH students (mean age years, SD 1.72, range 16 to 43 years), of whom 68% were women. The percentage of participants by year of data collection ranged from 8% to 16% of the total sample. Within the CH group, 53.6% were born in mainland China, Taiwan, or Hong Kong; 42.2% were born in Canada, 3.1% were born elsewhere, and 1.1% did not provide a country of birth. As the data were drawn from an annual prescreening battery that includes different questionnaires from year to year, no additional sociodemographic variables were available for the entire sample. Analysis Sequential sets of analyses were conducted for each of three pairs of between-group comparisons: CH versus EH, CH men versus CH women, and EH men versus EH women. In the first set of analyses, competing factor structures of the BDI II (see Table 1) were tested to establish the optimal BDI II measurement model across groups. Due to mounting evidence that a bifactor model best represents the BDI II structure (e.g., Al-Turkait & Ohaeri, 2010; Brouwer et al., 2013; Osman et al., 2008; Quilty et al., 2010), first-order oblique (F-O) and second-order (S-O) hierarchical models were also tested as bifactor models, similar to the method of Al-Turkait and Ohaeri (2010). F-O and S-O models with the same item composition of lower order (i.e., F-O) factors are nested within the related bifactor structure (see Chen et al., 2012). Two models are nested if one is a corresponding subset of the other, in that the model structures are similar and differ only in the number of free parameters included in the model (see Kline, 2011). Thus, a nested comparison test can be used to determine whether there is a significant difference in model fit. As an S-O model with only two F-O factors would be underidentified (Kline, 2011), we did not test this model. To identify each model, we set all factor variances to one in order to establish consistency across the CFAs being compared within the same samples (Kline, 2011). In the second set of analyses, MI testing of the optimal BDI II structure was assessed through increasingly restrictive MG-CFAs, corresponding to configural (factor structure), metric (loadings), and scalar (intercept or item mean) levels of invariance. MI was considered weak if the configural and metric levels were invariant and strong if the configural, metric, and scalar levels were invariant (Meredith, 1993). Strong MI was required in order to conduct group comparisons that would not be inflated or attenuated due to measurement error (Chen, 2008; Little, 1997; Meredith, 1993). In order to identify the models for MI testing, the loading of the first item of each factor was set to one (after running analyses to confirm that these items showed invariant properties). To estimate the unique contributions of each factor to the optimal model, we calculated explained common variance for each cultural group, with explained common variance being equal to the sum of squared factor loadings divided by the sum of squared model loadings (see Brouwer et al., 2013). In the third set of analyses, latent and observed group comparisons were conducted. To unpack group differences across cultural group and gender, two extra pairs of group comparisons were added: CH men versus EH men and CH women versus EH women. Variance and covariance invariance analyses for latent comparisons were planned if weak MI was met (i.e., factor loading invariance; Meredith, 1993). A nested comparison test with one degree of freedom was used to indicate a significant difference in variance across groups. Mean comparisons were planned if strong invariance was met (i.e., scalar or intercept invariance; Meredith, 1993); the mean of the reference group would be constrained to equal zero while being freely estimated in the comparison group, providing a relative difference z statistic for the means across groups, as well as an effect size (Hancock, 2001). For observed analyses, weighted factor scores based on the optimal cultural by gender group model loadings were formed. We then ran between-groups analyses using t tests with Bonferroni adjusted alpha levels of.003 (.05/18) and Levene s test for homogeneity of variance. The patterns of significant results found in the analyses using observed variables versus those using latent variables were compared in order to explore whether the observed analyses failure to control for measurement error led to different results than those found using latent analyses. All CFA models were tested with EQS 6.1 (Bentler, 2005) using maximum likelihood estimation; all other analyses were conducted with SPSS Several indices were used to assess goodness of fit. Because the BDI II scores were expected to violate normality assumptions due to the use of a nonclinical sample (i.e., positive skewness), a scaling correction using Satorra and Bentler s (1994) method was utilized and robust indices were specified. Three other fit indices were selected: root-mean-square error of approximation (RMSEA; Browne & Cudeck, 1993) and 90% confidence interval (90% CI); standardized root-mean-square residual (SRMR; Hu & Bentler, 1999); and comparative fit index (CFI; Bentler, 1990). The quality of each CFA model was evaluated according to the following fit criteria: RMSEA.08, SRMR.10, and CFI.90 for acceptable fit; RMSEA.05, SRMR.08, and CFI.95 for good fit (Browne & Cudeck, 1993; Hu & Bentler, 1999). Given the large sample size, the Satorra Bentler chi-square (S-B 2 ) was used for descriptive purposes only due to the sensitivity of 2 statistics to sample size (Kline, 2011). To compare the quality of nested models (i.e., F-O with bifactor and S-O with bifactor), we utilized change ( ) CFI.01 (Chen, 2007). To compare the quality of non-nested models, we utilized the Akaike information

