Comprehensive investigation of factor structure and gender equivalence of the Beck Depression Inventory-II among nonclinical adolescents

doi:10.21203/rs.3.rs-3235827/v1

Download PDF

Research Article

Comprehensive investigation of factor structure and gender equivalence of the Beck Depression Inventory-II among nonclinical adolescents

https://doi.org/10.21203/rs.3.rs-3235827/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 06 Jun, 2024

Read the published version in European Child & Adolescent Psychiatry →

You are reading this latest preprint version

The Beck Depression Inventory-II (BDI-II) is a tool widely used to rate the severity of depressive symptoms. Regarding the factor validity of the BDI-II in adolescents, the two-factor solution describing “cognitive” and “somatic-affective” dimensions has been shown to be the most reported. Recently, bifactor model analysis has become popular to describe psychological constructs such as depression. Furthermore, data related to measurement invariance across gender is limited. This study aimed to examine plausible factor structures and gender equivalence of the BDI-II in a Brazilian sample of nonclinical adolescents (N = 1,184, aged 13–18 years, 59.1% females). Cross-validation of the BDI-II was performed through exploratory (EFA) and confirmatory factor analysis (CFA). Measurement invariance was evaluated using multigroup-CFA (MG-CFA). EFA suggested an oblique two-factor model depicting “affective-cognitive” and “somatic" dimensions. Simple structure and bifactor models from literature, as well as the estimated EFA model and its corresponding bifactor model were examined. All factor models revealed adequate and similar fitness, well-defined factors, and good reliability. The goodness-of-fit of bifactor models was slightly superior when compared to simple structure models. Bifactor analyses indicated a robust general factor with low reliable variance in total scores attributed to multidimensionality caused by the group factors in bifactor models. MG-CFA supported invariance across gender, confirming that the same BDI-II construct could be applied to both female and male adolescents. This study provides evidence that the BDI-II could be used by researchers and clinicians as a unidimensional measure of depressive symptoms in the adolescent population.

adolescence

depression

beck depression inventory-II

bifactor model

measurement invariance

Early symptoms of depression have been associated with a long-term risk for worse personal cognitive, emotional, and social outcomes [1–4]. Notwithstanding the huge burden of depression among adolescents, findings from several studies indicate its under-recognition, low help-seeking behavior, and under-treatment [5, 6]. Therefore, the identification of depressive symptoms at this life stage constitutes the first step toward an early intervention with the potential to mitigate adverse outcomes later in life [7]. For this purpose, the use of well-validated and psychometrically sound rating scales could be of the utmost importance.

The Beck Depression Inventory (BDI) [8] is a common measure for evaluating depression, being used worldwide for clinical and non-clinical adults and adolescents aged ≥ 13 [9]. The current version of the BDI, Second Edition (BDI-II) [10], taps psychological and somatic manifestations of 2-week major depressive episodes (MDD), as operationalized in the DSM-IV [10]. The BDI-II has been translated into many languages with evidence of reliability and validity across cultures [11, 12]. Although the BDI-II has become a popular scale for assessing depressive symptomatology among adolescents, its factor structure remains an issue of debate [13].

Originally, Beck et al. [8] proposed a structure of an oblique two-factor model in undergraduates, including “cognitive-affective” and “somatic” dimensions. Subsequent studies conducted using exploratory factor analysis (EFA), also revealed a two-factor model, but items were loaded heterogeneously, while other researchers also proposed the one and three-factor models across different adolescent samples [9, 11, 14, 15].

Confirmatory factor analysis (CFA) represents a powerful hypothesis-testing method to compare the structure and the fitness of plausible models for the underlying construct of the BDI-II. To date, there are few CFA studies of adolescents [16–19], which results varied from a unidimensional to multidimensional two- and three-factor models as the best-fitting model. In addition, there is a study that revealed that none of the models tested met the adequacy-of-fit criteria [16]. As a result of controversy over the factor structure of the BDI-II and the fact that most models have shown highly correlated factors, bifactor models have been recently proposed as an alternative solution in adolescent samples [17, 18, 20].

In the bifactor model, each item loads on a general factor and is also allowed to load on one of the two or more specified group factors. With this model, it was possible to distinguish the variance that could be attributed to the general factor and the group factors. Thus, some authors regard it as a more realistic representation of complex psychological constructs such as depression [21, 22]. However, the same authors also pointed out problems with bifactor models, such as a tendency to overfit, and anomalous results with small factor loadings and zero or negative group factor variances. [21, 22].

Ward [20] was the first researcher to examine the BDI-II, using the bifactor approach in CFA to model six data set from five studies in the literature (three clinical adult and two college samples). Findings indicated that the bifactor model with a general depression factor and two group factors (“cognitive” and “somatic” dimensions) fitted as well as or better than the two-factor models. Available studies with adolescent samples have reported similar findings [16, 17]. However, when the fit indices were jointly evaluated with parameter estimates, such as factor loadings and reliability of the general and group factors, the authors concluded that the BDI-II measured a single latent construct. Therefore, this perspective highlights that an accurate assessment of the BDI-II from factor analysis, including bifactor modeling, could increase the understanding of depressive phenomena and avoid misleading conclusions.

