Dimensionality and differential item endorsement of depressive symptoms among aging Black populations in South Africa: Findings from the HAALSI study

Abstract

Background:

The Center for Epidemiologic Studies-Depression (CES-D) scale is a widely used measure of depressive symptoms, but its construct validity has not been adequately assessed in sub-Saharan Africa. This study validates the CES-D among an aging Shangaan-speaking and predominantly Black African sample in rural South Africa, with a special emphasis on gender differences.

Methods:

An 8-item CES-D scale was administered in Shangaan to 5,059 respondents, aged 40+ years, residing in Agincourt, South Africa. We used Cronbach’s alpha and exploratory and confirmatory factor analysis to examine and confirm dimensionality of the CES-D scale. Differential endorsement of CES-D items by gender were assessed using the Mantel-Haenszel (MH) odds ratio test.

Results:

Reliability of the CES-D scale differed by gender with women reporting higher internal consistency on items than men. A two-factor solution was retained and confirmed representing two latent factors: (1) Negative Affect (six items) and (2) Diminished Positive Affect (two items). MH results showed that men exhibited significantly higher odds of putting an effort in everything that they did (OR: 1.33, 95% CI: 1.15–1.54) and lower odds of feeling depressed (OR: 0.71, 95% CI: 0.56–0.89) and having restless sleep (OR: 0.67, 95% CI:0.58–0.77) than women.

Limitations:

Analyses were limited to a dichotomous, short form of the CES-D, a self-reported population-based measure.

Conclusion:

Aging Black Africans differ in endorsing affective and somatic items on the CES-D scale by gender, which may lead to skewed population-level estimates of depression in key subpopulations. These findings highlight the importance of continued research disentangling cross-cultural and gendered nuances of depression measurements.

Introduction

Depression is a primary contributor to the global burden of disease, impacting over 300 million people worldwide (World Health Organization, 2018). While the construct validity of depression measures has been well documented in high-income and Western countries, there is a paucity of data from low and middle-income countries (LMICs) (Bass et al., 2007; Haroz et al., 2017; Kleinman, 2004; Kohrt et al., 2014; Mutumba et al., 2014; Summerfield, 2012). Given the extensive body of literature linking low socioeconomic status to poor mental health, additional research assessing depression measurement in LMIC contexts is warranted (Miech et al., 1999; Yu & Williams, 1999). Researchers working to expand depression research to these regions must also contest with the lack of culturally appropriate psychometric assessment instruments that are adequately tailored to capture varying idioms of distress (Haroz et al., 2017; Kaiser et al., 2019). Indeed, many of our most well-known measures for capturing depression, which were developed in Europe and North America, are predominantly driven by Western and gender-specific idioms of depression. These measures are heavily skewed towards negative affective symptoms such as depressed mood, crying, and sadness (Haroz et al., 2017). For instance, the Center for Epidemiologic Depression Scale and Beck Depression Inventory, two of our most common depression measurement tools were originally created and validated in samples that were majority female or White American (Beck et al., 1988; Radloff, 1977). Moreover, depression prevalence is shown to vary considerably between countries, genders, and socioeconomic indicators, thus highlighting the variation in which this disorder is experienced across the globe (Lotfaliany et al., 2019; Mutyambizi et al., 2019).

Cultural, racial, and gender-related factors, such as normative attitudes towards emotional expression and stigmatized beliefs of mental health disorders, may obscure the construct validity of tools used to measure depressive symptoms in low-resource and indigenous African populations (Rumble et al., 1996; Swartz & Drennan, 2000). One such example of population-level depressive symptom heterogeneity among LMICs is among indigenous Black Africans residing in South Africa, where Afrocentric cultural and gender norms intersect with long-standing race-based marginalization that remains salient in the post-Apartheid era. In addition to these cultural differences, South Africa is home to vast linguistic diversity, with a total of eleven official languages (Swartz, 1998; Swartz & Drennan, 2000). Along with South Africa’s epidemiologic transition and the rising age of the population, depression has emerged as a leading cause of disability in this region (GBD 2015 DALYs and HALE Collaborators, 2016). Geldesetzer and colleagues (2019) determined that older age was the strongest correlate of depressive symptoms among elderly South Africans. Finally, despite research demonstrating that South African women have a higher prevalence of depression than men, there are limited studies that investigate whether this finding can be attributed to the higher likelihood of endorsing (e.g. responding “yes”) depressive symptoms compared to their male counterparts (Adjaye-Gbewonyo et al., 2016; Mutyambizi et al., 2019; Pengpid & Peltzer, 2018). Thus, these cultural, linguistic, and experiential factors warrant further investigation into the measurement and prevalence of depression in aging South African men and women.

