Factor Analysis of the Center for Epidemiological Studies Depression Scale in American Indian Women

Abstract

The Center for Epidemiological Studies Depression Scale (CES-D) is widely used to assess depressive symptoms in the general population. It lacks validation for widespread use within the American Indian population, however. To address this gap, we explored and confirmed the factor structure of the CES-D among a community sample of southeastern American Indian women. We analyzed data from a sample of 150 American Indian women ages 18 to 50 from a southeastern tribe who had complete responses on the CES-D as part of a larger cross-sectional, community-engaged study. We performed exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) to assess the measure’s validity. We examined EFA models ranging from one to five factors, with the four-factor structure yielding the best overall model fit (CFI = 1.00, TLI = 0.99, RMSEA = 0.03). Differences between the four-factor EFA-retained structure from our sample and Radloff’s four-factor structure emerged. Only the interpersonal factor was common to both factor structures. Our study findings confirm the validity of the original four-factor structure of the CES-D for younger adult American Indian women in the Southeast. Contrasting findings with the EFA-retained structure, however, provide a more nuanced interpretation of our results.

Depression is a growing public health concern in the United States. Trends in depression prevalence show a significant increase in the US population from 2005 to 2015 (Weinberger et al., 2018). Despite the general consensus that American Indians suffer a disproportionate burden of mental health problems compared to other racial and ethnic groups, few studies have examined the prevalence of depression in this population (Amparo et al., 2011; Asdigian et al., 2018; Gone & Trimble, 2012). American Indians are an extremely diverse population, representing 574 federal and 63 state recognized tribal nations in the United States (National Congress of American Indians, 2020). In one recent national epidemiologic study conducted by Hasin and colleagues (2018), American Indian and Alaska Native adults had the highest 12-month (15.9%) and lifetime (28.2%) prevalence of major depressive disorder—as defined by the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5)—of any racial and ethnic group. In two small community-based samples of American Indians from southeastern U.S. tribes, the prevalence of clinically significant depressive symptomatology ranged from 13.2% to 25% (McKinley et al., 2021; Schure & Goins, 2017). Some studies have revealed gender-related disparities in depressive symptoms, with women experiencing greater severity than men. For example, among American Indians in the Strong Heart Family Study, women had elevated rates of severe depressive symptoms (15.5%) compared to men (7.2%) (Zhao et al., 2016). Because few available published study results rely on population-based measures, estimates of depression prevalence in the American Indian population remain widely unexplored and—due to the immense diversity among tribal nations—have limited generalizability.

One of the most widely used, generalizable, population-based measures of depressive symptoms is the Center for Epidemiological Studies Depression Scale (CES-D). Although the CES-D is a well-established measure of the frequency of depressive symptoms in the general population (Radloff, 1977), its validation for use with American Indians remains inadequate (Gray et al., 2019; Schure & Goins, 2017). Only a few studies have assessed the validity of the CES-D in American Indian adults, and most focus on older adults or elders (Barbosa-Leiker et al., 2021; Gray et al., 2019, Schure & Goins, 2017; Somervell et al., 1993). In one recent study, Gray and colleagues (2019) reported construct and convergent validity of the CES-D in a sample of 473 Northern Plains Indians aged ≥ 18 years. The CES-D showed strong to moderate correlations with the Beck Depression Inventory-II (BDI-II), Tri-Ethnic Depression Scale (TEDS), Beck Hopelessness Scale (BHS), and SCL-90-R depression measures. In another study by Schure & Goins (2017), the CES-D demonstrated concurrent and divergent validity in a sample of 491 older (≥55 years) community-dwelling American Indians residing in the southeast. The CES-D was directly correlated with chronic pain and physical disability and indirectly correlated with social support and self-efficacy in this population.

Findings from a meta-analysis support the stability of the four-factor structure (i.e., depressed affect, positive affect, somatic symptoms, and interpersonal conflict) (Radloff, 1977) of the CES-D and congruence of structures across diverse racial and ethnic groups. These findings are inconsistent in minority populations, however, including American Indians (Kim et al., 2011; Schure & Goins, 2017). In examining the CES-D factor structure, Schure and Goins (2017) determined a two-factor solution—with 16 of the 20 items loading onto factor 1 (depressed affect) and three items loading onto factor 2 (positive affect)—to be the best fit model in their sample. In contrast, Barbosa-Leiker et al. (2021) found that the 20-item CES-D demonstrated poor model fit in their study of American Indian elders from the Northern Plains, Southern Plains, and Southwest. Instead, a three-factor model (i.e., depressed affect, somatic symptoms, and wellbeing) of the 12-item CES-D demonstrated the most valid factor structure. Wide variations in the factor structure across these samples warrant further investigations.

