Factorial structure of the Beck Depression Inventory for depression in university students

Objective: To explore the dimensionality of the Beck Depression Inventory (BDI) among Colombian college students. Methods: A validation study was designed, involving the participation of a sample of 786 health science students (medicine, nursing, and psychology) aged between 18 and 27 ( M = 20.0, SD = 1.9). The participants completed the 21-item BDI. Internal consistency was calculated (Cronbach’s alpha and McDonald’s omega) and dimensionality was demonstrated using factorial confirmatory analysis (CFA). Results : The Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy was high (0.898) and the Bartlett’s sphericity test gave excellent results (chi-square = 3,102.60; df = 210; p < 0.001). One-, two- and three-dimensional models were used. The unidimensional model performed best, representing 24.8% of the total variance, high internal consistency, a Cronbach’s alpha of 0.83 and a McDonald’s omega of 0.84. However, the CFA did not fit adequately (chi-square = 583.79; df = 189; p < 0.001, RMSEA = 0.052, CI 90% 0.047-0.056, CFI = 0.87, TLI = 0.85 and SMSR = 0.04). Conclusions : The best factor solution for the BDI is given by the unidimensional model, which presents high internal consistency. However, its adjustment in the CFA is not acceptable.


INTRODUCTION
Major depressive disorder is a global public health problem that leads to great impairments in school, work, family and social functioning and years of healthy life lost 1,2 . As such, it is important to be able to rely on the availability of valid and reliable instruments that allow us to screen depressive symptoms with possible clinical importance in different contexts 3 .
Currently, there are many screening instruments for major depressive episodes in different contexts for clinical or research purposes 4 . One of the most commonly used surveys is the Beck Depression Inventory (BDI) 5 , which has different versions with different numbers of items that quantify the cognitive symptoms of major depressive episodes [6][7][8] .
It is a well-known fact that these variations can present themselves in the psychometric performance of instruments such as the BDI, especially so in the response pattern that determines the dimensionality of the measurement scale 19 . As well as its theoretical implications, this also implies a need to interpret the results according to the characteristics of the people responding to the instruments 20 .
The purpose of this study was to test the dimensionality (AFC) of the BDI in a sample of health sciences students in Santa Marta, Colombia.

Design
A psychometric study was implemented to evaluate the performance of a construct quantification scale. Such studies are also known as instrumental methodological studies or evaluation screening or diagnostic tests according to the area of knowledge 21 . The study received the approval of the research ethics committee. Despite not presenting any risks according to the Ministry of Health Resolution 8,430 of 1993, all the participants signed the informed consent form. Confidentiality was guaranteed throughout the application and analysis of the socio-demographic data and findings 22 .

Population
A probabilistic sampling of health sciences students in Santa Marta, Colombia, was carried out in different phases. Probabilistic sampling was used given that this study is a secondary analysis of a cross-sectional research in which a number of different scales are applied. The sample was made up of a total of 706 students: 186 (23.7%) from nursing, 275 (35.0%) from medicine, and 325 (41.3%) from psychology. The stu-dents' ages ranged between 18 and 37 (M = 20.0, DE = 1.9). There was a participation rate of 616 women (78.4%) and 170 men (21.7%). The number of participants was sufficient for the calculation of internal consistency and to carry out a confirmatory factor analysis (CFA), which requires a minimum number of 400 participants 20 .

Instruments
The students completed the 21-item BDI. The items were originally qualified in two dimensions. The first (cognitive) was made up of the first 14 sections, and the second (somatic) was made up of the remaining seven sections. Each of the items offers four answer options ranging from "never" to "almost always", which are then qualified from 1 to 4. The higher the score, the higher the possibility of having presented a major depressive episode in the past two weeks 5 .

Procedure
The students completed the instrument in the classroom, in a group application. The objectives of the research were explained, as were the ethical considerations, voluntary participation, and the fact that the exercise would not be compensated with any kind of incentive beyond the usefulness of the findings for science and for knowledge generation.

Statistical analysis
Confirmatory factorial analyses were carried out using the maximum likelihood method. The analyses were carried out for the two dimensions proposed originally and for one and three dimensions, as suggested by more recent research 16,17 . The typical coefficients for the beginning of the factorial analysis were used, along with Bartlett's sphericity coefficient 23 and the KMO index 24 .
In the CFA, we determined the Satorra-Bentler chi square test, with degrees of freedom (DF) and probability value (p), the RMSEA coefficients (Root Mean Square Error of Approximation), and a confidence interval of 90% (CI 90%), CFI (Comparative Fit Index), TLI (Tucker-Lewis Index) and SMSR (Standardized Mean Square Residual). For the chi squared, we expected the probability value to be above 5%; for RMSEA and SRMR, below 0.06; and for CFI and TLI, values below 0.89 were expected.
Cronbach's alpha 25 and McDonald's omega 26 were calculated to find out the internal consistency according to the conceptualized dimensions. The McDonald test is more precise in estimating the internal consistency when the equivalence principle is not fulfilled 26 . Data analysis was carried out using STATA for Windows 27 .