The Center for Epidemiologic Studies Depression (CES-D) scale is one of the most widely used measures for assessing depression in population-based research (Anum et al., 2019). Since the original 20-item scale was created, the CES-D scale has been abbreviated to more efficiently capture depressive symptom structures and reduce respondent burden across diverse populations (Assari & Moazen-Zadeh, 2016a, 2016b; Kohout et al., 1993). Shortened CES-D scales, including 12, 10, and 8-item measures, are equivalent to the original 20-item measure in three ways; namely, the scales capture comparable dimensions of depression, demonstrate strong predictive validity to diagnostic tools in the clinical setting (Kohout et al., 1993). To reduce respondent burden in larger population-based surveys in aging populations, the psychometric properties of an 8-item, dichotomous CES-D scale has been assessed through psychometric evaluation and confirm the presence of two latent factors: depressed mood and somatic complaints (Steffick et al., 2000). Similar validation studies using abbreviated CES-D scales show acceptable, but not excellent, construct validity in Black American and male populations and demonstrate a varying factor structure compared to White American respondents (Adams et al., 2019; Assari & Moazen-Zadeh, 2016a; Powell et al., 2016; Torres, 2012). One potential reason for diminished psychometric functioning of the CES-D scale among Black populations is that certain items may conceptually overlap with other non-depression related constructs. This concept is partially confirmed by extant psychometric evidence showing that Black respondents are simultaneously more likely to endorse CES-D items related to interpersonal discord (“people disliked me”; “people were unfriendly”) and strain (“I felt that everything I did was an effort”), but less likely to have these items correlate with the remaining items of the CES-D scale compared to their racial and ethnic counterparts (Adams et al., 2019; Barnes et al., 2004; Lewis et al., 2015; Torres, 2012).

Population-based studies of depression in South Africa have shown that the sociodemographic and health correlates of depressive symptoms may differ from patterns reported in other parts of the world. In particular, cross-sectional analyses report weaker relationships between chronic health conditions and depressive symptoms in South Africa, and differential patterns with respect to gender (Geldsetzer et al., 2019). Given the rapidly aging population of South Africa, and the higher prevalence of older adults living with chronic diseases such as diabetes, hypertension, and HIV, it is important to understand how depressive symptoms manifest in this population. Additionally, older adults in South Africa lived through colonial-, Apartheid-, and post-Apartheid-era policies that aimed to systematically repress the educational and occupational opportunities afforded to Black men and women 9/2/2020 7:59:00 AM. These traumatic social policies likely shaped social relationships and mental health experiences of older South Africans, causing some depressive symptoms to be more or less salient compared to other populations. Thus, given that aging South African populations have differing comorbidities and odds of depressive symptoms compared to Western populations, it is critical to better understand the differential psychometric functioning of the CES-D scale in this region. Using a latent variable approach is one such methodology for capturing differences in depression measurement by identifying the underlying meaning or dimensions of a measure that may be distinguishable across groups (Bollen, 2002). To date, the factor structure and item endorsement of the 8-item CES-D scale remain unexplored in aging South African populations.

The present study

To address the aforementioned gaps, our study used a latent variable approach to address two objectives: (1) assess the appropriateness and usefulness of the Shangaan translation of the 8-item CES-D in an aging cohort of Black adults, and (2) describe differences in item endorsement by gender. Methodologically, we sought to assess the dimensionality and differential item endorsement of the CES-D using factor analyses and odds ratio tests. Guided by the previous U.S.-based validation study conducted by Steffick and colleagues (2000), we hypothesized that depressive symptoms, as measured by the Shangaan-translated 8-item CES-D measure, would cluster along two correlated latent factors. Additionally, we expected that our exploratory assessment of psychometric properties and differential item endorsement of the CES-D by gender would yield significant differences by gender.