8 CROSS-CULTURAL EXAMINATION OF THE BDI II 7 Table 2 Descriptive Statistics for BDI-II Items by Cultural Group and Gender Item Chinese heritage (n 933)/ European heritage (n 933) Chinese heritage: men (n 635)/ women (n 298) European heritage: men (n 635)/ women (n 298) M SD r(i-t) t(1864) M SD r(i-t) t(1864) M SD r(i-t) t(1864) 1. Sadness.36/.32.58/.53.66/ /.36.61/.57.67/ /.34.54/.52.61/ Pessimism.42/.38.59/.54.58/ /.44.58/.60.67/ /.39.53/.55.59/ Past failure.42/.25.70/.55.62/ /42.70/.70.67/ /.26.55/.55.64/ Loss of pleasure.33/.27.55/.52.58/ /.33.56/.55.59/ /.24.55/.51.52/ Guilty feelings.43/.42.61/.60.55/ /.42.64/.60.59/ /.42.58/.61.49/ Punishment feelings.23/.14.59/.48.61/ /.23.60/.58.71/ /.12.56/.43.38/ Self-dislike.49/.39.76/.65.73/ /.49.77/.75.76/ /.42.59/.67.67/ Self-criticism.52/.55.73/.72.69/ /.52.75/.72.74/ /.55.75/.70.57/ Suicidal thoughts.15/.11.44/.32.65/ /.15.52/.40.73/ /.11.32/.32.43/ Crying.37/.34.76/.65.59/ /.40.67/.79.60/ /.40.58/.67.56/ Agitation.47/.49.66/.62.62/ /.48.67/.65.63/ /.50.63/.61.48/ Loss of interest.29/.25.56/.52.59/ /.29.56/.56.64/ /.25.51/.52.56/ Indecisiveness.44/.33.71/.59.62/ /.44.70/.71.71/ /.30.59/.59.65/ Worthlessness.30/.20.64/.51.70/ /.32.61/.65.71/ /.21.50/.51.54/ Loss of energy.56/.50.66/.59.70/ /.57.70/.66.72/ /.53.57/.60.62/ Changes in sleep.84/.91.71/.73.48/ /.84.68/.72.46/ /.93.71/.74.49/ Irritability.27/.25.55/.52.69/ /.27.58/.54.77/ /.25.53/.51.60/ Changes in appetite.71/.72.70/.72.46/ /.70.70/.69.39/ /.72.71/.73.46/ Concentration difficulty.54/.48.69/.65.59/ /.55.66/.70.60/ /.51.64/.66.67/ Tiredness.64/.58.62/.60.62/ /.64.64/.61.64/ /.63.58/.61.64/ Loss of interest in sex.17/.15.52/.45.49/ /.18.48/.53.46/ /.16.40/.47.33/ BDI II total 8.93/ / / / / / Note. BDI II Beck Depression Inventory II; M mean; SD standard deviation; r(i-t) item total correlation coefficient; t(df) t statistic (degrees of freedom). p.003, Bonferroni correction. criterion (AIC; Akaike, 1987), in which a smaller number represents the more optimal model. Three statistics were used as indicators of invariance: S-B 2 (Byrne, 2006), where invariance was achieved if S-B 2 was nonsignificant when change in degrees of freedom [ df] was used to determine critical S-B 2 ; CFI; and RMSEA. As S-B 2 was included only for descriptive purposes, significantly better model fit was determined based on CFI.01 and RMSEA.015 (Chen, 2007; Cheung & Rensvold, 2002). The assumption of MI at each level was accepted if CFI and RMSEA did not show significant change between increasingly restrictive MI models. Results Preliminary Analyses Data were screened for outliers, and less than 1% of the sample was found to have total BDI II standardized scores greater than As this percentage was considered minuscule given the large sample size, outliers were not deleted (Tabachnick & Fidell, 2001). As expected, and similar to other student samples (e.g., Wu, 2010; Wu & Huang, 2014), the distributions of several BDI II items were positively skewed. Descriptive statistics (i.e., mean, standard deviation, item total correlation, and t tests) by cultural group and gender within cultural group are displayed in Table 2. T tests showed that CH students scored significantly higher than EH students on four items (past failure, punishment feelings, indecisiveness, and worthlessness). EH women scored higher than men on two items (crying and tiredness), whereas there were no differences between CH men and women. Model Comparisons of BDI II Factor Structures The goodness-of-fit indices for competing factor structures of the BDI II are presented in Table 3. In every instance, the bifactor model had significantly better fit than its related F-O and S-O models (i.e., CFI.01) and a lower AIC value than nonrelated F-O and S-O models. With all of the bifactor models showing adequate fit and minimal difference with respect to the AIC value, inspection of the parameter estimates indicated that the most parsimonious structure was Model 6; this was the sole model containing only significant and interpretable loadings. All other models had multiple nonsignificant and/or negative loadings across culture and culture by gender groups; moreover, several of these models did not have adequate goodness-of-fit indices (analyses available upon request). As such, Model 6 corresponding to Ward s (2006) bifactor structure with a general depression severity factor and two subfactors of cognitive (8 items) and somatic (5 items) domains was chosen as the BDI II structure for all further analyses. The standardized parameter estimates of this baseline model by cultural group are presented in Figure 1. 6 Measurement Invariance Testing The results of MI testing across pairs of comparison groups are presented in Table 4. Strong invariance was achieved across cul- 6 Loadings for culture by gender subgroups are available upon request.

