Category: Adult Anxiety
Background: An emerging debate in the psychometric literature concerns the limitations of applying independent clusters confirmatory factor analysis to estimate the relationships among psychological constructs as latent variables (Marsh, Ludtke, Nagengast, Morin & von Davier 2013). According to this critique, traditional structural equation modeling (SEM) draws on an overly restricted measurement framework with confirmatory factor analysis (i.e., no cross-loadings) and the problematic assumption of unidimensional factors with item parcels (i.e., sums of item responses). While such restrictions could build more parsimonious models to represent constructs of homogenous populations, they become less tenable in research interested in ascertaining accurately the heterogeneity among ethnic groups (e.g., SEM often leads to inflated estimates of factor correlations). To address these issues, Marsh and colleagues (2013) have proposed an alternative approach of exploratory structural equation model (ESEM), which integrates the strengths of EFA, CFA, and SEM to enable the testing of measurement invariance among multiple groups and more accurately estimate the correlations among latent psychological constructs. Adopting this flexible ESEM framework, this study re-examines the diathesis-stress model of vulnerabilities and anxiety symptoms, previously developed using a SEM approach, among three ethnic groups: Asians/Asian Americans (n = 197), Latinos/Latino-American (n = 207), and European Americans (n = 513).
Method: This is a follow-up study that compares prior SEM findings with a re-analysis of the same data within an ESEM framework. An ethnically diverse sample of adult participants was originally recruited from the undergraduate population, professionals in the Washington DC metropolitan area, and Mechanical Turk. Subjects were assessed with a battery of questionnaires: Ethnic identity was measured by Multigroup Ethnic Identity Measure – Revised; worry by Penn State Worry Questionnaire; social anxiety symptoms by Social Interaction Anxiety Scale and Social Phobia Scale; autonomic arousal by Beck Anxiety Inventory; trait vulnerabilities by Positive and Negative Affect Scale. ESEM and SEM estimates are compared for the following estimates: the relationships among the constructs of ethnic identity, trait vulnerabilities on the one hand, and the symptoms of anxiety on the other; the effects of the degree of ethnic identity (i.e. MEIM-R) within each groups.
Results: At the measurement model level, factor structures with and without cross-loadings of items for the individual measures revealed measurement variance among the groups. The comparisons at the structural regression level evidenced improved goodness of fit from the ESEM models and generally inflated relationships among the constructs from the SEM models.
Conclusion: Differences between the approaches indicate that the ESEM models could more accurately represent the relative independence of vulnerabilities and anxiety constructs. The practical implications of these findings and the general relevance of adopting the use of ESEM for ethnic minority research, especially for the investigation of population differences, are discussed.