Another requirement for a scale validation is that its factor structure should be invariant across subgroups of the population [23]. Because several studies have shown higher mean BDI-II scores in female than male adolescents [24, 25], the measurement invariance can help to tease apart actual gender variations, either from a measurement-driven artifact or psychometric differences in item response [23]. So far, few studies have explored the gender-related measurement invariance of the BDI-II in adolescent samples [26].

Therefore, we aimed to examine the factor structure and the gender-related measurement invariance of the BDI-II in a representative sample of Brazilian nonclinical adolescents. We performed cross-validation of the BDI-II through EFA and CFA, in random half samples. Subsequently, we evaluated the goodness-of-fit, factor loading, and reliability of three simple structure and one bifactor models from literature, as well as the estimated EFA model and its corresponding bifactor model. Bifactor statistics were estimated to clarify whether a multidimensional construct was more appropriate than a unidimensional one. Finally, the gender-related measurement invariance was tested by sequentially investigating configural, metric, and scalar equality. Our hypothesis is that (1) the best structural model would be an oblique two-factor or a bifactor model; (2) the bifactor modeling approach could reveal a robust general factor; and (3) there would be a measurement invariance across gender.

Study population

Participants in the present study comprised a subsample of adolescents from a larger cross-sectional study entitled: “Adaptation and validation of the Portuguese version of the Beck Depression Inventory-II in the Brazilian population: A psychometric study” [27]. Ethics approval was acquired by “Comissão de Ética para Análise de Projetos de Pesquisa” (CAPPesq), University of São Paulo Medical School, Brazil (nr 0006/08). The study recruited 1,880 adolescents, aged 10–18, from public and private schools in the city of Sao Paulo, the fourth largest city in the world, by the means of a multistage cluster stratified sampling procedure. The current study included only adolescents aged 13–18, totaling 1,184. Further details on the sampling frame and procedures are provided in Methods S1 (supplementary material).

Measures

Beck Depression Inventory-II (BDI-II)

The BDI-II is a 21-item self-report questionnaire for assessing the presence and severity of depressive symptoms. The items comprise four ordinal categories from 0 through 3 – the absence (or “as usual”) of symptom in each item is scored as “0” and its presence is scored between 1 and 3. The higher the total BDI-II scores, the more severe the depressive state [10].

We used the Brazilian-Portuguese version of the BDI-II [31], which has demonstrated good reliability (Cronbach’s alpha 0.93) and validity among the Brazilian adult population, with a 70% of sensitivity and 87% specificity. Its concurrent validity and the predictive ability of the severity level were acceptable [12, 27, 28].

Statistical analysis

Analyses were conducted in R version 3.6.1 [29]. All statistical analyses were conducted taking in account the ordinal nature of the BDI-II scores.

The dataset of 1,184 adolescents was split randomly in half, with one half used for developing (EFA) and the other for testing models (CFA). Descriptive analysis was performed across the whole sample and in both half-split samples, namely the EFA and the CFA sample, using the Chi-squared test to check differences between subsamples. The polychoric correlation matrix and an EFA were conducted before CFA (Methods S2 and Table S1).

In addition to estimated model from EFA (EFA Model) and its corresponding bifactor model (Bifactor Model), four competing models from the literature were compared in CFA, as following:

Araya et al. [24], a single general factor model based on 592 Chilean high-school adolescents (53% females; M = 15.5, SD = 0.9 years).
Beck et al. [8], a two-factor oblique model based on 120 Canadian undergraduates (56% women; M = 19.5, SD = 1.8 years). The “cognitive-affective” (items 1, 3–14, and 17) and the “somatic” (items 15, 16, and 18–20) factors were core dimensions. Two items (items 2 and 21) did not load saliently on any of two factors.
Uslu et al. [30], a two-factor oblique model based on 300 Turkish high-school adolescents (57% females; M = 15.7, SD = 1.4 years). In comparison with Beck’s model, item loadings were different and core dimensions were “cognitive” (items 1–3, 5–9, 12 and 14) and “somatic-affective” (items 4, 11, 13, 15–20) factors. Item 21 had no salient loading.
Osman et al. [18], a bifactor model based on 414 North American high-school adolescents (49% females; M = 15.8, SD = 1.0 years). This model includes one general and two group factors: “somatic” (items 15, 16, 18–20) and “cognitive-affective” (items 1–14, 17, and 21) factors.

We estimated the CFA model with the Weighted Least Square Mean and Variance (WLSMV) [31]. Thresholds used for good fit in CFA are CFI/TLI ≥ 0.95; RMSEA ≤ 0.06 and those considered acceptable are CFI/TLI ≥ 0.90; RMSEA ≤ 0.08, with its 90% confidence interval [32].