9 8 DERE, WATTERS, YU, BAGBY, RYDER, AND HARKNESS Table 3 Fit Indices for BDI II Factor Structures Outlined in Table 1 (N 933 for Each Group) Model S-B 2 df CFI RMSEA, 90% CI SRMR AIC 1. Chinese-heritage F-O [.032,.024] a. Chinese-heritage S-O [.027,.037] a. Chinese-heritage S-O (Item 21 deleted) [.033,.043] Chinese-heritage bifactor [.025,.036] European-heritage F-O [.028,.038] a. European-heritage S-O [.027,.037] a. European-heritage S-O (Item 21 deleted) [.028,.038] European-heritage bifactor [.025,.035] Chinese-heritage F-O [.037,.046] Chinese-heritage bifactor [.027,.037] European-heritage F-O [.033,.042] European-heritage bifactor [.026,.036] Chinese-heritage F-O [.037,.046] Chinese-heritage bifactor [.026,.036] European-heritage F-O [.034,.043] European-heritage bifactor [.026,.036] Chinese-heritage F-O [.037,.046] Chinese-heritage bifactor [.027,.037] European-heritage F-O [.034,.044] European-heritage bifactor [.025,.035] Chinese-heritage F-O [.038,.047] Chinese-heritage bifactor [.029,.039] European-heritage F-O [.033,.042] European-heritage bifactor [.026,.036] Chinese-heritage bifactor [.027,.037] European-heritage bifactor [.026,.036] Note. All S-B 2 values were significant at p.001. BDI II Beck Depression Inventory II; F-O first order model; S-O second-order model; S-B Satorra Bentler adjusted 2 test statistic; df degrees of freedom; CFI comparative fit index; RMSEA root-mean-square error of approximation; 90% CI 90% confidence interval for RMSEA; SRMR standardized root-mean-square residual; AIC Akaike information criterion. Signifies significantly better fit for bifactor versus related first-order and second-order model [i.e., change ( ) statistic exceeds critical value for S-B 2 ( when df 18 and when df 20) and CFI.01]. tural groups and for gender within each cultural group, meeting the requirements for latent comparison testing for all pairs of comparisons. CFI and RMSEA were nonsignificant in all instances, supporting both weak (i.e., factor loading invariance) and strong (i.e., scalar or intercept invariance; Meredith, 1993) levels of MI across cultural group and gender within cultural group. Explained common variance values were also similar across the CH and EH groups, with the general depression factor explaining 82% and 80%, the cognitive factor explaining 12% and 11%, and the somatic factor explaining 7% and 9% of the total variance in the two groups, respectively. 7 Latent and Observed Group Comparisons Variance invariance. Model fit and the results of latent variance invariance testing for Model 6 (Ward, 2006) are displayed in Table 4; the variance of all factors (i.e., general depression, cognitive, somatic) was equivalent for all five pairs of comparison groups. In contrast, Levene s test for the homogeneity of variance demonstrated that several pairs of comparison groups violated this assumption. The CH group showed significantly more variance on the general depression factor (Levene statistic), F(1, 1864) 27.06, p.001, and on the cognitive subfactor, F(1, 1864) 57.29, p.001, than did the EH group. The same pattern of results occurred for the CH versus EH women, F(1, 1268) 11.73, p.001; F(1, 1268) 50.73, p.001, respectively, and for the CH versus EH men, F(1, 594) 17.12, p.001; F(1, 594) 6.24, p.013, respectively. Within the CH group, women showed significantly more variance on the cognitive subfactor than did men, F(1, 931) 15.92, p.001. Because there were no significant differences in variance across groups when running latent mean comparisons, which control for measurement error, these results indicate that the violations of variance homogeneity were largely due to measurement error that is not controlled for with observed analyses. Mean comparisons. Fit indices and tests of latent mean differences are presented in Table 5. Descriptive statistics (i.e., mean, standard deviation, and coefficient alphas) for the BDI II factors by cultural group and gender within cultural group are displayed in Table 6, with a summary of significant latent and observed results. Both latent and observed comparisons found that the CH group scored significantly higher on the cognitive factor than the EH group, z 3.04, d.24; t(1864) 4.74, p.001, d.22, whereas no significant cultural group difference was found on the somatic factor. Further analyses showed that this difference was 7 Because several of the baseline models showed adequate fit across all fit indices (i.e., Model 1, F-O, S-O, and bifactor versions; bifactor version of Models 2 through 6), we also ran supplementary analyses to test the MI of these models (see Tables S1 through S6 in the online supplemental materials). In all cases, evidence of strong MI (Meredith, 1993) was found, across both culture and gender within culture. These results further suggest that the BDI II shows particularly robust invariance.