Insufficient exploration of the CES-D’s validity in American Indians is problematic given the cultural diversity among tribal groups. Consequently, depression measures may have limited construct validity, meaning the CES-D may not be as appropriate for this population since it was developed and normed with predominately White samples. Establishing the validity of the CES-D is further complicated by unique sociohistorical stressors among various American Indian tribes and age groups, which may cause varied expressions of depression (Brave Heart, 2003; Burnette et al., 2019; Tucker et al., 2016). Furthermore, the gendered experience of depression has not been well characterized in the American Indian population. Of the recent studies measuring depression, only one stratified their results by gender and found that women American Indian elders had higher levels on the depressed affect and somatic symptoms subscales than men American Indian elders (Barbosa-Leiker et al., 2021; Gray et al., 2019; Schure & Goins, 2017). Therefore, American Indian women may have psychological processes that manifest in ways researchers are not currently capturing. Reported variation in the conceptualization and expression of depression within other racial and ethnic populations underscores the scope of the problem in accurately measuring depressive symptoms in American Indians (Kim et al., 2011; Torres, 2012). Insufficient evidence on the CES-D’s validity and the gendered depression experience in the American Indian population necessitates examining the factors of depression in American Indian women, a population at high risk of mental health conditions. Although some studies have examined the construct validity of the CES-D in American Indians—reporting two to three factors as the most valid factor structure—these studies focused mainly on older adults and heterogeneous samples, leaving younger (<50 years) American Indian women understudied. Thus, in this study, the researchers aimed to explore and confirm the CES-D’s factor structure with a community sample of younger adult American Indian women in the southeast.

Methods

Data Source

We used data collected from the Hazardous Air Pollutants, Positivity, and Inflammation (HAPPI) study, a cross-sectional, community-engaged study conducted between 2018 and 2019 (Brooks et al., 2019). The goal of the HAPPI study was to explore whether positive psychological states buffer the cardiovascular-associated inflammatory effects of ambient air pollution in American Indian women residing in the southeastern United States. Briefly, we collected demographic, psychological, biological, and physical health data. For this secondary analysis, we included baseline data from a sample of 150 American Indian women ages 18–50 from a southeastern tribe who had complete responses on the CES-D. Minimum sample size recommendations for conducting factor analyses vary widely and generally lie within two categories—absolute numbers and ratios. The rule of thumb for minimum Ns in absolute numbers range from 100 to 500 (Kyriazos, 2018). Others rely on the ratio of participants (N) to variables (p), also known as the N:p ratio. The suggested N:p ratio ranges from 5 (with a minimum N > 100) to 10 participants per variable. Based on these criteria, our study met the recommended minimum sample size considered adequate to perform factor analyses.

Procedures

In the HAPPI study, we used convenience sampling methods to recruit a community sample of American Indian women ages 18 to 50 residing in southeastern North Carolina (Brooks et al., 2019). We recruited through various means, including flyer advertisements, direct contact, word of mouth, and community events, such as a health fair sponsored by a local academic institution. Women who expressed interest in the survey contacted the study personnel for informed consent and the completion of anonymous self-administered surveys and in-person study visits. Additional information regarding the data collection procedure is detailed elsewhere (Brooks et al., 2019). Participants completing the study received a $50 gift card. All study procedures were reviewed and approved by the tribal health board, Tribal Council, and the Institutional Review Boards at the University of North Carolina at Chapel Hill and local community healthcare organization.

Measures

Depressive Symptoms (CES-D, 20-item)

We administered the 20-item version of the CES-D, a self-report screening tool used to measure depressive symptoms in the general population, to eligible women (Radloff, 1977). Item responses range from (0) rarely or none of the time to (3) most or all of the time. Item scores are summed after reverse coding the four positively worded items (e.g., “I was happy”). Possible scores range from 0 to 60, with higher scores signifying a higher frequency of depressive symptoms. Risk for clinical depression, indicated by a typical cutoff score of ≥16, has high sensitivity (.87) and moderate specificity (.70); however, cut points vary by population (Vilagut et al., 2016).

Demographics

Demographic characteristics in these analyses included age, level of education (e.g., some high school, high school, some college, etc.), employment status, and household income.

