¤Current address: Escuela de Psicología, Pontifica Universidad Católica Madre y Maestra, Santiago de los Caballeros, República Dominicana.
* E-mail: od.ude.mmcup@aicragz, moc.liamg@aicrageolioz
Received 2017 Jun 22; Accepted 2018 Jun 13.
Copyright © 2018 García-Batista et al
This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Associated Data
S1 File: Dataset. BDI dataset in SPSS format.
[SAV]
pone.0199750.s001.sav [30K]
GUID: E6662881-3D30-436C-9369-FC735CB0014D
All relevant data are within the paper and its Supporting Information files.
Abstract
The Beck Depression Inventory-II [BDI-II] is currently one of the most widely used measures in both research and clinical practice for assessing depression. Although the psychometric properties of the scale have been well established through many studies worldwide, so far there is no study examining the validity and reliability of BDI-II in Republic Dominican. The purpose of the present study was twofold: [a] to examine the latent structure of BDI-II by testing several competing models proposed in the literature; and [b] to provide evidence of validity and reliability of the BDI-II in Republic Dominican. Confirmatory factor analysis indicated that a bifactor model with a general depression factor and three specific factors consisting of cognitive, affective and somatic showed the best fit to the data. Internal reliability was moderate to high for all subscales and for the total scale. Scores on BDI-II discriminated between clinical and general population, supporting for external validity. Practical implications are discussed and suggestions for further research are also made.
Introduction
Depression is a common mood disorder that affects individual’functioning individual ‘functioning across different domains. It is currently known that more than 350 million people suffer from depression worldwide and that it significantly contributes to the global burden of disease []. Depression stands out not only for its high prevalence, but also due to the probability of associated relapse and recurrence. Another setback is the high financial cost that it entails, which translates into low productivity, workplace absenteeism, outpatient care, hospitalizations and pharmacological treatments [].
According to the World Health Organization [] depression is the leading cause of years lived with disability [YLD], and the most prevalent disorder among serious psychiatric disorders in primary care setting. This disorder is characterized by changes in sleep, appetite and psychomotricity, decreased concentration and decision-making ability, loss of self-confidence, feelings of inferiority or worthlessness and guilt, as well as despair and recurrent thoughts of death with ideation, planning and/or suicidal acts.
So far, the Beck Depression Inventory-II [BDI-II] has become one of the most widely used measures to assess depressive symptoms and their severity in adolescents and adults []. The BDI-II [] is a 21-item self-report measure that taps major depression symptoms according to diagnostic criteria listed in the Diagnostic and Statistical Manual for Mental Disorders []. Items are summed to create a total score, with higher scores indicating higher levels of depression. It is worth noting that the BDI-II is not only extensively applied for research purposes but also in clinical practice, being the third test most used among Spanish professionals [].
Since its publication, a number of studies have examined the validity and reliability of BDI-II across different populations and countries []. Results have consistently shown good internal consistency and test-retest reliability of the BDI-II incommunity [,,] adolescent and adult clinical outpatients [] as well as in adult clinical inpatients []. Criterion-based validity have also shown acceptable sensitivity and specificity of the BDI-II for detecting depression, supporting its clinical utility as an aid measure for diagnostic purposes [,,]. In contrast, findings concerning BDI-II factor structure have been somewhat inconsistent. Particularly, while Beck et al. [] two-factor correlated model composed of a cognitive-affective and a somatic factors has been supported in many studies [,,]; there are others studies which identified a single factor [,], two alternative factors consisting of somatic-affective and cognitive [,], three factors corresponding to cognitive, somatic and affective [,,], and an alternative three-factor model including negative attitude, difficulty and somatic [–]. Less frequently, four [] and fivefactors [] have also been reported.
Additionally, more sophisticated analysis into the BDI-II factor structure including hierarchical and bifactor models have been tested. Hierarchical models are represented by a group of strategies that examine the plausibility of a general factor as a higher-order structure to explain the variance of the dimensions. Bifactor models, in contrast, allow to examine a non-hierarchical general factor independently of the specific factors and to simultaneously test the extent to which the common variance between items are explained by the orthogonal general factor and by the specific factors that are tested []. By doing so, bifactor models represent a useful strategy to examine if a construct of interest can be viewed primarily as unidimensional or multidimensional and, subsequently, the way in which scores should be computed.