10 CROSS-CULTURAL EXAMINATION OF THE BDI II 9 1. Sadness (.66,.59) 2. Pessimism (.50,.53) 3. Past failure (.51,.50) 4. Loss of pleasure (.58,.60) 5. Guilty feelings (.46,.40) 6. Punishment feelings (.55,.38) 7. Self-dislike (.66,.60) 8. Self-criticalness (.62,.49) (.33,.27) (.53,.39).90,,.80 (.27,.26) (.33,.21) (.38,.39) (.32,.36) (.21,.22) Cognitive General Depression 9. Suicidal thoughts (.62,.49) 10. Crying (.56,.51) 11. Agitation (.60,.48) 12. Loss of Interest (.61,.63) 13. Indecisiveness (.60,.61) 14. Worthlessness (.64,.56) 15. Loss of energy (.66,.58) 16. Changes in sleep (.39,.37) 17. Irritability (.71,.61) 18. Changes in appetite (.40,.37) 19. Concentration difficulty (.54,.58) 20. Tiredness (.57,.57) 21. Loss of interest in Sex (.48,.41) due to CH women scoring higher than EH women on cognitive symptoms (z 2.51, d.20). There were also several discrepant findings between the latent and observed analyses (see Table 6). Latent comparisons found that EH women scored higher than EH men on the somatic factor (z 2.78, d.031). Observed comparisons found that the CH group scored higher than the EH group on the general factor, t(1864) 4.39, p.001, d.20. More specifically, the CH men scored higher than the EH men on the general factor, t(1268) 4.74, p.001, d.27. The CH men scored lower than the CH women on the cognitive factor, t(931) 3.42, p.001, d.024. Although these group differences were small with respect to effect size, these results highlight the fact that observed analyses, which do not control for measurement error, can lead to different results from latent analyses, which do. In other words, significant group differences that are found using observed analyses (e.g., t tests, ANOVA), particularly those with small effect sizes, should be interpreted with caution, as such results could be (.35,.39) (.31,.38) (.36,.32) (.21,.25) (.22,.21) (.49,.54) Somatic Figure 1. Standardized parameter estimates for Chinese-heritage (CH) and European-heritage (EH) Beck Depression Inventory II baseline factor models (Ward, 2006). Numbers in parentheses are CH followed by EH loading estimates; all parameter estimates are significant at p.05. Explained common variance percentages for CH and EH groups: general factor (82, 80); cognitive (12, 11); somatic (7, 9). due to artifacts of measurement error rather than true group differences. Discussion In the current study we attempted to integrate several recent lines of research to provide a statistically rigorous examination of potential differences in depressive symptom reporting between Chinese- and European-heritage students using the BDI II. Our investigation builds upon recent work in the areas of cultural variation in depressive symptoms, the use of a bifactor approach to the factor structure of the BDI II, and the application of advanced statistical techniques in testing measurement invariance. We think that our findings not only contribute uniquely to each of these research areas but also have important methodological implications for cross-cultural psychopathology research more broadly. We also think that the findings of the current investigation contribute to the ongoing debate regarding the optimal factor

11 10 DERE, WATTERS, YU, BAGBY, RYDER, AND HARKNESS Table 4 Fit Indices and Difference Statistics for Measurement Invariance Models (Configural, Metric, Scalar) by Cultural Group and Gender Within Cultural Group Model S-B 2 df CFI RMSEA, 90% CI SRMR S-B 2 df CFI RMSEA Chinese heritage (n 933), European heritage (n 933) Configural (structure) [.028,.035].035 Metric (loadings) [.028,.034] Scalar (intercepts) [.028,.035] Variance depression [.029,.036] Variance cognitive [.030,.036] Variance somatic [.028,.035] Chinese heritage: men (n 298), women (n 635) Configural (structure) [.028,.039].042 Metric (loadings) [.028,.039] Scalar (intercepts) [.027,.038] Variance depression [.029,.039] Variance cognitive [.028,.038] Variance somatic [.029,.040] European heritage; men (n 298), women (n 635) Configural (structure) [.026,.038].045 Metric (loadings) [.025,.036] Scalar (intercepts) [.027,.038] Variance depression [.026,.037] Variance cognitive [.026,.037] Variance somatic [.026,.037] Note. S-B 2 Satorra Bentler adjusted chi-square test statistic; df degrees of freedom; CFI comparative fit index; RMSEA root-mean-square error of approximation; 90% CI 90% confidence interval for RMSEA; SRMR standardized root-mean-square residual; S-B 2 change in S-B 2 based on Byrne (2006) equation (asterisk indicates significant 2 value based on df change in df; where 21 critical value 32.67, 30 critical value 43.77, and 1 critical value 3.84); CFI change in CFI (asterisk indicates.01); RMSEA change in RMSEA (asterisk indicates.015); comparison analyses include (a) metric versus configural; (b) scalar versus metric; (c) variance versus metric model. structure of the BDI II. All bifactor models showed significantly improved fit (based on change in CFI of.01) over their nested first- and second-order counterparts, across all culture and gender subgroups. This result fits with those of other recent studies (e.g., Al-Turkait & Ohaeri, 2010; Brouwer et al., 2013; Osman et al., 2008; Quilty et al., 2010) to provide further evidence for the superiority of a bifactor approach to the BDI II. As this study was the first to test a bifactor model in a Chinese-heritage sample, our findings extend those of previous studies and add to a growing body of literature supporting the idea that depression is marked by a general factor of depression severity while simultaneously containing multiple components (e.g., Simms, Grös, Watson, & O Hara, 2008): the general factor accounted for approximately 80% of the explained common variance in BDI II scores. In terms of practical implications, this suggests that the calculation of a total BDI II score is useful and valid, a finding of considerable relevance to researchers and clinicians alike (Brouwer et al. 2013; Quilty et al., 2010). Statistically, a bifactor model allows for the investigation of dimensions of depression (e.g., cognitive and somatic) that contribute unique variance beyond the general severity factor, which other approaches such as second-order models do not (Chen et al., 2012). From a cross-cultural research perspective, such an approach is desirable in its ability to parse apart the variance associated with overall severity from that attributable to narrower symptom domains, which tend to be the focus of attention. As researchers seek to uncover those variables that contribute to cultural variation in depressive symptom reporting, stable and robust factor structures of the measures being used to assess depression are crucial. With regard to the cross-cultural use of the BDI II, we found strong measurement invariance (i.e., configural, metric, and scalar levels of invariance; Meredith, 1993) in our two groups, across both culture and gender. This finding is notable in terms of the strength of the invariance results and suggests that the BDI II represents a good choice as a self-report measure of depressive symptoms, at least among these two groups. It is important to note that the Chinese-heritage students in our study are likely to differ from other Chinese groups in North America and elsewhere, based on factors such as age, education level, and the level of Englishlanguage proficiency required to enter their university. Similarly, it is worth noting that all participants completed the BDI II in English, though Byrne et al. (2007) also found support for strong invariance comparing the English- and Chinese-language versions of the BDI II. The use of a non-help-seeking student sample in the current study limits the generalizability of these findings to samples with greater levels of depression severity or those in clinical settings. Although we cannot unequivocally recommend the use of the BDI II in cross-cultural comparisons of nonstudent or treatment-seeking samples, when combined with the findings of Byrne et al. (2007) and Whisman et al. (2013) our results suggest that there is mounting evidence that the BDI II shows robust measurement invariance.