Methods

Description of the HAALSI study

We used baseline data from the Health and Aging in Africa: A Longitudinal Study of an INDEPTH Community in South Africa (HAALSI) cohort. HAALSI is the first Health and Retirement Study (HRS) sister study in Africa, and is conducted in the Mpumalanga region of South Africa (Gómez-Olivé et al., 2018). HAALSI was created to establish a population-based longitudinal cohort of men and women aged 40 and over in a rural South African community (Gómez-Olivé et al., 2018). The cohort included 5,059 people (n = 2,345 men and 2,714 women). The study was conducted in the Agincourt sub-district in Mpumalanga province, South Africa, where the Medical Research Council (MRC)/Wits Rural Public Health and Health Transitions Research Unit has been running the Agincourt Health and Demographic Surveillance System (HDSS) since 1992. The Mpumalanga region is the second most homogenous province in South Africa, with over 90% of residents identifying as Black African. Within this province, the lifetime prevalence of a DSM-IV mood disorder is 9%, compared to 13.7% and 6.3% in the more racially diverse provinces of Western Cape and Limpopo, respectively (Herman et al., 2009). The HDSS conducts an annual census of all households and collects vital events for all household members (births, deaths and migration), and residency status. Sociodemographic characteristics are collected in alternating years (Kahn et al., 2012). The study area consists of 31 villages and covers approximately 450 km²; the total population is approximately 116,000 people. Overall, the demographic profile of the HAALSI cohort was typical of rural South Africa; life expectancy at older ages has improved in Agincourt as well as elsewhere in rural South Africa (Bor et al., 2013; Kabudula et al., 2017).

Only individuals who were aged 40 years or older as of July 1, 2014 and residing in the area during the 12 months prior to enrollment were eligible to participate in the study. We identified 12,484 (70% female) eligible individuals. Of those, 5,890 individuals (50% female) were selected through simple random sampling (stratified by sex). Participants were visited at home to seek informed consent and to be interviewed. Five thousand and fifty-nine (86%) individuals consented to the survey and were included in the final sample. Data for the current study were taken from the baseline assessment, which was administered in Shangaan to all participants using computer-assisted personal interviews between November 2014 and November 2015.

Measures

Depressive symptoms were assessed using an abbreviated 8-item version of the Center for Epidemiologic Studies-Depression (CES-D) screening tool (Radloff, 1977). The scale was translated into Shangaan by staff from the Agincourt MRC/Wits Rural Public Health and Health Transitions Research Unit’s Public Engagement Office. Binary responses were coded as Yes=1 and No=0.

Analysis

The analysis had several phases. First, we assessed the psychometric properties of the measure using means, frequencies and internal consistency, as measured by the Cronbach’s alpha, by item for the total sample and by gender. A Cronbach’s alpha value of 0.70 or higher is generally considered to be acceptable (Cortina, 1993). Item-to-total correlations were examined to evaluate how well each item mapped onto the overall scale and values of 0.30 or higher were evaluated as acceptable (Nunnally, 1994). A Kaiser-Meyer-Olkin (KMO) Test for sampling adequacy and Bartlett’s test for sphericity were performed in Stata v. 15 (StataCorp, 2017). A KMO index value of 0.50 or higher and a significant (p<0.05) Bartlett’s test indicate that the data was suitable for factor analyses (Williams et al., 2010).

Second, we conducted exploratory (EFA) and confirmatory factor analyses (CFA) in MPlus version 8.0 (Muthén & Muthén, 2019). Due to the mixed performance of the CES-D in the sub-Saharan context (Baron et al., 2017; Kilburn et al., 2018), an EFA was first conducted to establish a recommended factor structure that would be tested in the confirmatory phase of the analysis. We reverse-scored two positively worded symptoms: “I felt happy” and “I put an effort in everything that I did”, to reflect higher depressive symptomatology for the EFA/CFA phase of analysis. We tested a one, two and three-factor model, with direct oblimin rotation, to determine the appropriate number of correlated factors to retain; model fit was evaluated using criteria described by DeVellis (2016), which includes evaluation of eigenvalues greater than one. Additionally, we evaluated communalities (h²) with values at 0.4 or higher indicating acceptable variance with other items in the scale or factor (Costello & Osborne, 2005). Higher factor loadings for each CES-D item were used to signal the primary factor where the item would be loaded.

We confirmed the fit of the retained EFA model using CFA. Model fit was assessed using a weighted least square estimator (WLSMV) to account for the binary nature of the 8-item CES-D scale. Acceptable goodness of fit indices were based on cut-off values previously specified (Hu & Bentler, 1999), including the chi-square, comparative fit index (CFI ≥ 0.95), Tucker-Lewis index (TLI ≥ 0.95) root-mean square error of approximation (RMSEA ≤ 0.06)..