Analysis

We used STATA v. 15 to conduct a descriptive analysis of the data (i.e., means, standard deviations, and item-to-total correlations). We conducted two tests to determine feasibility of the factor analysis approach, a Kaiser-Meyer-Olkin (KMO) test and a Bartletts’ test for sphericity (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).

Due to inconsistent factor structure findings and limited investigations among American Indian women in the existing literature, we used Mplus version 8.0 (Muthén & Muthén, 2019) to perform exploratory factor analysis (EFA) and confirmatory factor analysis (CFA). We tested four factor models—ranging from one factor to five factors—using 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 the evaluation of eigenvalues greater than one. Additionally, we evaluated communalities (h2), with values of 0.4 or higher indicating acceptable variance with other items in the scale or factor (Costello and Osborne, 2005). We used the highest factor loadings for each CES-D item to signal the primary factor where the item would be retained for the confirmatory analyses. Finally, we tested the original four-factor CES-D structure as determined by Radloff (1977).

Model fit was assessed using a weighted least square mean and variance adjusted estimator (WLSMV) to account for the categorical response items of the 20-item CES-D. Acceptable goodness of fit indices was based on cutoff values previously specified (Hu and Bentler, 1999), including the chi-square, comparative fit index (CFI ≥ 0.95), Tucker-Lewis index (TLI ≥ 0.95), and Root Mean Squared Error of Approximation (RMSEA ≤ 0.06).

Results

Descriptive Statistics

This study’s sample included demographic and CES-D data from 150 American Indian women ages 18 to 50, with a mean age of 36.5 years (SD = 10.2). Table 1 includes demographic information for this sample. Most participants were currently employed (55.7%). Approximately 33% of women had earned a college or graduate degree. About 40% of participants reported an annual household income of > $35K.

Table 1.

Demographics (n = 150)

Demographic Measure	Mean	SD	Frequency	Percent
Age (in years)	36.5	10.2
Employment status
Currently employed			83	55.7
Unemployed			66	44.3
Highest level of education completed
Less than high school			1	0.67
High school or equivalent			39	26.2
Some college, but no degree			60	40.3
College (Associate/Bachelor’s degree)			30	20.1
Graduate (Master’s/Professional/Doctoral degree)			19	12.8
Annual income
<$35k			82	55.0
$35.1 – 50k			18	12.1
$50.1 – 75k			22	14.8
>75.k			20	13.4

Sample Means, Standard Deviations, and Item-to-Total Correlations of the CES-D

Table 2 presents the means, standard deviations, and item-to-total correlations (ITCs) for each CES-D item in our study sample. Mean CES-D item responses ranged from 0.33 (SD = 0.62 – “I felt like people dislike me”) to 1.52 (SD = 1.13 – “I felt everything I did was an effort”). Reliability analysis of the CES-D resulted in a Cronbach’s alpha of 0.91, indicating adequate internal consistency reliability. As listed in Table 2, the items generally showed very good discrimination, with ITCs exceeding acceptable standards (r > 0.30; Hajjar, 2018) on all except two items (“I felt that I was just as good as other people” and “I felt everything I did was an effort”).

Table 2.

CES-D Sample Means, Standard Deviations, and Item-to-Total Correlations

Scale Item		Mean	Standard Deviation (SD)	Item-to-Total Correlations
Depressed
3	Blues	0.57	0.83	0.65
6	Depressed	0.66	0.92	0.76
9	Failure	0.35	0.67	0.73
10	Fearful	0.38	0.63	0.57
14	Lonely	0.51	0.84	0.69
17	Cry	0.42	0.72	0.69
18	Sad	0.62	0.78	0.76
Somatic
1	Bothered	0.60	0.78	0.53
2	Appetite	0.59	0.82	0.65
5	Mind	0.79	0.92	0.66
7	Effort	1.52	1.13	0.02
11	Sleep	1.06	1.04	0.46
13	Talk	0.54	0.77	0.51
20	Get Going	0.62	0.83	0.74
Positive
4	Good	1.10	1.10	0.20
8	Hopeful	0.93	0.97	0.45
12	Happy	0.73	0.85	0.61
16	Enjoy	0.62	0.82	0.60
Interpersonal
15	Unfriendly	0.47	0.74	0.59
19	Dislike	0.33	0.62	0.61

Note. CES-D = Center for Epidemiologic Studies Depression Scale.

