Item-Level Refinement of the Short Center for Epidemiologic Studies Depression Scale: A Rasch Validation and Revision

Background

Depression remains a significant public health concern, with the point prevalence rate average estimated at around 15.4% between 2004 and 2014 (Lim et al., 2018). Different prevalence rates were also observed among women (14.4%), self-reported cases (17.3%), and individuals with cancer (32%; Lim et al., 2018; Massie, 2004). Reliable and valid measurement tools are essential for identifying and addressing mental health needs. The Center for Epidemiologic Studies-Depression Scale (CES-D; Radloff, 1977) is a cross-validated scale measuring a range of depressive severities. The single-factor short CES-D-10 (Powers et al., 2002) exhibits high reliability across age groups (Cosco et al., 2017; Haroz et al., 2014; Mohebbi et al., 2018). However, conventional validation methods (e.g., exploratory and confirmatory factor analyses) are considered sample-dependent, raising concerns about generalizability across diverse populations (Coste et al., 2014). To increase measurement invariance across populations, this study used the sample-independent Rasch modeling method to validate the CES-D-10.

Method

Data were obtained from the Athletic Involvement Study (Miller, 2013). A total of 795 students were recruited from a large northeastern university in the United States. Most participants identified as male (56.7%). Participants completed the CES-D-10, a 10-item scale assessing depressive symptoms on a 4-point Likert scale ranging from 1 (none of the time) to 4 (all the time). Items included “I felt lonely” and “My sleep was restless” (Andresen et al., 1994; Radloff, 1977). Analyses of item Infit and Outfit (recommended values between 0.5 and 1.5), Principal Component Analysis (PCA; λ < 2), Yen’s Q3 (local dependence; aQ3 < 0.2), and logistic Differential Item Functioning test (DIF) using Chi-square criterion (α = 0.01) were performed using Rating Scale Rasch Model with Marginal Maximum Likelihood estimation (Linacre & Wright, 1994).

Results

Initial results indicated that the two positive items—I felt hopeful about the future and I was happy—are inconsistent with the principal factor of depression (λ = 2.619). Additionally, these two items showed local dependence on item “I felt depressed” (aQ3 = -0.559 & -0.383 respectively). Furthermore, item “I felt depressed” functioned differently depending on participant sex (p < 0.001). After removing these three items, the resulting 7-item scale (CES-D-7) showed improved fit statistics. Latent trait variance (σ² = 0.905, Δ = 0.631) and reliability (r = 0.712 , Δ = 0.104) significantly improved. All item Infit and Outfit statistics fell within good-fit ranges (0.840–1.139), the PCA supported unidimensionality (λ = 1.401), and no evidence of local dependence or DIF by sex remained.

Conclusion

The CES-D-7 presents a psychometrically sound and sex-invariant tool for measuring depressive symptoms. For social work researchers and practitioners, this revised scale offers a reliable method to assess depression, enhancing the accuracy of mental health screenings and interventions. The CES-D-7 can also inform policies aimed at resource allocation for mental health services by offering a more consistent measure of depression. Future research should focus on validating the CES-D-7 across additional demographic variables such as race, gender identity, and socioeconomic status to ensure broader applicability and equity in mental health assessment.