We used the Mantel-Haenszel (MH) procedure, a statistically powerful method for detecting gender differences in the CES-D 8 scale. The MH procedure indicates a difference, illustrated by an odds ratio, in item performance between two comparable groups of examinees, that is, groups that are matched with respect to the construct being measured by the test (Holland & Thayer, 1986; Jin et al., 2018). Men and women were treated as the focal and reference groups, respectively, in the differential item functions (DIF) analysis. Thus, a MH odds ratio > 1 indicated higher endorsement of an item for men and < 1 indicated higher endorsement of the CES-D item for women.

Results

Sample characteristics included 5,059 total participants, 2,345 men and 2,714 women. Analysis was conducted on 4,951 individuals (97.9% of the total sample) with complete data on all CES-D items. Table 1 presents descriptive item responses of the CES-D 8-item scale, by gender. On average, respondents generally reported fewer depressive symptoms (e.g. reporting “No” on CES-D items or “Yes” on reverse coded items).

Table 1:

CES-D Item Responses for the Total Sample and by Gender

CES-D item	n (% yes)
	Total sample (n=5,059)	Men (n=2,345)	Women (n=2,714)	p value
1. I felt depressed	564 (11.4)	214 (9.4)	350 (13.2)	<0.001
2. Everything an effort	3,138 (63.4)	1,407 (61.4)	1,731 (65.1)	<0.001
3. Sleep was restless	1,626 (32.9)	662 (28.9)	964 (36.3)	<0.001
4. I was happy	3,822 (77.2)	1,762 (76.9)	2,060 (77.5)	0.62
5. I felt lonely	669 (13.5)	294 (12.8)	375 (14.1)	0.19
6. I did not enjoy life	464 (9.4)	202 (8.8)	262 (9.9)	0.21
7. I felt sad	474 (9.6)	206 (9.0)	268 (10.1)	0.20
8. I could not get going	433 (8.8)	186 (8.1)	247 (9.3)	0.15

Psychometric properties of CES-D scale by total sample and gender

Table 2 presents the mean, reliability, and item-to-total correlations by total sample and gender. Three items had an item-to-total correlation less than 0.50: “I put an effort in everything I did” (r= 0.47 & 0.48, among men and women respectively), “I was lonely” (r=0.49 in male sample), and “I was happy” (r=0.49 in male sample). The lowest item-to-total correlation was for the “effort” item among Black men (r=0.47). Cronbach’s alpha differed between genders such that women reported marginally improved internal consistency than men (0.68 vs.0.60, respectively). The reliability of the CES-D scale in the overall sample was 0.66. On average, men had lower average CES-D scores than women, with significant differences in two individual items (“depressed” and “restless sleep”) and total CES-D score between genders.

Table 2:

Psychometric properties of eight item CES-D scale, by gender

CES-D item	Total Sample (n=5,059)		Men (n=2,345)		Women (n=2,714)
	M(SD)	Item to total correlation	M(SD)	Item to total correlation	M(SD)	Item to total correlation
1. I felt depressed*	0.11 (0.32)	0.59	.09 (0.29)	0.50	0.13 (0.34)	0.62
2. Everything an effort*	0.37 (0.48)	0.48	0.39 (0.49)	0.47	0.35(0.48)	0.48
3. Sleep was restless*	0.33 (0.47)	0.53	0.29 (0.45)	0.53	0.36 (0.48)	0.52
4. I was happy	0.77 (0.42)	0.50	0.77 (0.42)	0.49	0.78 (0.42)	0.52
5. I felt lonely	0.14 (0.34)	0.52	0.13 (0.33)	0.49	0.14 (0.35)	0.53
6. I did not enjoy life	0.09 (0.29)	0.64	0.09 (0.28)	0.59	0.10 (0.30)	0.66
7. I felt sad	0.10 (0.29)	0.65	0.09 (0.29)	0.62	0.10 (0.30)	0.68
8. I could not get going	0.09 (0.08)	0.63	0.08 (0.27)	0.61	0.09 (0.29)	0.65
Total CES-D scale*	1.45 (1.61)		1.38 (1.52)		1.49 (1.68)
Cronbach’s alpha (standardized)		0.66		0.60		0.68