EFA Results

The KMO index of the overall scale was 0.91, and the Bartlett test yielded a significant result (X²=1506.17, df=190, p<0.001), which signaled adequacy for the subsequent factor analyses. We tested five exploratory factor models to determine the optimal fit for our confirmatory testing. The one-factor and two-factor EFA models yielded the poorest overall model fit (1-factor: CFI = 0.94, TLI = 0.94, RMSEA = 0.08; 2-factor CFI = 0.98, TLI = 0.97, RMSEA = 0.05). The three- and four-factor model yielded model fit closer to our predetermined standards described by Hu and Bentler (1999), (3-factor: CFI = 0.99, TLI = 0.98, RMSEA = 0.05; 4-factor CFI = 1.00, TLI = 0.99, RMSEA = 0.03). Finally, our five-factor exploratory solution also yielded good model fit (CFI = 1.00, TLI = 1.00, RMSEA = 0.02) but was not selected for further confirmatory analyses due to its eigenvalue (0.85), which falls below the optimal EFA fit criteria retention of the value one (DeVellis, 2016). Thus, we selected the four-factor model for subsequent confirmatory analyses in accordance with model fit and alignment with eigenvalue criteria.

The four-factor solution accounted for 61.1% of the variance in the overall CES-D. The eigenvalues were 10.40, 1.73, 1.10, and 0.98 for the first four factors. Direct oblimin rotated factor loadings for the four-factor EFA model are illustrated in Table 3. For the four-factor model, factor 1 comprised four items (bothered, mind, failure, sleep), factor 2 contained eight items (appetite, blues, depressed, effort, lonely, cry, sad, and get going), factor 3 included six items (good, hopeful, fearful, happy, talk, enjoy), and factor 4 had two items (unfriendly, dislike).

Table 3.

Exploratory Factor Analysis, 4-Factor Structure of the Full 20-Item CES-D (n = 149)

		Factors
Scale Items		1	2	3	4
1	I was bothered by things that usually don’t bother me	0.590*	0.110	−0.019	0.151
2	I didn’t feel like eating; my appetite was poor	0.295	0.427*	0.087	0.122
3	I felt that I could not shake off the blues even with help from my family or friends	0.217	0.674*	−0.023	0.016
4	I felt that I was as good as other people	−0.230	−0.168	0.516*	0.382*
5	I had trouble keeping my mind on what I was doing	0.798*	−0.019	0.115	0.117
6	I felt depressed	0.172	0.855*	0.078	−0.216*
7	I felt everything I did was an effort	0.023	0.323*	−0.243	−0.161
8	I felt hopeful about the future	−0.032	0.007	0.786*	−0.041
9	I thought my life had been a failure	0.475*	0.099	0.456*	0.100
10	I felt fearful	0.326*	0.200	0.346*	−0.009
11	My sleep was restless	0.458*	0.243	−0.064	−0.016
12	I was happy	0.032	0.115	0.804*	−0.005
13	I talked less than usual	0.262	0.283	0.239*	−0.047
14	I felt lonely	−0.008	0.660*	0.121	0.197*
15	People were unfriendly	0.221	0.171	−0.007	0.664*
16	I enjoyed life	0.120	0.085	0.724*	−0.019
17	I had crying spells	−0.092	0.880*	−0.003	0.175
18	I felt sad	−0.009	0.838*	0.121	0.112
19	I felt that people dislike me	0.173	0.215	0.082	0.642*
20	I could not get “going”	0.375	0.431*	0.079	0.162

Note. CES-D = Center for Epidemiologic Studies Depression Scale

^*= significant at p<0.05.

CFA Results and Factor Loadings

Based on the results of the EFA, we ran four-factor confirmatory analyses to assess model fit. Additionally, we tested the original four-factor structure proposed by Radloff (1977). Table 4 summarizes fit indices of our confirmed models. The four-factor EFA-retained model yielded acceptable model fit as demonstrated by model fit indices (WLSMV χ² = 239.37, CFI = 0.98, TLI = 0.97, RMSEA = 0.06, 90% CI: 0.04–0.07). The four-factor Radloff model yielded a slightly improved model fit (WLSMV χ² = 225.17, CFI = 0.98, TLI = 0.98, RMSEA = 0.05, 90% CI: 0.05–0.08). Figure 1 illustrates the final measurement models of the four-factor EFA-retained model and the original Radloff factor configuration. In the EFA-retained model, factors 1–3 corresponded to distracted thoughts, sadness, and mixed affect; factor 4—equivalent to Radloff’s fourth factor—corresponded to interpersonal conflict.

Table 4.