The reliability of the models was assessed by McDonalds’s omega (ω), with values > 0.80 indicating good reliability [21, 33]. For bifactor models, we calculated (ω), omega-hierarchical (ω_H) and omega-hierarchical-subscale (ω_H−S), with values ≥ 0.70 for good reliability [21]. We also compared the ω and ω_H to investigate the degree of reliable variance in item responding [22]. For the variance attributed to the general factor, the value of ω_H should be divided by ω (i.e., ω_H/ω); for the remaining reliable variance attributable to the specific factors, ω should be subtracted from ω_H (i.e., ω – ω_H) [22]. Reise et al. [21] suggest 75% or higher accounted for by the general factor as the ideal amount of variance to justify the use of a total score, despite the presence of multidimensionality in the data.

In bifactor models, the dimensionality of the general and group factors was examined by the following bifactor statistical indices: Common Variation Explained (ECV); Item Common Variation Explained (I-ECV); and Percentage of Uncontaminated Correlations (PUC) [21, 22]. To the extent that PUC is high (> 0.80), the values of the indices such as ECV and ω_H are less important in predicting bias, and the common variance can be regarded as essentially unidimensional [22]. When PUC is < 0.80, researchers may consider ECV > 0.60 and ω_H > 0.70 as tentative benchmarks, indicating an essentially unidimensional rather than a multidimensional structure [21, 22]. Some authors recommend selecting items with I-ECV values > 0.80 to create a more unidimensional measure [21]. Finally, to assess possible distortions in factor loadings caused by the multidimensionality of a bifactor model, we calculated the relative bias as the average difference between the factor loadings on the general bifactor dimension and those on a strictly unidimensional model, values < 10–15% are acceptable [22].

Configural, metric and scalar invariance across gender was evaluated through multigroup-CFA (MG-CFA) [34]. First, the goodness of fit of the configural model was assessed. If configural invariance is held (CFI ≥ 0.95, TLI ≥ 0.95, and RMSEA ≤ 0.06), this model could be established as the baseline model against which subsequent models were compared. Metric and scalar invariance across gender groups is established when, after imposing a constraint, there is no significant fit deterioration (ΔRMSEA < 0.015 and ΔCFI < 0.01) compared to the previous model [35]. We also reported the Chi-square test with the correction suggested by Satorra & Bentler (S-Bχ²) [36], and its associated p-value (> 0.05 indicates that the compared models have the same fit).

Descriptive results

The whole sample comprised 1,184 adolescents, aged 13–18 (M = 14.9, SD = 1.5), 59.1% females, 73.4% from public schools, and 52.1% attending high school (Table S2). Means and standard deviations BDI-II item scores, and proportion of response choices 1, 2, or 3 for the whole sample, EFA and CFA sample are shown in Table S3. There were no differences between EFA and CFA samples (p ≥ 0.05)

EFA

The estimated EFA solution (EFA Model) was an oblique two-factor model (Table S4), explaining 42% of the total variance, with a factor correlation of 0.77. All items loaded saliently > 0.35 on one factor, without cross-loading. Factor-I represented the affective-cognitive dimension (composed by items 1–10, 12, 14 and 21), while Factor-II represented the somatic dimension (composed by items 11, 13, 15–20). The core items of Factor-I were “sadness”, “self-dislike” and “worthlessness”. The Factor-II was determined by items “tiredness” and “changes in sleep patterns” (Tables S1and S4).

The corresponding bifactor model of EFA Model (Bifactor Model) included a general depression factor and two group factors composed as follows: affective-cognitive with items 1–10, 12, 14 and 21; and somatic with items 11, 13 and 15–20. Four items (13, 17, 19 e 20) did not load significantly on the somatic factor.

CFA

Table 1 shows the results of six competing models. Simple structure models included the solutions proposed by: Araya et al. [24], Beck et al. [8], Uslu et al. [30], and the EFA Model. Bifactor models included the bifactor model proposed by Osman et al. [18] (hereafter, the Osman model) and the Bifactor Model.

All six models indicated satisfactory and similar fit indices to the sample data. The two bifactor models presented the best model fit, followed by oblique two-factor models and the one-factor model, consecutively.

The standardized factor loadings for all models were statistically significant, indicating that each latent construct was well-defined by its respective factor loadings, except for the bifactor models, where some items loaded with p < 0.05 only on the general factor. In addition, the items showed a moderate to high loading (average of 0.63, ranging from 0.39 to 0.80) for one- and oblique two-factor models. For bifactor models, the loadings were moderate to high on the general factor (average of 0.61, ranging from 0.39 to 0.80) and weaker in the group factors (average of 0.23, ranging 0.01 to 0.65).

The ω and ω_H ranged from 0.79 to 0.94, demonstrating good reliability of all models tested. The ω_H−S reduced substantially after controlling for a general factor in both bifactor models (ranging from 0.01 to 0.18). The reliable variance in total scores attributed to the general factor (ω_H/ω) was 91% for the Bifactor Model and 99% for the Osman model. The reliable variance in total scores attributed to the multidimensionality of the group factors (ω–ω_H) was 9% for the Bifactor Model and 1% for the Osman model.

