South African Journal

South African Journal

South African Journal of Psychology 2016, Vol. 46(2) 203 –217 © The Author(s) 2015 Reprints and permissions: sagepub.co.uk/journalsPermissions.nav DOI: 10.1177/0081246315611016 sap.sagepub.com

Validation of the Beck Depression Inventory–II in South Africa: factorial validity and longitudinal measurement invariance in university students

M.S Makhubela1 and S Mashegoane2

Abstract This study evaluated the factorial validity of the Beck Depression Inventory–II with a diverse sample of 919 university students. A randomised split of the data was done, and then exploratory and confirmatory factor analyses were conducted on Group 1 (n = 460). Thereafter, confirmatory factor analysis was performed on Group 2 (n = 459) to cross-validate the determined factor structure. A lower-order factor structure that comprised three factors, namely, Negative Attitude, Performance Difficulty, and Somatic Complaints was found. A hierarchical second-order analysis indicated that the lower-order factors tap into a higher-order general factor of Depression. Results based on multigroup confirmatory factor analysis further indicated evidence of factorial invariance for this three lower-order factor structure across time. Evidence for convergent and discriminant validity were provided by predicted associations with subscales from the Hopkins Symptom Checklist–25. It is concluded that the Beck Depression Inventory–II is a reliable and valid measure that can be used to assess the severity of depressive symptoms over time among South African university students.

Keywords Construct validity, depression, factorial validity, latent mean, reliability

The psychometric properties of the Beck Depression Inventory–II (BDI-II; Beck, Steer, & Brown, 1996), a widely used measure for detecting depressive symptoms in adolescents and adults, are contentious. Studies have reported the factor structure of the BDI-II in different populations with

1Department of Psychology, University of Pretoria, South Africa 2Department of Psychology, University of Limpopo, South Africa

Corresponding author: M.S Makhubela, Department of Psychology, University of Pretoria, Pretoria 0002, South Africa. Email: [email protected]

611016 SAP0010.1177/0081246315611016South African Journal of PsychologyMakhubela and Mashegoane research-article2015

Article

at Apollo Group – UOP on August 11, 2016sap.sagepub.comDownloaded from

mailto:[email protected]
http://sap.sagepub.com/
204 South African Journal of Psychology 46(2)

varying consistency (e.g., Beck et al., 1996; Byrne, Stewart, Kennard, & Lee, 2007). While a num- ber of studies have reported two-factor solutions, encompassing cognitive and somatic symptoms, others have supported a three-factor model (Shafer, 2006). Beck et al. (1996) originally validated the BDI with outpatient and undergraduate participants, with both samples yielding two factors using exploratory factor analysis (EFA). The solutions were called the ‘Somatic-Affective’ (SA) and ‘Cognitive’ (C) model, and the ‘Cognitive-Affective’ (CA) and ‘Somatic’ (S) model (abbrevi- ated as SA-C and CA-S models, respectively). The models were subsequently confirmed with depressed geriatric inpatients, college students, and the general population in confirmatory factor analysis (CFA) studies (Vanheule, Desmet, Groenvynck, Rosseel, & Fontaine, 2008; Wu, 2010). Dozois, Dobson, and Ahnberg (1998) formulated another two-factor structure in a student sample with a CA (10 items) factor and a Somatic-Vegetative (SV) (11 items) factor.

However, several authors have also identified a three-factor solution with adolescent psychiatric outpatients and college students (i.e., Negative Attitude [NA], Performance Difficulty [PD], and Somatic Complaints [S]) (Steer, Kumar, Ranieri, & Beck, 1998; Vanheule et al., 2008). This is consistent with the already identified hierarchical structure, consisting of three first-order and one second-order factor structures of the BDI-II with high-school and college students (e.g., Byrne et al., 2007; Byrne, Stewart, & Lee, 2004; Whisman, Judd, Whiteford, & Gelhorn, 2013). Bos et al. (2009) reported three-factor structures in pregnancy and postpartum samples (namely, CA, Somatic-Anxiety [SAnx], and Fatigue [F]; CA, SA and Guilt [G]).

Not only have the BDI-II factor structure findings differed, but items representing the factors and the weight of item loadings within factors have also varied across studies. Several methodo- logical studies have explored the potential for a more complex factor structure of the BDI-II using CFA. Findings from these studies generally indicate that second-order (Byrne et al., 2007; Grothe et al., 2005) and general-factor structures (Ward, 2006) are better fits for the BDI-II than simple two- or three-factor structures. Ward’s (2006) model comprises a general (G) factor underlying the BDI-II, as well as C (8 items) and S (5 items) factors that are all orthogonal. Byrne and Baron (1993) also reported a four-factor model that assumed one higher-order factor of general Depression and three lower-order factors that represented NA, PD, and S, with Canadian adolescents. The validity of this structure of the BDI has subsequently been tested for Swedish (Byrne, Baron, Larsson, & Melin, 1995) and Bulgarian (Byrne, Baron, & Balev, 1998) adolescents.

