Time and gender measurement invariance in the modified Calderon depression scale

Abstract

Background Assessing change and comparing groups requires high quality and invariant scales. However, there is limited evidence of simultaneous longitudinal and gender measurement invariance for depression scales. This evidence is even more scant with long-established panel studies from low and middle-income countries. Methods In this paper, we used three waves (years 2002, 2005, and 2009) of a nationally representative panel study to examine the psychometric properties of the modified Calderon Depression Scale (CAL-DM)—a one-item exclusion of a depression scale designed for a population residing in a middle-income country (i.e., Mexico). Our analytical sample included 16,868 participants: 7,696 men and 9,172 women. Using Confirmatory Factor Analysis (CFA), we first examined overall fit in each wave, and then we tested time, gender, and time-gender measurement invariance across three waves. We also estimated and compared depression score means by gender and time. Finally, we examined the association between depression scores and self-rated health. Results Our analyses indicated the CAL-DM is a robust scale, suitable for time, gender, and time by gender comparisons. Mean comparisons exemplified how the scale can be used as a latent variable or a summative score. Women have higher depression scores than men and the gap is narrowing from 3.4 in 2002 to 2.5 in 2009. Conclusions The CAL-DM is a reliable instrument to measure depression in the Mexican general population that can be used for epidemiological research. Our results will contribute to a burgeoning line of research that examines the social determinants of depression, and the risk factors associated with different individuals’ depression trajectories over the life course.

Longitudinal measurement invariance of the Patient Health Questionnaire-9 across racial/ethnic groups: Results from the traumatic brain injury model system study.

“…The longitudinal MI analyses comprised a series of models with increasing constraints estimated using a CFA framework (Liu et al, 2017). We extended the steps proposed by Li et al (Liu et al, 2017), following the analytic structure of Arenas et al (2022). Therefore, we started with the configural model and subsequently added further constraints: (a) thresholds, (b) loadings, (c) intercepts, (d) residuals, and (e) latent means.…”