Confirmatory Factor Analysis Fit Statistics for the CES-D (n = 149)

	Comparison of Factor Models
Model	χ 2	df	χ2 p value	CFI	TLI	RMSEA (90% CI)
4-Factor Model (EFA-retained)	239.37	164	<0.001	0.98	0.97	0.06 (0.04–0.07)
4-Factor Model (Radloff)	225.17	164	<0.001	0.98	0.98	0.05 (0.05–0.08)

Note. CES-D = Center for Epidemiologic Studies Depression Scale; CFI = Comparative Fit Index; TLI = Tucker-Lewis Index; RMSEA = Root Mean Squared Error Approximation; Confidence Interval.

Figure 1.

Measurement Model of Four-Factor Solutions of (a) EFA-retained model and (b) original Radloff four-factor solution (n = 149)

Note. ^# = non-significant at p<0.05; * = reverse-coded items.

Discussion

To our knowledge, this study was the first to assess the factor structure of the CES-D using EFA and CFA with a community sample of younger American Indian women from the southeast. The four-factor structure derived from Radloff’s original four-factor configuration demonstrated the best model fit of the CES-D in our sample. This finding differs from previous studies reporting a two-factor (Schure & Goins, 2017) or three-factor (Barbosa-Leiker et al., 2021) solution as the best fitting models in samples of older American Indians. The contrast in model fit—a four-factor versus two- or three-factor structure—could reflect not only the heterogeneity in gender but also the cultural diversity among tribal groups. Indeed, studies using heterogeneous samples of racial and ethnic minorities show that the CES-D may differ by key characteristics of within-group populations, such as nationality, tribe, and gender (Adams et al., 2020; Goodwill, 2020; Rivera-Medina et al., 2010; Torres, 2012). Most of this research, however, has focused on people of African descent. Our study offers important parallel conclusions to the variations in CES-D construct validity among American Indian women residing in the Southeast region of the United States.

Differences between the four-factor EFA-retained structure from our sample and Radloff’s (1977) four-factor structure also emerged. Compared to Radloff’s original four-factor structure of the CES-D, we found different factor loadings for the “fearful,” “failure,” “talk,” “appetite,” and “effort” items. Notably, rather than into domains related to negative affect and somatic factors, these items were dispersed onto three factors characterized by distracted thoughts, sadness, and mixed affect. Additionally, the interpersonal factor domain was consistent across factor solutions, containing two items: “I felt that people dislike me” and “people were unfriendly.” As noted by Kim et al. (2011), the different factor configurations may be due, in part, to the underlying assumptions of the exploratory compared to confirmatory approaches. EFAs are primarily data-driven, while a priori hypotheses and researcher-imposed models guide CFAs. These important differences may inform our findings and demonstrate the importance of considering a wider variety of confirmatory models when testing the construct validity of measures among racial and ethnic minority samples. Additionally, in the case of measuring mental health outcomes, there may be no equivalent concepts of depression across cultures.

We offer several rationales to potentially explain our study findings. First, of the studies that have examined the CES-D factor structure, few have compared the more data-driven EFA-retained structures with the original, hypothesized Radloff configuration. Consequently, published research may skew towards presenting more uniform findings (e.g., publication bias) that support the original four-factor structure instead of allowing the data to drive a new configuration. Next, consistent with several other CES-D validation studies, racial and ethnic minority women experiencing depression may present differently than their White counterparts. Indeed, a recent systematic review illustrates that African American and Latina/Hispanic women were more likely to report physical symptoms associated with depression than their White counterparts (Lara-Cinisomo et al., 2020). In contrast, one study concluded that there was no significant difference in somatization between American Indian and Alaska Native women and other racial and ethnic groups (Uebelacker et al., 2009). In a separate study, Makambi et al. (2009) evaluated the factor structure of the CES-D in a subset of US Black women enrolled in a large prospective cohort study—the Black Women’s Health Study. Compared to their sample of healthy Black women, American Indian women in our sample had higher mean scores on all CES-D items except three (i.e., fearful, lonely, and talk). Moreover, we found incongruencies in factor loadings of the “fearful” and “talk” items across the two samples. Thus, American Indian women appear to present with more severe and culturally nuanced somatization of depressive symptoms than Black women. Divergences in symptom severity and factor loadings across these two racial and ethnic groups suggest that women’s experiences of depression reflect conceptualizations and expressions unique to their culture. Inferences drawn from this comparison must also consider the mean age difference (36.5 years, SD = 10.2 versus 53.8 years, SD = 9.7) between the American Indian and Black women in these samples since middle-aged adults generally report lower scores on the CES-D than young adults (Johnson et al., 2008). Given the mixed evidence, additional studies are needed to clarify the cause-effect relationship, namely, whether (a) cultural presentations of depression include more somatization or (b) if depression causes more somatization among racial and ethnic minority women.