The construct validity of the BDI-II, examined through the scale’s correlations with related external criteria, has also been an area of interest in the exploration of the validity of the instru- ment. According to theory and empirical research, scores on depression should correlate with risk factors and environmental concomitants and precipitants of depression (e.g., stressful life events). Cognitive theories of depression (Beck, Rush, Shaw, & Emery, 1979) suggest that core beliefs, such as the inclination to interpret events to support negative predictions (cognitive distortions) and to attribute negative events to stable causes (hopelessness), are central to the development of depressed mood. For this reason, it is anticipated that hopelessness and stressful life events will display strong, positive correlations with depression scores (Byrne et al., 2007).

Self-efficacy is an important protective variable for depression and correlates negatively with depression scores (Bandura, 1997). Self-efficacy has an indirect effect on depressive symptoms but only influences behaviours that decrease the likelihood of increased environmental stress (Bandura, 1997). Also, cross-cultural theory on depression suggests that, in collective cultures (e.g., South Africa), beliefs that accentuate internal sense of personal worth, efficacy, and control may be less significant than in individualistic cultures (Markus & Kitayama, 1994); therefore, they should be weakly protective against depressed mood.

Moreover, within a diathesis-stress framework, the vulnerability model suggests that negative self-evaluations (i.e., concomitants of low self-esteem; Beck et al., 1979) constitute a causal risk

at Apollo Group – UOP on August 11, 2016sap.sagepub.comDownloaded from

http://sap.sagepub.com/
Makhubela and Mashegoane 205

factor of depression (Stewart et al., 2004). For instance, according to Beck et al.’s (1979) cognitive theory of depression, negative self-beliefs are not just a symptom of depression but a diathesis wielding causal influence in the onset and maintenance of depression. The extant research provides unequivocal and consistent evidence for this relationship. Studies suggest that low self-esteem prospectively predicts depression (Orth, Robins, & Roberts, 2008).

Other subcategories of construct validity are convergent and discriminant validity. Essentially, the correlations between the BDI-II and discriminant measures are expected to be negative as opposed to correlations between the BDI-II and convergent measures (i.e., rs > .50). Several studies have reported significant positive correlations between the BDI-II, Depression Anxiety Stress Scales (DASS) depression scale, 90-item Symptom Checklist–Revised (SCL-90-R) depression subscale and the Zung Self-Rating Depression Scale, and Hopkins Symptom Checklist–25 (HSCL-25) depression subscale (e.g., Aasen, 2001; Al-Turkait & Ohaeri, 2010; Dozois et al.,1998; Lovibond & Lovibond, 1993). While others have reported significant negative correlations with the HSCL-25 anxiety subscale, Beck Anxiety Inventory (BAI), and the Hamilton Anxiety Rating Scale (Al-Turkait & Ohaeri, 2010; Beck et al., 1996; Osman, Barrios, Gutierrez, Williams, & Bailey, 2008).

Given the still contentious factorial and construct validity, and inconsistent assignment of items to the factors (e.g., Al-Turkait & Ohaeri, 2010; Wu, 2010), there is a need to investigate the factor structure and validity of the BDI-II in South Africa, especially among Africans. Besides, the majority of partici- pants in previous studies on the BDI-II were White, with no study including more than 15% African American or non-Western participants in their samples. Van de Vijver and Hambleton (1996) caution that the fact that a measuring instrument has demonstrated adequate validity in one culture does not necessarily mean that the same psychometric properties will prevail in another; such evidence needs to be empirically established. Although the BDI-II is a popular measure of the severity of depression in adolescents and adults among both researchers and clinicians in South Africa, its validity remains inde- terminate. The purpose of the current study, therefore, was to investigate the validity of the BDI-II for use among South African university students. Respectively, we (1) sought to establish whether there will be a theoretically justifiable factor structure of the BDI-II using both EFA and CFA, (2) whether the resulting model will remain invariant over time, and (3) examined whether there will be empirically and theoretically justifiable correlations between depression and related external criteria.

Method

Participants

Participants were a purposive longitudinal sample of 919 students (Mage = 21.70 years, standard devi- ation [SD] = 13.51 years) from both the University of Limpopo (n = 493) and the University of Pretoria (n = 425). The first is a predominantly Black institution and the latter is a historically White institution with a large component of Blacks in its student body. Besides accessibility, sampling from the institutions ensured that we had a heterogeneous sample (i.e., race and socio-economic status [SES]), which also approximated the ones used in previous studies on the BDI-II (Beck et al., 1996; Byrne et al., 2004). Only 304 (33.08%) of the total follow-up questionnaires distributed were returned. Of all the 919 participants, 579 (63.4%) selected the classification of ‘Black’, 291 (31.9%) said they were ‘White’, 26 (2.8%) were ‘Asian’, and 17 (1.9%) were ‘Coloured’.