^*p<0.05 between genders

Exploratory factor analysis

The KMO index of the overall scale was 0.80 and the Bartlett test yielded a significant result (X²=6872.11, df=28, p<0.001), which signaled suitability for the subsequent factor analysis. One, two, and three-factor EFA models were conducted to determine the optimal number of factors to retain for confirmatory analysis. The one-factor EFA model yielded the poorest overall model fit (CFI=0.900, TLI=0.860, RMSEA=0.097, 90% CI: 0.02–0.103), while the two-(CFI=0.991, TLI=0.981, RMSEA=0.035, 90% CI: 0.029–0.042) and three-factor model (CFI=0.996, TLI=0.984, RMSEA=0.033, 90% CI: 0.024–0.043) demonstrated excellent fit. The two-factor model was selected for subsequent analyses due to higher factor loadings and fewer cross-loaded items compared to the three-factor model. Moreover, the eigenvalue for the three-factor model (0.624) was lower than optimal EFA fit criteria retention of one, which is detailed in a previous validation study (Van de Velde, Bracke, Levecque, & Meuleman, 2010). Communalities ranged between 0.378 (“I put an effort in everything that I did”) to 0.799 (“I felt sad”). Additional factor loadings for varimax and promax rotations are available in the Supplementary Materials.

The final two-factor solution accounted for 50.6% of the variance in the overall scale. Factor 1, represented as Negative Affect, consisted of items related to feeling “depressed”, “restless sleep”, “lonely”, “not enjoying life”, “sad”, and “having trouble getting going”. The Negative Affect dimension explained 34.8% of the variance in the overall sample, and encompass negative affect and somatic items related to depression. Factor 2, represented as Diminished Positive Affect, was comprised of reverse coded items related to feeling “happy” and “putting forth an effort”. The Diminished Positive Affect dimension explained 15.8% of the variance in the overall sample. Direct oblimin rotated factor loadings, communalities, and correlations for the final two-factor EFA model are illustrated in Table 3.

Table 3:

Direct oblimin rotated standardized factor loadings of the exploratory factor analysis

CES-D Item	Factor 1: Negative Affect	Factor 2: Diminished Positive Affect	h2
1. I felt depressed	0.69*	0.08*	0.54
2. Everything an effort	0.01	0.61*	0.38
3. Sleep was restless	0.77*	−0.38*	0.46
4. I was happy	0.01	0.83*	0.69
5. I felt lonely	0.76*	−0.25*	0.46
6. I did not enjoy life	0.87*	0.02	0.78
7. I felt sad	0.90*	−0.01	0.80
8. I could not get going	0.85*	0.06	0.78
Correlation between factors	0.48*

^*p<0.001

Bolded values signal the designated factor selected for each CES-D item

h²= communality value

Confirmatory factor analysis

Given the results of the EFA, we ran a two-factor confirmatory analysis to assess model fit. Figure 1 summarizes factor loadings and fit indices of the two-factor model. Figure 1 illustrates the final measurement model and confirmatory factor loadings of the 8-item CES-D scale. Standardized factor loadings were all statistically significant at p<0.001 and ranged between 0.562 (“My sleep was restless”) and 0.898 (“I felt sad”). Standardized factor loadings in the Negative Affect dimension ranged from 0.562–0.898 and Diminished Positive Affect was comprised of factor loadings ranging from 0.606–0.856. The results of the intercorrelated model CFA also showed a significant positive correlation between the two factors (r = 0.414, p<0.001). This model demonstrated reasonable model fit as demonstrated by model fit indices (WLSMV χ²=360.88, TLI=0.946, CFI=0.963, RMSEA=0.060, 90% CI: 0.055–0.066).

Figure 1: Confirmatory Factor Loadings and Model Fit.

Model fit indices:

RMSEA: 0.060 (90% CI 0.055–0.066, p<0.001); CFI: 0.963; TLI: 0.946

*Reverse coded items

Gender differences in item endorsement

Mantel-Haenszel DIF analysis revealed several significant differences in item endorsement by gender (Table 4). Items 1–3 (e.g. feeling depressed, effort, and restless sleep) exhibited DIF by gender based on a 5% significance level. The results show that men exhibited significantly higher odds of “putting an effort in everything that they did” (OR: 1.33, 95% CI: 1.15–1.54). Moreover, males in our study sample reported lower odds of “feeling depressed” (OR: 0.71, 95% CI: 0.56–0.89) and having “restless sleep” (OR: 0.67, 95% CI:0.58–0.77) than their female counterparts. The remaining CES-D items exhibited no significant differences between gender.