The ECV was 0.82 for the Bifactor Model and 0.88 for the Osman model, meaning that the general factor explains 82% and 88% of the common variance extracted with 18% and 12% (respectively) of the common variance spread across group factors. Most I-ECV values for items of both bifactor models were ≥ 0.80. The average relative bias across factor loadings between the unidimensional model and the general factor in the bifactor models was 8% for the Bifactor Model and 1.4% for the Osman model. The PUC was 0.49 for the Bifactor Model and 0.38 for the Osman model.

In summary, all models revealed adequate and nearly identical fit in CFA, well-defined factors with strong loadings, and good reliability. The goodness-of-fit of the two bifactor models was slightly superior when compared to simple structure models. The bifactor models demonstrated good reliability and a robust general factor. The bifactor indices in both models were PUC < 0.50, ECV > 0.80, most of I-ECV > 0.80, ω_H > 0.80, ω_H−S < 0.20, ω_H/ω > 90%, ω–ω_H < 10%, and relative bias < 10%, which we interpreted as meaning the general depression factor was like the unidimensional model. Accordingly, this unidimensional measurement model should be regarded as the most acceptable, considering the fitness performance of the six CFA models, further appraisal of bifactor statistical indices, and the parsimonious criteria.

Table 1

Factor Loadings, Reliability Indices and Fit Statistics of the CFA Models
	Simple structure models							Bifactor models
BDI-II items	Araya et al. (2013)	Beck et al. (1996)		Uslu et al. (2008)		EFA Model¹		Bifactor Model²				Osman et al. (2008)
		CA	S	C	SA	AC	S	G	g_AC	g_S	I-ECV	G	g_S	g_CA	I-ECV
1	0.66	0.67		0.67		0.67		0.57	0.39		0.68	0.66		0.13	0.96
2	0.67	0.67		0.68		0.68		0.59	0.33		0.76	0.67		0.05	0.99
3	0.60	0.60		0.61		0.61		0.53	0.32		0.73	0.60		-0.01	1.0
4	0.61	0.61			0.62	0.62		0.57	0.23		0.86	0.60		0.29	0.81
5	0.58	0.59		0.60		0.59		0.54	0.24		0.83	0.59		-0.23	0.86
6	0.55	0.55		0.56		0.55		0.52	0.15		0.92	0.56		-0.36	0.71
7	0.78	0.78		0.79		0.79		0.65	0.51		0.62	0.78		0.11	0.98
8	0.66	0.66		0.67		0.67		0.60	0.28		0.84	0.66		-0.12	0.96
9	0.74	0.75		0.75		0.75		0.68	0.29		0.85	0.75		-0.24	0.91
10	0.58	0.59			0.60	0.59		0.54	0.24		0.83	0.59		0.07	0.98
11	0.39	0.39			0.40		0.41	0.40		0.40	0.50	0.39		-0.15	0.87
12	0.74	0.74		0.75		0.75		0.69	0.25		0.88	0.74		0.38	0.79
13	0.57	0.57			0.58		0.59	0.59		0.07	0.98	0.57		0.04	0.99
14	0.79	0.80		0.80		0.80		0.70	0.40		0.75	0.80		0.05	0.99
15	0.69		0.73		0.71		0.72	0.75		-0.43	0.75	0.66	0.27		0.85
16	0.54		0.57		0.56		0.57	0.57		0.16	0.92	0.52	0.23		0.84
17	0.66	0.66			0.68		0.69	0.69		-0.01	1.0	0.66		-0.03	0.99
18	0.55		0.58		0.56		0.57	0.57		0.29	0.79	0.54	0.12		0.95
19	0.65		0.68		0.66		0.68	0.67		0.10	0.98	0.64	0.07		0.99
20	0.66		0.69		0.67		0.69	0.69		-0.10	0.98	0.61	0.65		0.47
21	0.56	0.57		0.57		0.57		0.52	0.23		0.83	0.56		0.19	0.89
χ²	427.23	402.89		395.14		376.15		329.47				334.51
df	189	188		188		188		168				168
CFI	0.966	0.964		0.965		0.968		0.973				0.972
TLI	0.955	0.959		0.961		0.965		0.966				0.965
RMSEA [90%-CI]	0.046 [0.040–0.052]	0.044 [0.038–0.050]		0.043 [0.037–0.049]		0.041 [0.035–0.047]		0.040 [0.034–0.047]				0.041 [0.035–0.047]
ω	0.93	0.92	0.79	0.91	0.85	0.91	0.83	0.94	0.91	0.85		0.94	0.92	0.80
ω_H/ ω_H−S								0.85	0.18	0.08		0.93	0.01	0.13
ECV (%)								82	13	5		88	6	6
PUC								0.49				0.38

Significant factor loading values (p < 0.05) shown in bold; ω = omega reliability coefficient; ω_H = omega hierarchical; ECV = explained common variance; I-ECV = Item Common Variation Explained; PUC = Percentage of Uncontaminated Correlations; CFI = comparative fit index; TLI = Tucker–Lewis Index; RMSEA = root-mean-square error of approximation; df = degrees of freedom; ¹model with two-factors correlated derived from EFA; ² bifactor model derived from EFA. CA = Cognitive-Affective; S = Somatic; C = Cognitive; SA = Somatic-Affective; AC = Affective-Cognitive; G = general factor; gx = group factors for the bifactor models.