Results from hierarchical and bifactor BDI-II models supported both models. For example, Byrne et al. [] found that a hierarchical model comprising one general factor of depression and three factors of negative attitude, performance difficulty and somatic elements fitted well to data and were fully invariant across Hong Kong and American adolescents. Subica et al. [] compared a unidimensional model, three alternative two-factor models and three bifactor models including an independent general depression factor and specific factors. They found that none of the two-factor models have acceptable fit and, in contrast, all the corresponding bifactor models showed good fit indices, concluding that only BDI-II total score should be used to measure the severity of depression. Similarly, McElroy et al. [] tested fifteen competing BDI-II models including unidimensional, multidimensional and bifactor models, and revealed that bifactor models provided the best fit to the data, supporting the view that BDI-II assesses a single latent construct. Finally,Vanheule et al. [] did not find confirmatory evidence for bifactor models but, instead, they found that a three-factor model consisting of affective, cognitive and somatic factors provided better fit to data in clinical and non-clinical samples.
Current study
In summary, although factorial data suggests that bifactor models outperform multidimensional models–regardless of the number of specific factors–findings are not conclusive [,,]. Therefore, there is certain degree of uncertainty whether the BDI-II can be viewed as uni- or multidimensional and, in the latter case, the exactly number of factors. Thus, further research is needed into the latent structure of BDI-II. Addressing this issue may have not only practical implications [i.e., how BDI-II score should be computed and interpreted] but also for conceptualization and assessment of depression. Moreover, despite that cultural convergence is being accelerated due to increased globalization [] and that major depression has been reported worldwide [] there are considerable cross-cultural differences in the symptomatology of depression []. To the extent that depression’ symptoms and inner experience may differ across cultural backgrounds [], findings cannot be generalized. More importantly, the detection and treatment of depression have become a matter of high priority in low and middle-income countries [] such as Dominican Republic, despite psychometrically validated measures are currently lacking. Therefore, the purpose of the present study was twofold. First, to determine the most appropriate BDI-II factor structure by examining several competing factor models that have been reported in previous studies. Secondly, to examine the validity and reliability of BDI-II in Dominican Republic.
Method
"This research had the revision and approval of the National Council of Bioethics in Health/ Consejo Nacional de Bioética en Salud [CONABIOS] of the Dominican Republic. The protocol registration number in CONABIOS was 028–2014."
Participants
One thousand and forty individuals [54.9% women and 45.1% men] from Dominican Republic participated in the study. The mean age was 27.07 [SD = 11.18]. Participants were selecting by convenience from general population [N = 797] and hospital population [N = 243]. Within the hospital sample, 76.5% came in for routine checkups, 15.3% sought help for cardiac and hypertension conditions and 8.2% went to the psychiatric service.
Measure
Depression Inventory–II [BDI–II] []. The Self-report study based on the symptoms described by the Diagnostic and Statistical Manual of Mental Disorders [DSM-IV] [], which makes measuring depressive severity possible. This version of the inventory consists of 21 items, in which four response options are presented on a scale of 0 to 3. For example, to measure pessimism [item 2] the response options used range from “I am not particularly discouraged about the future” [score of 0] to “the future is hopeless and things cannot improve” [score of 3]. In this study we are using the Spanish version of Beck Depression Inventory-II [], which has an excellent reliability coefficient of .92. Its content validity is ensured because most of its items are equivalent to the DSM-IV criteria for depression. Its construct validity has also been tested successfully by comparing scores with other measures for depression.
Procedure and data analysis
While Dominicans’ native language is the same that the language BDI-II version used in this study [i.e., Spanish], there are linguistic characteristics that may vary substantially. Thus, even using the exactly same words the interpretation and meaning may be quite different []. Therefore, a pilot study was first conducted to ensure that participants correctly understood the content of BDI-II items. Fifteen people were asked to complete the scale and write down items that were unclear or incomprehensible, as well as any other aspect of the scale that may deem relevant. Once the activity was completed, a focus group was used to enable individuals to share their appreciations concerning items, response format, instructions, and to check for discrepancies in the interpretation or meanings. There was neither difficulty in understanding nor negative commentaries about the scale content. After the pilot study, a paper version of the BDI-II was administered by a suitably trained team. All participants agreed to participate voluntarily and provided written consent prior to complete the inventory and after information about purposes of the study were provided. Preliminary analysis using SPSS v20 was carried out to examine outliers, missing values and to test assumptions of univariate and multivariate normality. Next, internal structure of the BDI-II was assessed using confirmatory factor analysis [CFA] through AMOS v20 []. Since Mardia’s kurtosis multivariate coefficient was 338.70 –thus indicating a significant deviation from multivariate normality according to benchmarks []–the Asymptotic Distribution-Free method was used for model estimation. To examine model fit,the chi-square value [χ2], the comparative fit index [CFI], the goodness-of-fit index [GFI], the Tucker-Lewis index [TLI] and the Root Mean Square Error of Approximation [RMSEA]. The cut-off points of values greater than .95 reported by Hu and Bentler [] and Joreskog and Sorbom’s [] were used for the CFI and GFI indices in order to consider an optimal fit, and greater than .90 for an acceptable fit. On the other hand, values lower than .06 for the RMSEA are considered optimal and lower than .08 are considered acceptable. As an additional criterion, the χ² value was divided by the degrees of freedom [χ²/df], with the aim of obtaining values lower than 3 in order to consider the model a good fit [,]. In order to compare the fit of the models, the Akaike Information Criterion [AIC] was also considered. According to this index, those models that present values lower than AIC provide a better fit. Also, when verifying the fit of a bifactor model, it makes sense to consider additional indices, mainly the hierarchical omega [ωH], the percentage of explained common variance [ECV] and the percentage of uncontaminated correlations [PUC].
