Measuring depression in adolescence: Evaluation of a hierarchical factor model of the Children’s Depression Inventory and measurement invariance across boys and girls

Abstract

Background One of the most widely used instruments to measure depression in childhood and adolescence is Kovacs’s Children’s Depression Inventory (CDI). Even though this particular measure sparked massive interest among researchers, there is no clear consensus about its factorial structure. It has been suggested that inconsistencies in findings can be partly ascribed to the cultural context. The aim of this study was a) to examine and verify the factor structure of CDI in the Czech population and b) to assess gender-related psychometric differences using the mean and covariance structure (MACS) approach and differential item functioning (DIF) analysis. Methods The research sample consisted of 1,515 adolescents (ages 12 to 16 years, 53.7% female) from a non-clinical general population. Based on exploratory factor analysis (EFA) on a random subsample (N = 500), we proposed a model that was subsequently tested on the rest of the sample (N = 1,015) using confirmatory factor analysis (CFA). Following the MACS procedure, we assessed measurement invariance in boys and girls. The between-group comparison was further supplemented by a DIF analysis. Results The proposed hierarchical four-factor model (General Symptoms, Negative Self-Concept, Inefficiency, and Social Anhedonia) with a second-order factor of depression fitted the data reasonably well (χ2 = 1281.355; df = 320; RMSEA = 0.054, CFI = 0.925). Regarding gender differences, we found no substantial signs of measurement invariance using the MACS approach. Boys and girls differed in first-order latent means (girls scored higher on General Symptoms with a standardized mean difference of 0.52 and on Negative Self-Concept with a standardized mean difference of 0.31). DIF analysis identified three items with differential functioning. However, the levels of differential functioning were only marginal (in two items) or marginal/moderate and the presence of DIF does not substantially influence scoring of CDI. Conclusion In the general adolescent population in the Czech Republic, the CDI can be considered a reliable instrument for screening purposes in clinical settings and for use in research practice. Instead of the originally proposed five-factor model, we recommend using the newly established four-factor structure. The measure seems to show only marginal psychometric differences with respect to gender, and overall measurement invariance in boys and girls seems to be a tenable assumption.