Table 4:

Mantel Haenszel odds ratios of differential item functioning of CES-D 8 items by gender^*

CES-D item	Odds Ratio (95% CI)
1. I felt depressed	0.71 (0.56–0.89)
2. Everything an effort	1.33 (1.15–1.54)
3. Sleep was restless	0.67 (0.58–0.77)
4. I was happy	1.12 (0.96–1.32)
5. I felt lonely	1.01 (0.84–1.39)
6. I did not enjoy life	1.21 (0.92–1.58)
7. I felt sad	1.20 (0.91–1.58)
8. I could not get going	1.18 (0.89–1.56)

^*Men and women were treated as the focal and reference groups, respectively, in the Mantel Haenszel (MH) analysis. Thus, a MH odds ratio > 1 indicates higher endorsement of an item for men and an MH of <1 indicates higher endorsement of an item for women.

Bolded values are significant at p<0.05.

Discussion

The current study seeks to provide additional insight towards cultural and gender-influenced differences in the CES-D, a commonly used population-based depression measure. To our knowledge, this is the first psychometric evaluation of the abbreviated CES-D 8 scale in a sub-Saharan sample of aging Shangaan-speaking adults. Furthermore, this study uniquely identified differences in item endorsement by gender in an exclusively Black African study sample. Our approach and findings yield additional evidence for researchers to weigh when interpreting depression measures in aging populations in sub-Saharan Africa, such as the psychometric functioning of abbreviated measures in predominantly Black African populations and item endorsement between gender.

Our guiding hypothesis that the Shangaan-translated 8-item CES-D measure would cluster into positive affect and negative affect, was confirmed, as results demonstrated optimal fit for the two-factor model compared to 1- and 3-factor models. The two factors emerged as (1) a combination of negative affect (e.g. crying, sadness, lonely) and somatic symptoms associated with depression (e.g. restless sleep) and (2) positive affect (e.g. happiness). The Negative Affect factor mirrored that of previous studies among Black Americans and Zulu and Xhosa-speaking South Africans; notably, the finding that affective-somatic symptoms merge into a single factor (Adams et al., 2019; Baron et al., 2017; Kilburn et al., 2018). Comparatively, studies including White American populations, Dutch, and Afrikaans-speaking South Africans yielded a similar integration of negative and somatic symptoms, suggesting that the psychological and physical experience of depression may be intrinsically linked across both culture and language (Assari & Moazen-Zadeh, 2016a; Baron et al., 2017; Kwakkenbos et al., 2013).

The Diminished Positive Affect factor included two reverse-coded items: “I felt happy” and “I put an effort in everything I did”. Although methodologists recommend at least three items per factor (Kline, 1986, 2015; Raubenheimer, 2004), our two-item factor was supported through our CFA findings and maintain significantly higher bivariate correlation values compared to other CES-D items (see Supplementary Materials). Particularly, the “effort” item was retained in the final factor structure despite poor fit, as demonstrated by a low communality value, EFA and CFA factor loadings and item-to-total correlation, for two reasons. First, item-to-total correlations of the “effort” item fit the methodologic standard for acceptable values (0.20 and above) for both men and women in our study sample (Kline, 1986). Second, the “effort” item was the only item that changed meaning as a result of the Shangaan translation, moving from a negative framing in the original Radloff (1977) study (e.g. “I felt that everything I did was an effort”) to positive framing (e.g. “I put an effort in everything that I did”). As a result, it is no surprise the “effort” item loaded onto the same factor as the “happy” item to collectively represent lack of positive affect. Consequently, we were interested in exploring the implications of this linguistic change on the construct validity of the item and positioning the findings in extant literature which highlights a consistently poor performance of the “effort” item in Black populations (Adams et al., 2019; Kilburn et al., 2018; Torres, 2012). Building on these previous findings, our results suggest that a single item measure of effort, either positively or negatively framed, may not fully capture the depression experience in Black populations. Moreover, because these results were found in previous studies involving both Black Americans and indigenous Black sub-Saharan Africans, the “effort” item may indicate a broader misfit in capturing the underlying experience of depression across the African Diaspora. Public health researchers and mental health providers working with these populations should consider the psychometric limitations of the CES-D by race, and interpret the “effort” item with caution.