Measurement Invariance

The unidimensional model was selected for assessing the measurement invariance across gender groups. The configural model (M1) showed good fit indexes (CFI = 0.961, TLI = 0.956 and RMSEA = 0.046). The fit indices differences between M1 and metric model (M2), and M2 and scalar model (M3) indicated that the compared models have the same fit (ΔCFI = 0.001, ΔRMSEA < 0.001 and p-value of S-Bχ² > 0.05) (Table 2). Therefore, configural, metric, and scalar invariance was established across gender.

Table 2

Measurement invariance across gender
Models	S-Bχ²	df	p	CFI	TLI	RMSEA [90%-CI]	ΔCFI	ΔRMSEA
M1: Configural invariance (no constraints imposed)	609.35	378	-	.961	.956	.046	-	-
M2: Metric invariance (thresholds equal)	634.43	399	0.96	0.960	0.958	0.045 [0.038–0.052]	0.001	0.001
M3: Scalar invariance (thresholds/loadings equal)	661.96	419	0.45	0.959	0.959	0.044 [0.038–0.051]	0.001	0.000
S-Bχ² = Satorra–Bentler scaled χ²; df = degrees of freedom; CFI = comparative fit index; RMSEA = root-mean-square error of approximation; TLI = Tucker–Lewis Index; ΔCFI = difference in CFI between the compared models; ΔRMSEA = difference in RMSEA between the compared models; p-value > 0.05

The present study provides unique data on depressive symptoms in a Brazilian sample of nonclinical adolescents. To the best of our knowledge, this study was the first to comprehensively examine the reliability, factor validity, and gender equivalence in diverse models of BDI-II, including bifactor models. Our analyses yielded several promising findings. In the EFA, the best model was the oblique two-factor one, with affective-cognitive and somatic dimensions. In the CFA, after checking measurement indicators of competing models, the unidimensional structure emerged as the best representative in our dataset. In fact, a robust general factor conspicuously emerged in tested bifactor models. In addition, we found evidence of measurement invariance across gender. Examination of bifactor models along with indicators of reliability, item variance and residuals provided fresh insights on the applicability of the BDI-II. A standardized and widely used instrument, such as the BDI-II, is pivotal for helping clinicians and researchers to understand the expression of depression in adolescents in different cultures. Further exploration of moderators that potentially affect the factor structure, such as culture, language, and urbanicity, among others, could support the utility of a measurement tool in different settings and contexts.

Regarding the EFA, our findings support a similar factor structure to the oblique two-factor solution (affective-cognitive and somatic) proposed by Beck et al. [8], Uslu et al. [30] and Huang & Chen [15]. Although these researchers also investigated nonclinical youths, the item loadings vary slightly across studies. Some items such as sadness, past failure, self-dislike, suicidal thoughts, and worthlessness have consistently loaded on the dominant cognitive-affective dimension. In contrast, other items such as loss of energy, changes in sleep patterns, changes in appetite, concentration difficulties, and tiredness have defined the somatic dimension. Importantly, previous clinical descriptions have reported somatic symptoms, such as loss of energy, changes in appetite and weight, fatigue, and insomnia as particularly common presentations among adolescents diagnosed with depression [4, 37]. While symptoms, such as concentration difficulties, worthlessness, guilt, and changes in sleep patterns could be viewed as early warning signs for MDD, symptoms like sadness, anhedonia and suicidal thoughts could indicate the severity of the depressive condition [38]. Although the affective-cognitive dimension in adolescents determined the core dominant data covariance, the somatic dimension could be a helpful marker of a potential depressive episode.

The oblique two-factor solution could not always be replicated across different samples. Some factor models failed to converge on the underlying structure; therefore, an alternative representation of two distinct dimensions should be considered. When highly correlated two-factor models were found in an EFA solution, this could indicate the presence of a general latent variable of depression partitioning for data distribution. Thus, the bifactor model has emerged as a sophisticated alternative for structural analysis of the BDI-II. However, so far, there are no clear guidelines or benchmarks for evaluating this measurement approach. To address this gap, we assessed bifactor CFA models by aligning fit criteria with bifactor indices reported in the literature.

Our CFA results revealed the unidimensional model as the best representation of observed data, considering substantive, statistical, and practical fit. The CFA is a theoretically grounded method. In our study, the two bifactor models showed a slightly better fit than simple structure models. The results of our parameter estimators for both bifactor models confirmed the existence of a strong general factor and indicated that most of the reliable variance in the total score was attributable to these factors. Regarding group factors, we observed several low loadings across all subgroups. For example, the indecisiveness, irritability, and concentration difficulties items clearly loaded on general depression factors and disappeared on specific factors.

