مرکز مقیاس افسردگی مطالعات اپیدمیولوژیک: اندازه گیری و تغییر ناپذیری ساختاری در سرتاسر رتبه های مردان و زنان بالغ مسن

The current study examined the measurement and structural invariance of the Center for Epidemiological Studies Depression Scale (CES-D; Radloff, 1977) across ratings provided by older adult men (n = 573) and women (n = 605). Ratings were modeled in terms of the original four-factor oblique model: a simple structure with correlated factors for Depressed Affect, Positive Affect, Somatic Symptoms, and Interpersonal Difficulties. Multiple-group confirmatory factor analysis supported full measurement and structural invariance, and no sex difference for the four latent factors. These findings indicate good support for measurement and structural invariance of CES-D ratings across older adult men and women. The psychometric and practical implications of the findings are discussed.

High levels of depression symptoms in older adults have serious negative consequences, including increased burden of physical illness, impaired functioning, and risk of suicide (Fiske, Wetherell, & Gatz, 2009). Thus we need reliable and valid measures for screening depression symptoms in older adults. The Center for Epidemiological Studies Depression Scale (CES-D) is a 20-item self-rating questionnaire, developed by Radloff (1977) that is used widely in research and clinical settings for screening depressive symptoms in adults, including older adults (Ros et al., 2011). The current study used multiple-group confirmatory factor analysis (CFA) to examine the measurement and structural invariance of ratings of the CES-D items provided by older adult men and women. Additionally, it examined the differences in latent mean scores across these groups. The original theoretical structure of the CES-D is a four-factor oblique model, with factors for Depressed Affect (DA), (lack of) Positive Affect (PA), Somatic Symptoms and Retarded Activity (SS), and Interpersonal Difficulties (ID) (see Fig. 1; Radloff, 1977). Although alternate factor models for the CES-D have been proposed, two separate meta-analyses of factor analyses studies of the CES-D have concluded most support for the original four-factor model (Kim et al., 2011 and Shafer, 2006). Currently there are CFA studies showing support for this structure in older adults (Foley et al., 2002, Hertzog et al., 1990 and Ros et al., 2011). Full-size image (28 K) Fig. 1. Schematic path diagram of the 4-factor oblique CES-D model. Note. The residuals are not shown in the figure. DA = Depressed Affect; PA = Positive Affect; SS = Somatic Symptoms; ID = Interpersonal Difficulties. The numbers in the rectangles refer to the item numbers in the CES-D. Figure options Despite having four separate factors, when scoring the CES-D, most studies have used the total score based on all items (Edwards, Cheavens, Heiy, & Cukrowicz, 2010). Some recent studies with older adults have however shown that the different CES-D scales scores are differentially associated with relevant external variables (e.g., St. John et al., 2013 and Yu et al., 2012). For instance, Yu et al. found that while age was associated positively with SS, it was associated negatively with DA and ID, and it had no relation with the total score. Also, while family support was the strongest predictor for DA, health status was the strongest predictor for SS. Findings such as these highlight the relevance, usefulness and importance of using the CES-D scales in studies involving older adults. If the CES-D scale scores are to be used with confidence in future studies with older adults, then it is critical that there is good support for the psychometric properties of the four-factor oblique model in this group. A psychometric property that could foster more confidence in the use of the CES-D factor scales in this group would be support for measurement invariance for the CES-D in terms of the four-factor oblique model across men and women. Measurement invariance refers to groups reporting the same observed scores when they have the same level of the underlying trait (Reise, Widaman, & Paugh, 1993). Invariance would mean that for the groups being compared, the measure in question has the same measurement and scaling properties. If there is weak or no support for invariance, then it follows that the groups in question cannot be justifiably compared in terms of observed scores, as the same observed scores for the groups do not reflect the same levels of the underlying trait. When applied to the CES-D, support for measurement invariance for sex would mean that men and women have the same observed scores for the same level of the latent trait. If the majority of items in the scale show measurement invariance, the observed scale scores can be compared directly across the groups, as the scores across the groups are not confounded by differences in measurement and scaling properties. For CES-D ratings from older adults, there are reasons to suspect that they could lack measurement invariance. Item response theory (IRT) based studies involving older adults have consistently shown lack of invariance for the item “I had crying spells”, with women endorsing higher ratings for the same level of depression (Cole et al., 2000, Gelin and Zumbo, 2003, Yang and Jones, 2007 and Yang et al., 2009). The Yang et al. study also showed lack of measurement invariance for “I felt like a failure”, “People were unfriendly” and “I felt that people disliked me”. For all three items, men endorsed higher ratings for the same level of depression as women. A powerful method for examining invariance is the multiple-group CFA approach. Assuming that the indicator-ratings are treated as ordered-categorical (as in the CES-D), this approach can test for configural invariance (same overall factor structure), item factor loadings invariance (same strengths of the associations of items with the first-order factors), invariance for item thresholds (equivalency for responses for the different item categories), and error variances invariance (equivalency in the error variances of the items). When there is support for invariance for item factor loadings and thresholds, structural invariance or invariance for latent variance and covariance can be tested. In addition, the equivalency for the factor mean scores can be computed after removing the items that lack invariance. Using a multiple-group CFA approach, in a study by Stommel et al. (1993) that included some older adults (percentage not reported), non-invariance was found for the factor loadings of the “crying spell” (higher for women) and “I talked less than usual” (higher for men) for a three-factor CES-D model that excluded the two ID items. O’Rourke (2005) also examined measurement invariance in a group of adult men and women that included some older adults (percentage not reported). This was examined for a higher order factor model, in which the lower order factors were the four original CES-D factors. The study found lack of invariance for the factor loadings for the items on “I felt that I could not shake off the blues even with help from my family or friends” (higher for men), “I felt lonely” (higher for men), and “I enjoyed life” (higher for women). Currently, no study has examined measurement invariance across sex for a group of exclusive older adult men and women in terms of the original four-factor model. For the original CES-D four-factor oblique model, the current study used multiple-group CFA approach to examine measurement (configural, factor loadings, intercepts and error variances) and structural (factor variances and covariances) invariance across the ratings of older adult men and women. Additionally, the study examined the differences in latent mean scores across the groups.

Prior to the multiple-group analyses, the goodness-of-fit values for the four-factor CFA model for all participants together, and for men and women separately, were examined. The fit values for all participants together were WLSMV χ2(df = 164) = 546.536, p < .001; RMSEA = .044 (90% CI = .040–.049); CFI = .978; and TLI = .974. The values for men were WLSMV χ2(df = 164) = 341.564, p < .001; RMSEA = .043 (90% CI = .037–.050); CFI = .980; and TLI = .977. The values for women were WLSMV χ2(df = 164) = 339.671, p < .001; RMSEA = .042 (90% CI = .036–.048); CFI = .978; and TLI = .975. Thus for both groups together and separately, the RMSEA, CFI and TLI values indicated good fit for the four-factor model.