12 CROSS-CULTURAL EXAMINATION OF THE BDI II 11 Table 5 Tests for Latent Mean Differences by Cultural Group, Gender Within Cultural Group, and Gender Across Cultural Group (Women, N 635, and Men, N 298, for Chinese-Heritage and European-Heritage Groups) Model and subscales S-B 2 CFI RMSEA, 90% CI SRMR Difference estimate z Chinese heritage, European heritage; full sample (reference group European heritage) General depression [.028,.035] Cognitive subscale Somatic subscale Chinese heritage; men, women (reference group men) General depression [.029,.040] Cognitive subscale Somatic subscale European heritage; men, women (reference group men) General depression [.028,.039] Cognitive subscale Somatic subscale Chinese heritage, European heritage; men (reference group Chinese heritage) General depression [.028,.043] Cognitive subscale Somatic subscale Chinese heritage, European heritage; women (reference group European heritage) General depression [.027,.036] Cognitive subscale Somatic subscale Note. S-B 2 Satorra Bentler adjusted chi-square test statistic; df degrees of freedom; CFI comparative fit index; RMSEA root-mean-square error of approximation; 90% CI 90% confidence interval for RMSEA; SRMR standardized root-mean-square residual; reference group group with latent mean constrained to be zero, where the difference estimate is in relation to the reference group and the z value represents the strength of this difference estimate. p.05. Examining the issue of cultural variation in symptom reporting, the current results are not consistent with the ideas of either Chinese somatization or Western psychologization. We found, based on both the latent and measured group comparisons, that Chinese-heritage students scored higher on the cognitive factor than did European-heritage students. Additional group comparisons revealed that this result was largely due to Chinese-heritage women scoring higher than European-heritage women on the cognitive factor, whereas no difference was found between men across cultural group. This result echoes the earlier discussion regarding a lack of consistent support for the idea of Chinese somatization and discrepancies between clinical and nonclinical samples (e.g., Yen et al., 2000). This finding also fits with previous studies that have found a greater emphasis on cognitive symptoms among Chinese-heritage adolescents and young adults under certain circumstances (Chang, 2007; Stewart et al., 2002). The cultural group difference on the cognitive factor of the BDI II was the only result that was consistent across the latent and observed group comparisons. Discrepancies between the analyses using latent versus observed variables highlight the importance of controlling for measurement error, as failing to do so can potentially lead to spurious results, artifacts of measurement error, and group differences that do not truly exist. A recommendation that emerges from these findings, therefore, is for researchers conducting group comparisons (cultural and otherwise) to make use of the MG-CFA framework when possible. When sample size constraints do not allow for this approach, researchers are urged to be wary of group differences with small effect sizes, which may be artifacts of measurement error. Given the lack of additional variables in the current study that might help to explain our group comparison findings, any proposed explanation remains speculative. However, one issue that may be important to consider is the academic context in which the data were collected; all participants were first-year undergraduates attending a prestigious Canadian university with a highly competitive admission process. Other authors have previously highlighted the important value that tends to be placed on academic success in traditional East Asian cultural contexts and the potential link to depressive symptomatology (Stewart et al., 2002; Young et al., 2010). It is possible that cognitive symptoms (e.g., past failures) would be especially salient among students for whom academic success is particularly important, who may also find any first-year academic challenges to be particularly distressing. This line of inquiry may be a fruitful avenue for future research. As our results add to the pattern of discrepant findings regarding Chinese somatization and Western psychologization between clinical and nonclinical samples, it is worth briefly reflecting on potential explanations for this discrepancy. A number of authors

