A systematic review of the latent structure of the Center for Epidemiologic Studies Depression Scale (CES-D) amongst adolescents

All nine studies that provided evidence that Radloff’s four-factor structure was an appropriate fit for the data used CFA as their initial approach. Studies that proposed an alternative structure performed EFA or PCFA as their primary analyses. Six studies exclusively tested Radloff’s four-factor structure using CFA and did not consider other factor structures [11, 18, 19, 21, 22]. The studies that did compare Radloff’s structure to alternative options suggested that the four-factor structure remained the best fit, although there was adequate model fit in several other two and three- factor structures [12, 20]. These studies were primarily based in an American sample (n = 5), with one study from each of Germany, Turkey, Taiwan and China.

Four studies proposed distinct two or three-factor structures. In contrast to studies above, each sample was ethnically diverse, drawn either from schools in the Republic of Korea, Malaysia or Jordan or from an in-patient psychiatric facility in USA. How these ethnic or cultural differences may underlie major differences is explored in further detail below.

Heo et al. [17] and Ghazali et al. [16] used EFA to identify distinct three-factor structures as the optimal factor structure for the CES-D in school-based samples in the Republic of Korea and Malaysia, respectively. Heo and colleagues [17] subsequently used CFA to compare the three-factor structure to several other structures that had been suggested in the literature, including Radloff’s four-factor structure. There was no evidence that one structure was a better fit than another, suggesting that multiple factor structures may be valid for use in adolescents. Conversely, Ghazali et al. [16] did not assess the fit of other factor structures; thus, did not determine if the four-factor structure would have been an acceptable fit. Although there are some similarities in factor loadings between these three-factor structures, there remain distinct differences in items that could be attributable to cultural differences between these two Asian countries.

Faulstich et al. [15] applied a principal component analysis (PCA) approach to identify three distinct factors: a happiness dimension (identical to the positive affect factor in Radloff’s proposed structure), behavioural component of depression (similar to depressive affect) and a cognitive component of depression (combined somatic and interpersonal). However, the generalisability of this factor structure to the general adolescent population is limited for several reasons. First, the sample was drawn from two psychiatric inpatient facilities in the Southern USA and consequently is heterogeneous both in age (8 to 17 years old) and in diagnoses, which included conduct disorder, major depression, schizophrenia and atypical psychosis amongst others. This may explain why interpersonal items (i.e. item 15: people were unfriendly and item 19: “people disliked me”) loaded on to the somatic factor and why items 3 (i.e. “could not shake off the blue”) and 13 (i.e. “talked less than usual)” failed to load on to any factor. Furthermore, the authors utilised a version of the CES-D that was translated to be more comprehensible for children and adolescence. For example, “I felt that I could not shake off the blues even with help from my family or friends” in the original CES-D was translated to “I wasn’t able to feel happy, even when my family or friends tried to help me feel better.” This could have changed how individuals responded to each item. Finally, PCA assumes that the total variance is equal to the common variance, whereas EFA and CFA consider both common variance and unique variance. Differences in both the operationalisation of the questionnaire and the subsequent analyses may limit generalisability.

Dardas et al. [14] was the only study to propose a two-factor structure of positive and negative affect. The sample of Arab adolescents failed to distinguish between depressed affect symptoms, somatic symptoms and interpersonal difficulties of positive and negative affect items. Cultural differences such as decreased expression of emotions or inability to recognise both cognitive and physical symptoms of depression may underlie these differences [25].

The review summarises emerging evidence that the CES-D is not psychometrically equivalent across adolescent cultural and ethnic groups. These cross-cultural findings are consistent with those seen in reviews of CES-D in older adults [26]. Positive effect was the only domain consistently seen across all cultures. Evidence suggests that depression may be driven by psychological aspects in Western cultures (i.e. USA) and by somatic factors in Eastern cultures (i.e. China, Republic of Korea) [27]. Evidence suggests that Western cultures can better distinguish between psychological, somatic and interpersonal symptoms of depression, compared to Asian or Arab samples [25, 28‐30]. Individuals from Western cultures are thought to overemphasise the distinction between depressive symptoms of the mind and body, with a particular emphasis on the affective or psychological aspects [30].

Even within a single country, there can be considerably diversity in depressive symptoms. For example, Crockett et al. [13] examined how the structure may change amongst different American ethnic groups and found that the four-factor structure fit in Anglo and Mexican Americans, whilst a separate four-factor structure demonstrated the best fit in Cuban Americans and a five-factor structure was optimal in Puerto Rican Americans. Originally developed in an adult Caucasian sample, Radloff’s four-factor structure appears to be appropriate for an ethnically similar adolescent sample but further investigation of structural invariance across culture is necessary to support its translation and continual use in different cultures.

Several limitations must be considered in the interpretation of these results with respect to the manner in which the search was conducted and the findings in the studies captured in the review. We searched the Scopus database, which includes MEDLINE; however, Scopus does not offer the same subject heading search feature offered by MEDLINE and suggested by the Cochrane Handbook. Although best efforts were made to be comprehensive in the search strategy, articles may not have been captured by the search due to the keywords used in the search strategy and/or the absence of subject heading searches. Additionally, the total number of items captured in the search strategy was not reported, only the number of articles remaining after duplicates were removed.

First, studies that were published in a non-English language or that captured adolescents in a wider age range (e.g. aged 13–20) were not included; this may have resulted in study selection bias. Next, there was substantial heterogeneity of sample characteristics- most notably, country, age, and language of the CES-D. These differences could have induced differences in factor structure, limiting comparability between studies. While different fit indices and software appeared to have no effect on the final results, inference is limited due to differing reports of goodness of fit indices. Finally, there was a bias towards a four-factor structure in studies who utilised a hypothesis-driven approach (i.e. CFA). While this approach suggested that a four-factor structure was appropriate, a data-driven approach (i.e. EFA) may have identified more optimal factor structures.