The Mantel-Haenszel analysis confirmed our second hypothesis by illustrating that item endorsement differed by gender such that men were more likely to endorse the effort item (“I put an effort in everything that I did”) and less likely to endorse sleep problems and depressed mood than women. This finding stands in parallel to our item-to-total correlation findings and descriptive statistics, which demonstrated that women in our study sample were more likely to endorse depressive symptoms than men. To explain these differences, psychological literature has identified gender and racial socialization potential rationale for gendered idioms of depression and, particularly, Black men’s lower endorsement of depressive symptoms. A small, but robust, body of literature in the United States posits that traditional masculine norms, which censures crying, sadness, and vulnerability, may explain a lower endorsement of depressive symptoms among African American men (Hammond et al., 2010; Hammond & Mattis, 2005; Lindsey et al., 2017). Conversely, Black women may actually experience higher rates of depression in this and other parts of rural South Africa as a result of historic oppression of women and girls in this context, a product of Colonial- and Apartheid-era policy coupled with patriarchal family structures and societal norms (Oberhauser & Pratt, 2004). Little is known about the intersection of gender norms and depressive symptoms in the South African context. Future research should explore predictors and norms that influence differential item endorsement by gender in sub-Saharan African countries.

A major strength of the HAASI dataset is its harmonization with the U.S.-based Health and Retirement Study (HRS). The parallel use of measures in multiple countries lends itself to future comparative studies of aging Black populations in the United States and South Africa. Indeed, our findings related to factor analyses both align and diverge from HRS studies in the United States, which showed acceptable fit for two- (psychosomatic and lack of positive affect) and three- (dysphoria, psychosomatic, and lack of positive affect) factor models (Steffick et al., 2000; Yang & Jones, 2008). These previous studies also found higher reliability scores for CES-D items among female participants than men. Despite these parallel findings, U.S.-based studies also report higher Cronbach’s alphas than our results illustrate, suggesting that the CES-D may have diminished reliability to adequately capture the full range of depressive symptoms and psychological wellbeing in South Africa (Van de Velde et al., 2010).

There are limitations that may influence the interpretation of our results. First, the use of the 8-item CES-D scale excludes depressive symptoms related to broader somatic changes (e.g., weight gain, appetite change) that is reflected in the original 20-item scale (Radloff, 1977). Future studies are needed to extend findings to the full CES-D measure to assess whether higher endorsement of somatic symptoms is evident in study samples comprised of Black South Africans. Despite these limitations, this study provides an important contribution towards further understanding heterogeneity in the experience of depressive symptoms in aging, rural, and LMIC settings. Although there is large body of evidence that examines depression among counterparts in the U.S. and other Western countries, this study builds on the dimensional structure of a widely used scale by introducing key psychometric and dimensional differences in CES-D item endorsement that are unique to aging sub-Saharan African men and women.

Despite these limitations, the current study provides important insights as to how older Black African adults residing in South Africa, who have disproportionately endured systemic oppression and racial marginalization as a function of Apartheid-era policies, may differentially experience or make sense of depression. Our findings suggest that the use of an abbreviated CES-D scale with this population is nuanced and requires additional contextual and gender-focused considerations in future studies. While our results found that the 8-item measure demonstrated acceptable psychometric properties and confirmatory model fit, forthcoming research should also disentangle gendered differences in item endorsement, diminished construct validity of the “effort” item, and the contrasting dimensionality of the scale compared their Western counterparts in the harmonized Health and Retirement study (Steffick et al., 2000). We also recommend future qualitative studies to deeply explore the experience of depression in the community which could precede robust validity and reliability studies. Researchers should also center these studies in the context of South Africa’s growing aging population and, most importantly, the Apartheid-related racial trauma that this population has endured. Such insights would provide researchers with clarified evidence on idioms of depression in LMIC and predominantly Black African contexts, which, in turn, will enhance sustainable strategies to promote mental wellbeing in these populations.

Highlight.

• The Center for Epidemiologic Depression Scale (CES-D) may have diminished reliability and validity to adequately capture the full range of depressive symptoms among aging adults in South Africa

• CES-D scale items clustered into two latent factors: Negative Affect and Diminished Positive Affect.

• A single item measure of effort, as captured in the CES-D scale, may not fully describe the depression experience in Black and African populations

• Aging Black Africans differ in endorsing affective and somatic items on the CES-D scale by gender.