13 12 DERE, WATTERS, YU, BAGBY, RYDER, AND HARKNESS Table 6 Descriptive Statistics for the BDI II, Ward s (2006) Bifactor Model, by Total Sample and Gender Measure CH (n 933) EH (n 933) CH men (n 298) CH women (n 635) EH men (n 298) EH women (n 635) BDI II total (21 items) M 8.93 a 8.04 a SD 8.13 a 6.83 a 8.59 d 7.91 e 6.59 d 6.91 e Cronbach s alpha MIC Cognitive (8 items) M 2.97 A 2.45 A E E SD 3.62 a 2.87 a 3.87 d 3.51 e 2.78 d 2.91 e Cronbach s alpha MIC Somatic (5 items) M e 2.93 C 3.32 C,e SD Cronbach s alpha MIC Note. A superscript lowercase letter indicates significant observed mean difference (based on t-test statistic) or violation of homogeneity of variance (based on Levene s statistic); a superscript bold and capital letter indicates significant latent and observed mean or variance difference. BDI II Beck Depression Inventory II; CH Chinese heritage; EH European heritage; MIC mean interitem correlation. Comparison groups: a CH/EH; b CH men/women; c EH men/women; d CH/EH men; e CH/EH women. have pointed toward help seeking and the patient role as important explanatory factors contributing to cultural differences in depressive symptom reporting among Chinese and Western groups (e.g., Chang, 2007; Ryder et al., 2008; Yen et al., 2000). Others have also suggested that cultural variations in symptom reporting reflect culturally shaped differences in the types of symptoms that are experienced as most salient and troubling during times of profound distress (Ryder & Chentsova-Dutton, 2012). The current data were collected in a classroom setting from students who were unlikely to be experiencing severe levels of depressive symptoms, suggesting that neither help seeking nor experiences of profound distress would be particularly applicable factors in their pattern of symptom reporting. Further work is required to understand more clearly and precisely those variables that play a role in cultural variations in depressive symptom reporting across both clinical and nonclinical samples. Future work should also explore the potential role of acculturation and related processes in helping to understand the cultural shaping of depressive symptom reporting. Acculturation is a particularly relevant construct when examining samples that include different generations of immigrants and migrants, such as international students. Other processes relating to cultural change, such as modernization, would also be an important area of future study. In particular, recent research on depression among Chinese-heritage samples highlights the need to examine historical forces in better understanding apparent shifts in the prevalence and symptom presentation of depression in Chinese cultural contexts in recent decades (see, e.g., Ryder, Sun, Zhu, Yao, & Chentsova-Dutton, 2012). The influence of Western psychiatric models of depression, along with related norms concerning emotional expression and psychologization, has been discussed as an influential factor in mainland China (e.g., Lee & Kleinman, 2007); such forces are likely to be all the more influential for young adults of Chinese heritage who are born, raised, and/or studying in North America. Indeed, such factors may help to explain the lack of evidence for Chinese somatization in the current results. As suggested by the proposed areas of future research, an important limitation of the current study is the lack of more detailed sociodemographic information about our participants, as well as a lack of additional variables that could serve to help explain our findings more fully. Although the use of an archival database allowed for the large samples sizes required for our statistical analyses, it limited our ability to gather such variables. It might also be argued that the comparison of a Chinese-heritage sample of university undergraduates in North America with European-heritage counterparts is not a particularly compelling cross-cultural comparison. However, previous research has demonstrated significant differences between Asian Americans and European Americans on a variety of fundamental emotion-related processes, including ones specific to depression (e.g., Chentsova- Dutton et al., 2007), suggesting that important differences can emerge from such comparisons. Furthermore, the study of North American undergraduates of diverse cultural backgrounds is crucial to ongoing efforts to improve the ability to assess and treat mental health concerns on university campuses. References Akaike, H. (1987). Factor analysis and AIC. Psychometrika, 52, doi: /bf Al-Turkait, F. A., & Ohaeri, J. U. (2010). Dimensional and hierarchical models of depression using the Beck Depression Inventory-II in an Arab college student sample. BMC Psychiatry, 10, Article 60. doi: / X American Psychiatric Association. (1994). Diagnostic and statistical manual of mental disorders (4th ed.). Washington, DC: Author. Bayram, N., & Bilgel, N. (2008). The prevalence and socio-demographic correlations of depression, anxiety and stress among a group of univer-

14 CROSS-CULTURAL EXAMINATION OF THE BDI II 13 sity students. Social Psychiatry and Psychiatric Epidemiology, 43, doi: /s x Beck, A. T., Steer, R. A., Ball, R., & Ranieri, W. F. (1996). Comparison of Beck Depression Inventories IA and II in psychiatric outpatients. Journal of Personality Assessment, 67, doi: / s jpa6703_13 Beck, A. T., Steer, R. A., & Brown, G. K. (1996). Manual for the Beck Depression Inventory 2. San Antonio, TX: Psychological Corporation. Bentler, P. M. (1990). Comparative fit indices in structural models. Psychological Bulletin, 107, doi: / Bentler, P. M. (2005). ESQ 6.1 structural equations program manual. Los Angeles, CA: Multivariate Software. Brouwer, D., Meijer, R. R., & Zevalkink, J. (2013). On the factor structure of the Beck Depression Inventory II: G is the key. Psychological Assessment, 25, doi: /a Browne, M. W., & Cudeck, R. (1993). Alternative ways of assessing model fit. In