Disparities in self-reported geriatric depressive symptoms due to sociodemographic differences: an extension of the bi-factor item response theory model for use in differential item functioning

Abstract

This study evaluates the measurement noninvariance, or differential item functioning (DIF), in the Center for Epidemiological Studies-Depression (CES-D) items attributable to age, sex, and race/ethnicity among community-dwelling older adults. Participants (N=2773) were from the New Haven site of the Established Populations for Epidemiologic Studies in the Elderly. Statistical analyses included exploratory factor analysis, bi-factor confirmatory factor analysis, and a bi-factor multiple indicator and multiple causes (MIMIC) model to address measurement noninvariance. Blacks, compared with whites, were more likely to endorse items loading on the interpersonal factor, which include "people dislike me" and "people are unfriendly." Women were less likely to endorse the interpersonal items and the "I felt like a failure" item (odds ratio [OR]=0.63, 95% confidence interval [CI]: 0.42, 0.94) than men. But women had a higher proportional odds than men for endorsing the "crying" item (OR=1.86, 95% CI: 1.17, 2.96). Those 75 years and older (relative to those aged 65-74) were less likely to endorse the "I felt like a failure" item (OR=0.65, 95% CI: 0.43, 0.97). However, measurement noninvariance found in both the "crying" and "failure" items were attributable to women and to those aged 75 and older were trivial after controlling for the underlying level of depressive symptomatology. Therefore, the interpersonal items showed measurement noninvariance attributable to sex and race. The bi-factor MIMIC model is useful for examining measurement noninvariance due to sociodemographics in a multidimensional depression instrument.

Fig. 1

This is a depiction of a generic unidimensional scale with a covariate, which is shown as the box marked with an “x.” This is called a multiple indicator, multiple causes (MIMIC) model with one general construct (g) made up of five factor indicators (y₁–y₅). There is a DIF effect found in the direct effect from the covariate (x) to the third item (y₃) in the scale. The general effect is the indirect effect between x and y₃ mediated by the general latent factor (g).

Fig. 2

The bi-factor multiple indicator, multiple causes (MIMIC) model adds one specific construct (s), measured by three items of the scale (y₁–y₃), to the generic unidimensional MIMIC model shown in Fig. 1. The general effect is the indirect effect between the covariate (x) and item y₃ mediated by the general latent factor (g). The DIF effect is the direct effect from x to y₃. The unidimensional DIF effect is decomposed in the bi-factor model into a specific effect and a (residual) DIF effect. The specific effect is the pathway from x to s to y₃.

Fig. 3

This is a depiction of the bi-factor item response theory for the 20-item Center for Epidemiological Studies-Depression (CES-D) scale. This schematic conceptual representation uses the structural equation modeling path diagram to show the latent factors (depicted by oval shapes) that include the general depression factor and the specific factors of depressed affect, interpersonal difficulties, low positive affect, and somatic complaints. The latent factor indicators are depicted by rectangles, which are the 20 CES-D items asked of each subject.