Taken together, these results pointed to the possible existence of a factor over extraction and reinforced the BDI-II as a unidimensional scale. This is in line with the results of studies in adult samples [15]. In contrast, the scant studies available that conducted CFA with one, two- and three-factors, and second-order or bifactor models in adolescents revealed diverse findings. In clinical settings, among the two existing studies of CFA, one [16] concluded that none of the tested models met all adequacy-of-fit criteria, while another [17] supported a bifactor and a two-factor model as plausible solutions. To date, there are five studies of nonclinical samples. Byrne et al. [11] proposed a second-order factor model defined by three lower order factors comprising “negative attitude”, “performance difficulty”, and “somatic elements” and one higher order factor representing general depression. The best-fitting model, that of Wu and Huang [25], relies on three similar factors Byrne et al. model [14], but without the second-order factor structure. Lee et al. [19] also concluded that the BDI-II was best represented by the three-factor model. Osman et al. [18] found that all tested models showed good fit to the sample data, with superior fitness to the bifactor model (one general factor and two factors, i.e, somatic and cognitive-affective). After examining the loadings, they confirmed this model as a plausible alternative. Previously, Whisman et al. [39] had reported the Beck et al. [8] original oblique two-factor model (somatic-affective and cognitive), slightly modified, as the best fit model to older adolescents. It is important to note that, except for the Osman et al. study [18], the best CFA model was chosen only by the fit criterion. Thus, further CFA studies are needed to improve our understanding of the underlying construct of the BDI-II in nonclinical adolescents, considering further reliability indicators and bifactor statistics. The present study’s findings demonstrated that the one-factor model provided a better representation of the BDI-II to the data than more complex (i.e., bifactor) models that are becoming increasingly popular approaches for measuring depression psychopathology. However, a general depression factor does not invalidate the differentiation of specific depression factors as a distinct depression group factor, but it could prompt advancement in the comprehension of different etiologies and associated symptomatology.

Based on the MG-CFA, the unidimensional factor structure was invariant across gender. In other words, our findings suggested that the same BDI-II construct can be applied to both female and male adolescents. This accords with our original hypothesis and one study that also used a one-factor model for invariance analysis [24], but is somewhat at odds with others [16, 17, 25]. Although these studies found some degree of partial invariance, most of them concluded the effect of non-invariance on observed mean differences could be negligible. There are two possible reasons for the varied findings among the studies: the chosen baseline model and the characteristic of the samples. Measurement invariance is critically important when comparing groups. If measurement invariance cannot be established, then the finding of a between-group difference cannot be unambiguously interpreted [37]. In summary, our findings call attention to the need for further investigations, which could drive more specific care for female and male adolescents with depressive symptoms.

Limitations and future research

The findings should be interpreted considering the following limitations. We used a large and representative sample, and our results allow some generalization to the general adolescent population. However, it comprised only urban school-age adolescents and thus, the findings may need further validation in rural and dropped-out-of-school adolescents. The methods applied are well-established in the literature, suitable for data type, and are reproducible in other studies. Nevertheless, the examination of gender was based on a categorical variable (gender binary). Recent advances in measurement invariance analysis have enabled the test of invariance within a dimensional perspective [40]. Thus, further studies should investigate the invariance of the BDI-II across gender dimensions that may allow for more inclusive conclusions. Furthermore, the effect of puberty on the factor structure of the BDI-II was not examined in our data. The increasingly high level of lifetime prevalence of depression in females first arises after puberty and seems to be associated with hormonal changes [4]. Thus, further fine-grained gender-related invariance analyses across early and late adolescence may help to understand this gender gap. Lastly, given that Sao Paulo is a global city, the effect of race and ethnicity on the factor structure of depression is an important issue and should be addressed by future researchers.

This study suggested the BDI-II is a unidimensional measure of depressive symptoms in nonclinical adolescents, being invariant across gender. Fit indices evaluated jointly with a set of parameter estimates could uncover valuable information on measurement models, which in turn scale up our understanding of dimensionality and usefulness of the scale by researchers and clinicians.

Ethical approval Ethics approval was acquired by “Comissão de Ética para Análise de Projetos de Pesquisa” (CAPPesq), University of São Paulo Medical School, Brazil (nr 0006/08). All participants and their parents or guardian provided fully informed consent.

Competing interests The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Authors' contributions APFM played a lead role in conceptualization, data curation, formal analysis, investigation, methodology, project administration, and writing of original draft. CG played a supporting role in methodology and writing of review and editing. YPW played a supporting role in methodology and writing of review and editing.

Funding This research was supported in part by a grant from the Fundação de Amparo a Pesquisa do Estado de São Paulo – Brasil (grant number: 08/11415-9) to author YPW and Coordenação de Aperfeiçoamento de Pessoal de Nível Superior – Brasil (grant number: 001) to author APFM. The funders were not involved in the study design, collection, analysis, interpretation of data, the writing of this article or the decision to submit it for publication.

Availability of data and materials The datasets and other materials used during the current study are available from the corresponding author upon reasonable request.