K. A. Bollen & J. S. Long (Eds.), Testing structural equation models (pp ). Newbury Park, CA: Sage. Byrne, B. M. (2006). Structural equation modeling with EQS (2nd ed.). New York, NY: Psychology Press. Byrne, B. M., Stewart, S. M., Kennard, B. D., & Lee, P. W. H. (2007). The Beck Depression Inventory-II: Testing for measurement equivalence and factor mean difference across Hong Kong and American adolescents. International Journal of Testing, 7, doi: / Chang, H. (2007). Depressive symptom manifestation and help-seeking among Chinese college students in Taiwan. International Journal of Psychology, 42, doi: / Chen, F. F. (2007). Sensitivity of goodness of fit indexes to lack of measurement invariance. Structural Equation Modeling, 14, doi: / Chen, F. F. (2008). What happens if we compare chopsticks with forks? The impact of making inappropriate comparisons in cross-cultural research. Journal of Personality and Social Psychology, 95, doi: /a Chen, F. F., Hayes, A., Carver, C. S., Laurenceau, J.-P., & Zhang, Z. (2012). Modeling general and specific variance in multifaceted constructs: A comparison of the bifactor model to other approaches. Journal of Personality, 80, doi: /j x Chentsova-Dutton, Y. E., Chu, J. P., Tsai, J. L., Rottenberg, J., Gross, J., & Gotlib, I. H. (2007). Depression and emotional reactivity: Variation among Asian Americans and European Americans. Journal of Abnormal Psychology, 116, doi: / x Cheung, G. W., & Rensvold, R. B. (2002). Evaluating goodness-of-fit indexes for testing measurement invariance. Structural Equation Modeling, 9, doi: /s sem0902_5 Cuijpers, P., Beekman, A. T. F., & Reynolds, C. F., III. (2012). Preventing depression: A global priority. JAMA: Journal of the American Medical Association, 307, doi: /jama Dere, J., Sun, J., Zhao, Y., Persson, T. J., Zhu, X., Yao, S.,...Ryder, A. G. (2013). Beyond somatization and psychologization : Symptom-level variation in depressed Han Chinese and Euro-Canadian outpatients. Frontiers in Psychology, 4, Article 377. doi: /fpsyg Dozois, D. J. A., Dobson, K. S., & Ahnberg, J. L. (1998). A psychometric evaluation of the Beck Depression Inventory II. Psychological Assessment, 10, doi: / Eisenberg, D., Gollust, S. E., Golberstein, E., & Hefner, J. L. (2007). Prevalence and correlates of depression, anxiety, and suicidality among university students. American Journal of Orthopsychiatry, 77, doi: / Hancock, G. R. (2001). Effect size, power, and sample size determination for structured means modeling and MIMIC approaches to betweengroups hypothesis testing of means on a single latent construct. Psychometrika, 66, doi: /bf Horn, L., Peter, K., Rooney, K., & Malizio, A. G. (2002). Profile of undergraduates in U.S. postsecondary institutions: Washington, DC: U. S. Department of Education, Office of Educational Research and Improvement, National Center for Education Statistics. Hu, L. T., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6, doi: / Ibrahim, A. K., Kelly, S. J., Adams, C. E., & Glazebrook, C. (2013). A systematic review of studies of depression prevalence in university students. Journal of Psychiatric Research, 47, doi: /j.jpsychires Kessler, R. C., Berglund, P., Demler, O., Jin, R., Koretz, D., Merikangas, K. R.,... Wang, P. S. (2003). The epidemiology of major depressive disorder: Results from the National Comorbidity Survey Replication (NCS-R). JAMA: Journal of the American Medical Association, 289, doi: /jama Kirmayer, L. J. (2001). Cultural variations in the clinical presentation of depression and anxiety: Implications for diagnosis and treatment. Journal of Clinical Psychiatry, 62, Kleinman, A. (1982). Neurasthenia and depression: A study of somatization and culture in China. Culture, Medicine and Psychiatry, 6, Kline, R. B. (2011). Principles and practice of structural equation modeling (3rd ed.). New York, NY: Guilford Press. Lam, C. Y., Pepper, C. M., & Ryabchenko, K. A. (2004). Case identification of mood disorders in Asian American and Caucasian American college students. Psychiatric Quarterly, 75, doi: /b: PSAQ b Lee, S., & Kleinman, A. (2007). Are somatoform disorders changing over time? The case of neurasthenia in China. Psychosomatic Medicine, 69, doi: /psy.0b013e31815b0092 Little, T. D. (1997). Mean and covariance structures (MACS) analyses of cross-cultural data: Practical and theoretical issues. Multivariate Behavioral Research, 32, doi: /s mbr3201_3 Mathers, C. D., & Loncar, D. (2006). Projections of global mortality and burden of disease from 2002 to PLoS Medicine, 3(11), e442. doi: /journal.pmed Meredith, W. (1993). Measurement invariance, factor analysis and factorial invariance. Psychometrika, 58, doi: /bf Milfont, T. L., & Fischer, R. (2010). Testing measurement invariance across groups: Applications in cross-cultural research. International Journal of Psychological Research, 3, Osman, A., Barrios, F. X., Gutierrez, P. M., Williams, J. E., & Bailey, J. (2008). Psychometric properties of the Beck Depression Inventory II in nonclinical adolescent samples. Journal of Clinical Psychology, 64, doi: /jclp Parker, G., Chan, B., Tully, L., & Eisenbruch, M. (2005). Depression in the Chinese: The impact of acculturation. Psychological Medicine, 35, doi: /s Parker, G., Cheah, Y.-C., & Roy, K. (2001). Do the Chinese somatize depression? A cross-cultural study. Social Psychiatry and Psychiatric Epidemiology, 36, doi: /s Patten, S. B., Wang, J. L., Williams, J. V. A., Currie, S., Beck, C. A., Maxwell, C. J., & el-guebaly, N. (2006). Descriptive epidemiology of major depression in Canada. Canadian Journal of Psychiatry/Revue canadienne de psychiatrie, 51, Quilty, L. C., Zhang, K. A., & Bagby, R. M. (2010). The latent symptom structure of the Beck Depression Inventory II in outpatients with major depression. Psychological Assessment, 22, doi: / a Radloff, L. S. (1977). The CES-D scale: A self-report depression scale for research in the general population. Applied Psychological Measurement, 1, doi: /