RESULTS
Initially, we examined the indicators to determine CFA pertinence. The analysis indicated sampling adequacy through

DISCUSSION
This study demonstrates that the factorial solutions for one-, two-and three-dimensional BDI do not adjust adequately to the sample of health sciences students in Santa Marta, Colombia.
We can see that the BDI presented a Cronbach's alpha of 0.84 when measured across the 21 items. This observation is consistent with previous research that has shown values within the desired range which falls between 0.70 and 0.95 [9][10][11][12][13][14][15] . This approach is correct if we consider a BDI with a unidimensional scale 28 .
The original proposal of the BDI was for a bi-dimensional scale. In this study, the internal consistency for the first dimension was of 0.79 and for the second, it was 0.61. This finding is inconsistent with research using other populations that showed internal consistency values within the desired range for both dimensions [9][10][11][12][13][14][15] . It is evident that this disparity found in terms of the values of the coefficients has practical implications for the acceptance of the dimensionality of the scale 19,28 . We recommend the use of internal consistency only for one-dimensional scales. The calculation should be carried out separately for each dimension in bi-or multidimensional scales 28 . Furthermore, it is highly likely that a consistency of above 0.80 for the 21 items as a set may be given directly by the number of items and not the high correlation between them 19,28 . It is known that this coefficient is very sensitive to the number of items and, as such, the calculation is not recommended for a set of over 15 items for the more conservative, or up to 20 items for the more liberal. The reason is simple: as from 15 items, the internal consistency increases rapidly and tends to steer away from the real value 19 .
For the CFA, this analysis shows that in none of the three factorial solutions do the five goodness of fit models adjust to the data, with high chi squared and CFI, and TLI of lower than 0.90. Other research has shown that not all goodness of fit coefficients were adequate for the BDI 16,17 . However, it was concluded that the solution for one, two or three factors was the most promising [12][13][14][15][16][17][18] . These divergences in the conclusions are caused for many different aspects. The first is that there is no absolute agreement for the quantitative and qualitative interpretation of the factorial solutions. The second is that there is an evident lack of consensus in terms of the interpretation of the goodness of fit coefficients 19 . Finally, the third is that there is notable variability of the factorial solutions according to the characteristics of the population for those scales with more than 15 items 19,28 .
These findings encourage a consideration of the current limitations of factorial analyses and, as such, the factorial structure of the BDI-21. The weaknesses of this approach have led to the reduction of the number of items in the instrument and, as a result, currently, 10 or less item scales which measure the essential part of the construct and that show greater stability and better performance indicators are preferred 20,29 . A 7-item version is now available for the BDI, which has one dimension confirmed in CFA, achieving very good fit indices (RMSEA = 0.058, and both CFI and TLI = 0.99) 30 . The findings encourage us to consider the reduction of the number of items in the BDI-21 for this population 29 .
This study's strength is that it involved a large sample of participants chosen at random and that it considered strict interpretation criteria for the indicators, in particular, the goodness of fit indicators. However, the study was limited in that the number of students in the possible segments did not allow for a reliable analysis.

CONCLUSIONS
We conclude that for the BDI, the one-, two-and three-dimensional factorial solutions do not adequately adjust to the health sciences students in Santa Marta, Colombia. Care should be taken in the interpretation of BDI results for this population. This performance needs to be corroborated in another group of university students.

INDIVIDUAL CONTRIBUTIONS
Adalberto Campo-Arias -Contributed substantially to design, analysis and interpretation of data, make substantially contributed to and the drafting of the article. He also gave final approval of the version to be published. Yuly Suárez-Colorado -Contributed substantially to conception, design and interpretation of data, and to the intellectual content. She also gave final approval of the version to be published. Carmen C. Caballero-Dominguez -Contributed substantially to conception, design, and interpretation of data, and critically reviewed the paper for important intellectual content. She also gave final approval of the version to be published.

CONFLICTS OF INTEREST
Drs. Adalberto Campo-Arias, Yuly Suárez-Colorado and Carmen C Caballero-Domínguez have no conflicts of interest to declare.