GBD 2019 Mental Disorders Collaborators (2022) Global, regional, and national burden of 12 mental disorders in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet Psychiatry 9(2):137–150. https://doi:10.1016/S2215-0366(21)00395-3
Clayborne ZM, Varin M, Colman I (2019) Systematic Review and Meta-Analysis: Adolescent Depression and Long-Term Psychosocial Outcomes. J Am Acad Child Adolesc Psychiatry 58(1):72–79. https://doi:10.1016/j.jaac.2018.07.896.
Lewinsohn PM, Clarke GN, Seeley JR, Rohde P (1994). Major depression in community adolescents: age at onset, episode duration, and time to recurrence. J Am Acad Child Adolesc Psychiatry 33(6):809–818. https://doi:10.1097/00004583-199407000-00006
Thapar A, Eyre O, Patel V, Brent D (2022). Depression in young people. Lancet 400(10352):617–631. https://doi:10.1016/S0140-6736(22)01012-1
Merikangas KR, He JP, Burstein M, Swanson SA, Avenevoli S, Cui L, et al (2010) Lifetime prevalence of mental disorders in U.S. adolescents: results from the National Comorbidity Survey Replication–Adolescent Supplement (NCS-A). J Am Acad Child Adolesc Psychiatry 49(10):980–989. https://doi:10.1016/j.jaac.2010.05.017
Morris J, Belfer M, Daniels A, Flisher A, Villé L, et al (2011) Treated prevalence of and mental health services received by children and adolescents in 42 low-and-middle-income countries. J Child Psychol Psychiatry 52(12):1239–1246. https://doi:10.1111/j.1469-7610.2011.02409.x
Whiteford HA, Degenhardt L, Rehm J, Baxter AJ, Ferrari AJ, et al (2013). Global burden of disease attributable to mental and substance use disorders: findings from the Global Burden of Disease Study 2010. Lancet 382(9904):1575–1586. https://doi:10.1016/S0140-6736(13)61611-6
Beck AT, Steer RA, Brown GK (1996) Manual for the Beck Depression Inventory-II. San Antonio, TX: Psychological Corporation).
Steer RA, Kumar G, Ranieri WF, Beck AT (1998) Use of the beck depression inventory-II with adolescent psychiatric outpatients. J Psychopathol Behav Assess 20:127–137. https://doi:10.1023/A:1023091529735
American Psychiatric Association (1994) Diagnostic and Statistical Manual of Mental Disorders (4th ed.). Washington: DC.
Byrne BM, Stewart SM, Lee PWH (2004) Validating the beck depression inventory– II for Hong Kong community adolescents. Int J Test (4):199–216. https://doi:10.1207/s15327574ijt0403_1
Wang YP, Gorenstein C (2013) Psychometric properties of the Beck Depression Inventory-II: a comprehensive review. Br J Psychiatry 35(4):416–431. https://doi:10.1590/1516-4446-2012-1048
Stockings E, Degenhardt L, Lee YY, Mihalopoulos C, Liu A, et al (2015) Symptom screening scales for detecting major depressive disorder in children and adolescents: a systematic review and meta-analysis of reliability, validity and diagnostic utility. J Affect Disord 15(174):447–463. https://doi:10.1016/j.jad.2014.11.061
Bennett DS, Ambrosini PJ, Bianchi M, Barnett D, Metz C, et al (1997) Relationship of Beck Depression Inventory factors to depression among adolescents. J Affect Disord 45(3):127–134. https://doi:10.1016/s0165-0327(97)00045-1
Huang C, Chen JH (2015) Meta-Analysis of the Factor Structures of the Beck Depression Inventory-II. Assessment 22(4):459–472. https://doi:10.1177/1073191114548873
Osman A, Kopper BA, Barrios F, Gutierrez PM, Bagge CL (2004) Reliability and validity of the Beck depression inventory–II with adolescent psychiatric inpatients. Psychol Assess 16(2):120–132. https://doi:10.1037/1040-3590.16.2.120
Keller F, Kirschbaum-Lesch I, Straub J (2020) Factor Structure and Measurement Invariance Across Gender of the Beck Depression Inventory-II in Adolescent Psychiatric Patients. Front Psychiatry 23(11):527–559. https://doi:10.3389/fpsyt.2020.527559
Osman A, Barrios FX, Gutierrez PM, Williams JE, Bailey J (2008) Psychometric properties of the beck depression inventory–II in nonclinical adolescent samples. J Clin Psychol 64(1):83–102. https://doi:10.1002/jclp.20433
Lee EH, Lee SJ, Hwang ST, Hong SH, Kim JH (2017) Reliability and Validity of the Beck Depression Inventory-II among Korean Adolescents. Psychiatry Investig 14(1):30–36. https://doi:10.4306/pi.2017.14.1.30
Ward LC (2006) Comparison of factor structure models for the beck depression inventory–II. Psychol Assess 18(1):81–88. https://doi:10.1037/1040-3590.18.1.81
Reise SP, Scheines R, Widaman KF, Haviland MG (2013) Multidimensionality and structural coefficient bias in structural equation modeling: A bifactor perspective. Educ Psychol Meas 73(1):5–26. http://dx.doi.org/10.1177/0013164412449831
Rodriguez A, Reise SP, Haviland, MG (2016b) Evaluating bifactor models: Calculating and interpreting statistical indices. Psychol Methods 21(2):137–150. https://doi.org/10.1037/met0000045