15 14 DERE, WATTERS, YU, BAGBY, RYDER, AND HARKNESS Reise, S. P., Bonifay, W. E., & Haviland, M. G. (2013). Scoring and modeling psychological measures in the presence of multidimensionality. Journal of Personality Assessment, 95, doi: / Ryder, A. G., & Chentsova-Dutton, Y. E. (2012). Depression in cultural context: Chinese somatization, revisited. Psychiatric Clinics of North America, 35, doi: /j.psc Ryder, A. G., Sun, J., Zhu, X., Yao, S., & Chentsova-Dutton, Y. E. (2012). Depression in China: Integrating developmental psychopathology and cultural-clinical psychology. Journal of Clinical Child and Adolescent Psychology, 41, doi: / Ryder, A. G., Yang, J., Zhu, X., Yao, S., Yi, J., Heine, S. J., & Bagby, R. M. (2008). The cultural shaping of depression: Somatic symptoms in China, psychological symptoms in North America? Journal of Abnormal Psychology, 117, doi: / x Satorra, A., & Bentler, P. M. (1994). Corrections to test statistics and standard errors in covariance structure analysis. In A. von Eye & C. C. Clogg (Eds.), Latent variables analysis: Applications for developmental research (pp ). Thousand Oaks, CA: Sage. Simms, L. J., Grös, D. F., Watson, D., & O Hara, M. W. (2008). Parsing the general and specific components of depression and anxiety with bifactor modeling. Depression and Anxiety, 25, E34 E46. doi: / da Stewart, S. M., Lewinsohn, P. M., Lee, P. W., Ho, L. M., Kennard, B., Hughes, C. W., & Emslie, G. J. (2002). Symptom patterns in depression and subthreshold depression among adolescents in Hong Kong and the United States. Journal of Cross-Cultural Psychology, 33, doi: / Storch, E. A., Roberti, J. W., & Roth, D. A. (2004). Factor structure, concurrent validity, and internal consistency of the Beck Depression Inventory Second Edition in a sample of college students. Depression and Anxiety, 19, doi: /da Tabachnick, B. G., & Fidell, L. S. (2001). Using multivariate statistics (5th ed.). Boston, MA: Allyn & Bacon. Üstün, T. B., Ayuso-Mateos, J. L., Chatterji, S., Mathers, C., & Murray, C. J. L. (2004). Global burden of depressive disorders in the year British Journal of Psychiatry, 184, doi: /bjp van de Vijver, F. J. R., & Leung, K. (2000). Methodological issues in psychological research on culture. Journal of Cross-Cultural Psychology, 31, doi: / Vanheule, S., Desmet, M., Groenvynck, H., Rosseel, Y., & Fontaine, J. (2008). The factor structure of the Beck Depression Inventory-II. Assessment, 15, doi: / Ward, L. C. (2006). Comparison of factor structure models for the Beck Depression Inventory II. Psychological Assessment, 18, doi: / Watters, C. A., Keefer, K. V., Kloosterman, P. H., Summerfeldt, L. J., & Parker, J. D. A. (2013). Examining the structure of the Internet Addiction Test in adolescents: A bifactor approach. Computers in Human Behavior, 29, doi: /j.chb Whisman, M. A., Juss, C. M., Whiteford, N. T., & Gelhorn, H. L. (2013). Measurement invariance of the Beck Depression Inventory Second Edition (BDI-II) across gender, race, and ethnicity in college students. Assessment, 20, doi: / Whisman, M. A., Perez, J. E., & Ramel, W. (2000). Factor structure of the Beck Depression Inventory Second Edition (BDI-II) in a student sample. Journal of Clinical Psychology, 56, doi: / (SICI) (200004)56:4 545::AID-JCLP7 3.0.CO;2-U Wiebe, J. S., & Penley, J. A. (2005). A psychometric comparison of the Beck Depression Inventory II in English and Spanish. Psychological Assessment, 17, doi: / Wu, P. (2010). Measurement invariance and latent mean differences of the Beck Depression Inventory II across gender groups. Journal of Psychoeducational Assessment, 28, doi: / Wu, P.-C., & Huang, T.-W. (2014). Gender-related invariance of the Beck Depression Inventory II for Taiwanese adolescent samples. Assessment, 21, doi: / Yen, S., Robins, C. J., & Lin, N. (2000). A cross-cultural comparison of depressive symptom manifestation: China and the United States. Journal of Consulting and Clinical Psychology, 68, doi: / x Young, C. B., Fang, D. Z., & Zisook, S. (2010). Depression in Asian- American and Caucasian undergraduate students. Journal of Affective Disorders, 125, doi: /j.jad Zhou, X., Dere, J., Zhu, X., Yao, S., Chentsova-Dutton, Y. E., & Ryder, A. G. (2011). Anxiety symptom presentations in Han Chinese and Euro-Canadian outpatients: Is distress always somatized in China? Journal of Affective Disorders, 135, doi: /j.jad Received October 9, 2013 Revision received July 23, 2014 Accepted August 1, 2014