Additionally, Lewis and colleagues (2015) elaborate that the two interpersonal items in the CES-D measure (“I felt that people dislike me” and “people were unfriendly”) may identify feelings of discrimination rather than clinically relevant symptoms of major depressive disorder, particularly among racial and ethnic minorities. Other researchers’ findings corroborate the importance of considering social influences and their role in shaping depressive symptoms among American Indians (Dinges & Duong-Tran, 1992). For American Indians, studies show that family closeness and, more broadly, connection to tribal culture are unique cultural factors that may be protective against depressed mood and other negative emotions (Hodge et al., 2009; Martin & Yurkovich, 2014). Family systems theory, which posits that families are relationally and emotionally interconnected, may be applied in future research to provide critical insights into the connection between tribal communities, families, and mental wellbeing among American Indians (Brown, 1999; Morgaine, 2001). Inversely, the endorsement of interpersonal conflict in the CES-D among American Indians may reflect disrupted relational ties within their family or community. Further research should consider disentangling these potential explanations by employing more robust longitudinal designs to examine the temporal sequence between disrupted relational ties among family members within tribal communities, discriminatory experiences, somatic presentations, and depressive symptoms.

Finally, we found that the “effort” item in our EFA-retained confirmatory model was nonsignificant; additionally, its factor loading was lower than the remaining CES-D items (i.e., 0.03 versus 0.33–0.94). Our findings are consistent with previous research in diverse populations (Adams et al., 2019; Brooks et al., 2022; Kim et al., 2011) and may imply that a single-item measure of effort demonstrates diminished reliability and validity due to its convergence with other multidimensional concepts of strain. Several qualitative studies affirm the interconnectedness of strain and negative emotions is a complex experience that may explain its inconsistency in the CES-D (Christiansen et al., 2019; Walls et al., 2007). Of the studies focused on the strain-emotion relationship among American Indian populations, most have highlighted its influence on suicide outcomes (Ivanich & Teasdale, 2018; LaFromboise & Malik, 2016; Walls et al., 2007; Yang et al., 2018). More research is needed to clarify how strain influences the expression of depressed mood among American Indians.

Limitations

The limitations of our study include the generalizability of findings, small sample size, lack of a separate confirmatory dataset, and minimal clinical implications. First, we based our findings on a gender-homogenous study sample within one tribal group. Thus, we caution against broad applicability to other diverse tribal populations across the United States. Next, the small sample size may have influenced the interpretation of our model fit indices, mainly the chi-square fit criteria, which is most sensitive to sample size. To reduce interpretative bias from a small sample size, we included relative and absolute measures of model fit that are considered less sensitive to sample size. It is also important to note that our sample size is comparable to samples in other studies. Another limitation was our inability to confirm the factor structure using a different dataset. Given our sample size of 150, splitting the sample would not make it viable for a separate confirmatory dataset based on the sample size parameters detailed in Kyriazos (2018). Finally, the CES-D is a population-based measure of depressive symptoms rather than a clinical diagnostic tool, which limits our ability to consider clinical implications. Apart from the mentioned limitations, our study contributes important findings on the construct validity of the CES-D among American Indian women. To our knowledge, this is the first study to explore and confirm the CES-D factor structure in a gender-homogenous sample of younger American Indian women. Findings from our study provide evidence on the measurement of depression from a population-based perspective; however, future studies should also focus on validating tools that are more appropriate for clinical diagnoses, such as the Patient Health Questionnaire (PHQ-9) and Composite International Diagnostic Interview (CIDI).

Conclusion

Our study confirms that the four-factor structure of the CES-D is the best model fit for our community sample of American Indian women. Future studies should examine the validity of the CES-D for measuring depressive symptoms in other tribal groups and gender-homogenous study samples. Consistent with other studies, a single-item measuring effort may not align consistently with other items in the CES-D. Researchers should consider whether effort or strain is conceptually distinct from depression, particularly among historically marginalized populations. Our study builds on extant research showing the original four-factor structure of the CES-D is well-suited for American Indians. Still, contrasting the Radloff structure with our EFA-retained structure provides an alternative interpretation. Applying our findings to clinical settings, the divergence between our EFA-retained model and the original Radloff four-factor structure may imply that providers should not rely on summary scores but rather on the symptom structure expressed by the individual.