Meredith W (1993) Measurement invariance, factor analysis and factorial invariance. Psychometrika 58:525–543. https://doi.org/10.1007/BF02294825
Araya R, Montero-Marin J, Barroilhet S, Fritsch R, Gaete J, et al (2013) Detecting depression among adolescents in Santiago, Chile: sex differences. BMC Psychiatry (23):13:122. https://doi:10.1186/1471-244X-13-122
Wu PC, Huang TW (2014) Gender-related invariance of the Beck Depression Inventory-II for Taiwanese adolescent samples. Assessment 21(2):218–226. https://doi:10.1177/1073191112441243
Whisman MA, Perez JE, Ramel W (2000) Factor structure of the Beck Depression Inventory-Second Edition (BDI-II) in a student sample. J Clin Psychol 56(4):545–551. https://doi:10.1002/(sici)1097-4679(200004)56:4<545::aid-jclp7>3.0.co;2-u
Gorenstein C, Wang YP, Argimon IL, Werlang BSG (2011) Manual do Inventário de Depressão de Beck - BDI-II. São Paulo: Casa do Psicólogo.
Gomes-Oliveira MH, Gorenstein C, Lotufo Neto F, Andrade LH, Wang,YP (2012) Validation of the Brazilian Portuguese version of the Beck Depression Inventory-II in a community sample. Br J Psychiatry 34(4):389–394. https://doi:10.1016/j.rbp.2012.03.005
RStudio Team (2020) Rstudio: Integrated Development Environment for R [Computer software manual]. Boston, MA. Available online at http://www.rstudio.com/
Uslu RI, Kapci EG, Oncu B, Ugurlu M, Turkcapar H (2008) Psychometric properties and cut-off scores of the beck depression inventory-II in Turkish adolescents. J Clin Psychol in Medical Settings 15(3):225–233. https://doi:10.1007/s10880-008-9122-y
Li CH (2016) Confirmatory factor analysis with ordinal data: Comparing robust maximum likelihood and diagonally weighted least squares. Behav Res Methods 48:936 – 949. http://dx.doi.org/10.3758/s13428-015-0619-7
Hu L, Bentler PM (1999) Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Struct Equ Modeling 6(1):1–55. https://doi:10.1080/10705519909540118
McDonald RP (1999) Test theory: A unified approach. Mahwah, NJ: Erlbaum.
Wu H, Estabrook R (2016) Identification of confirmatory factor analysis models of different levels of invariance for ordered categorical outcomes. Psychometrika 81(4):1014–1045. https://doi:10.1007/s11336-016-9506-0.
Chen FF (2007) Sensitivity of goodness of fit indexes to lack of measurement invariance. Struct Equ Modeling 14(3):464–504. http://dx.doi.org/10.1080/10705510701301834
Satorra A, Bentler PM (2010) Ensuring Positiveness of the Scaled Difference Chi-square Test Statistic. Psychometrika 75(2):243–248. https://doi:10.1007/s11336-009-9135-y
Nardi B, Francesconi G, Catena-Dell'osso M, Bellantuono C, (2013) Adolescent depression: clinical features and therapeutic strategies. Eur Rev Med Pharmacol Sci 17(11):1546–1551.
Cole DA, Cai L, Martin NC, Findling RL, Youngstrom EA, et al (2011). Structure and measurement of depression in youths: applying item response theory to clinical data. Psychol Assess 23(4):819–833. https://doi:10.1037/a0023518
Whisman MA, Judd CM, Whiteford NT, Gelhorn HL (2013) Measurement invariance of the Beck Depression Inventory-Second Edition (BDI-II) across gender, race, and ethnicity in college students. Assessment 20(4):419–428. https://doi:10.1177/1073191112460273
Kolbe L, Molenaar D, Jak S, Jorgensen TD (2022) Assessing measurement invariance with moderated nonlinear factor analysis using the R package OpenMx. Psychol Methods 4. https://doi:10.1037/met0000501

No competing interests reported.

SupplementaryMaterial.docx

Download PDF

Journal Publication

published 06 Jun, 2024

Read the published version in European Child & Adolescent Psychiatry →

Editorial decision: Revision requested
22 Dec, 2023
Reviews received at journal
22 Dec, 2023
Reviewers agreed at journal
16 Oct, 2023
Reviews received at journal
28 Aug, 2023
Reviewers agreed at journal
14 Aug, 2023
Reviewers agreed at journal
14 Aug, 2023
Reviewers invited by journal
14 Aug, 2023
Editor assigned by journal
08 Aug, 2023
Submission checks completed at journal
08 Aug, 2023
First submitted to journal
04 Aug, 2023

You are reading this latest preprint version

Comprehensive investigation of factor structure and gender equivalence of the Beck Depression Inventory-II among nonclinical adolescents

Status:

Journal Publication

Version 1

Abstract

Introduction

Method

Study population

Measures

Beck Depression Inventory-II (BDI-II)

Statistical analysis

Results

Descriptive results

EFA

CFA

Measurement Invariance

Discussion

Limitations and future research

Conclusions

Declarations

References

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1