In addition, the internal consistency was evaluated using Cronbach's α statistic, and validity evidence was provided by comparing the BDI-II scores of the general population and the hospital population; to do so, successive Student's t tests were carried out for independent samples applying a Holm–Bonferroni adjustment to control for Type 1 error. In this case, a stepwise procedure is used where each p-value is compared with α/[n—i + 1] for rejection. The comparison continues in a sequential increasing order [from i-1 and proceeding in order] until the first nonrejection. This method has demonstrated to be statistically more powerful for controlling Type 1 error compared to Bonferroni adjustment []. In order to assess the effect size of these differences, Cohen's d was calculated, with values around .30 considered as small effects, values around .50 as medium effects and values greater than .80 as large effects [].
Results
Model comparisons
Based on previous BDI-II research findings, several competing models were tested including one, two, three-factor models and bifactor models. In particular, Model 1 assumes depression as a unitary construct and, therefore, all BDI-II items were allowed to load into a single factor [“Depression”] []; Model 2 tested a two-factor model represented by “cognitive-affective” and “somatic” factors []; Model 3 tested the original two-factor model identified by Beck et al. [], namely, “somatic-affective” and “cognitive” factors; Model 4included three factors corresponding to “cognitive”, “affective” and “somatic” []; Model 5 tested an alternative three-factor model consisting of “negative attitude”, “difficulty” and “somatic” [] [Table 1]. Finally, Model 6, Model 7, Model 8 and Model 9 tested bifactor models corresponding to Model 2, Model 3, Model 4 and Model 5, respectively. In these models, an orthogonal general factor called “depression” was tested along with the specific proposed factors. Results are summarized in Table 2. In general, neither the unidimensional model nor the one, two and three factor models reached acceptable fit indices. In contrast, all the corresponding bifactor models fitted well to the data. However, findings show that the bifactor model consisting of a general depression factor and three specific factors including cognitive, affective and somatic provided the best fit to data [see Fig 1]
Open in a separate window
Fig 1
Bifactor model with a general depression factor and three specific factors consisting of cognitive, affective, and somatic factors.
Table 1
Models of Beck Depression Inventory-II [BDI-II].
Model 1Model 2Model 3Model 4Model 5ItemsDC-ASS-ACCASNADS1XXXXX2XXXXX3XXXXX4XXXXX5XXXXX6XXXXX7XXXXX8XXXXX9XXXXX10XXXXX11XXXXX12XXXXX13XXXXX14XXXXX15XXXXX16XXXXX17XXXXX18XXXXX19XXXXX20XXXXX21XXXXX
Open in a separate window
Note: D: Depression; C-A: Cognitive-affective; S: Somatic; S-A: Somatic-Affective; C: Cognitive; NA: Negative Attitude; D: Difficulty
Table 2
Fit indices for all the specified models of Beck Depression Inventory-II [BDI-II].
χ2dfχ2/dfCFIGFITLIRMSEA
[90% CI]Akaike Information Criterion [AIC]Model 1: Unidimensional model968.61***1895.12.87.91.85.063 [.059-.067]1052.61Model 2: Cognitive-affective and somatic823.48***1884.38.89.93.88.057 [.053-.061]909.48Model 3: Somatic-affective and cognitive795.97***1884.23.90.93.88.056 [.052-.060]881.97Model 4: Cognitive, affective and somatic772.84***1864.15.90.93.88.055 [.051-.059]862.84Model 5: Negative attitude, difficulty and somatic792.94***1864.26.90.93.88.056 [.052-.060]882.94Model 6: Bifactor model [with cognitive-affective and somatic as specific factors]560.59***1683.33.93.95.92.047 [.043–0.52]686.59Model 7: Bifactor model [with cognitive and somatic-affective as specific factors]550.04***1683.27.94.95.92.047 [.042-.051]676.04Model 8: Bifactor model [with cognitive, somatic and affective as specific factors]541.57***1683.22.94.95.92.046 [.042-.051]667.57Model 9: Bifactor model [with negative attitude, difficulty and somatic as specific factors]561.28***1683.34.93.95.92.047 [.043-.052]687.29
Open in a separate window
Note: df: degree of freedom; CFI: comparative fit index; GFI: goodness of fit index; TLI: Tucker-Lewis index; RMSEA: root mean square error of approximation.
***p