Instruments

BDI-II. The BDI-II is a 21-item self-report Guttman scale (Beck et al., 1996) used to measure the severity of depressive symptoms in adolescents and adults. The examinee receives a score of 0–3

at Apollo Group – UOP on August 11, 2016sap.sagepub.comDownloaded from

http://sap.sagepub.com/
206 South African Journal of Psychology 46(2)

for each item; the total raw score is the sum of the 21 items, with the highest possible score of 63. Cronbach’s alpha was .84, and test–retest reliability was r = .55 in this study. Its correlations with clinical ratings and scales of depression from measures such as the Minnesota Multiphasic Person- ality Inventory (MMPI) are typically in the range of r = .60–.76 (Conoley, 1992).

HSCL-25. The HSCL-25 (Mollica, Wyshak, de Marneffe, Khuon, & Lavelle, 1987) is a widely used self-report screening measure derived from the 90-item Symptom Checklist (SCL-90). It consists of two subscales, namely, the 10-item HSCL-10 and the 15-item HSCL-15, measuring anxiety and depression symptoms, respectively. A full-scale score can also be computed. The measure is scored on a severity scale from ‘1’ (not at all) to ‘4’ (extremely), and has demonstrated its usefulness as a screening tool in various cross-cultural settings (Kaaya et al., 2002). Respond- ents scoring higher than a mean of 1.75 on the HSCL-25 (full scale), the HSCL-10 (anxiety), or the HSCL-15 (depression) are classified as having significant emotional distress (Mollica et al., 1987). The full-scale HSCL-25 has a demonstrated internal consistency of .95 (Coyne, 1994). The Cron- bach’s alpha of the HSCL-25 is in the upper .90s in South Africa (Halvorsen & Kagee, 2010). The full scale and the two subscales of the HSCL-10 and HSCL-15 displayed high internal consisten- cies of α = .89, .81, and .84 respectively, in this study.

Sherer General Self-Efficacy Scale. The Sherer General Self-Efficacy Scale (SGSES; Sherer et al., 1982) is a 17-item scale primarily developed for clinical and personality research. The response format is a 5-point scale (1 = strongly disagree, 5 = strongly agree). The higher the total score is, the more self-efficacious the respondent. Sherer et al. (1982) developed the SGSES scale to measure ‘a general set of expectations that the individual carries into new situations’ (p. 664). The SGSES has been the most widely used of the self-efficacy measures (Chen, Gully, & Eden, 2001).

Reviewing various studies, Chen et al. (2001) found the internal consistency reliabilities of the SGSES to range from moderate to high (α = .76–.89). While the scale’s reliability was compara- tively low at α = .53 in the present study, it was nevertheless acceptable (Kline, 2000). Moreover, removal of item 13 (Failure makes me try harder) because of its poor item-to-total correlation with the total scale would improve reliability to α = .60. This was not done to maintain the integrity of the scale. Besides, the results conducted with the two versions do not differ substantially. In two of their studies using samples of university students and managers, Chen et al. (2001) reported a high internal consistency reliability for the SGSES (α = .88–.91, respectively). With regard to the tem- poral stability of the scale, Chen et al. (2001) found high test–retest reliability estimates (r = .74 and .90). Research results show that SGSES negatively correlates with negative affect, anxiety, depres- sion, anger, and physical symptoms (e.g., Luszczynska, Gutiérrez-Donã, & Schwarzer, 2005).

Rosenberg Self-Esteem Scale. The Rosenberg Self-Esteem Scale (RSES) (Rosenberg, 1965) is a 10-item Likert-type scale that refers to self-respect and self-acceptance rated from 1 (totally disa- gree) to 4 (totally agree). Items 1, 3, 4, 7, and 10 are positively worded, while items 2, 5, 6, 8, and 9 are negatively worded. The RSES is the most used measure of global self-esteem (Hagborg, 1993). The RSES had a high internal reliability in previous studies (α = .92) (Hagborg, 1993) and also a moderate internal consistency at α = .73 in the present study.

Perceived Stress Scale–10-Item Version. The Perceived Stress Scale–10-Item Version (PSS-10; Cohen & Williamson, 1988) is a self-report instrument designed to assess the degree to which situations and circumstances in one’s life are appraised as stressful. It was designed to tap how unpredictable, uncontrollable, and overwhelming respondents find their lives. The PSS-10 requires participants to respond to a series of questions using a 5-point Likert scale (never = 0; almost never = 1;

at Apollo Group – UOP on August 11, 2016sap.sagepub.comDownloaded from

http://sap.sagepub.com/
Makhubela and Mashegoane 207

sometimes = 2; fairly often = 3; very often = 4), with higher scores reflecting greater stress levels. The PSS-10 has high coefficient alpha reliabilities (Cohen & Williamson, 1988). However, the scale displayed a moderate internal consistency in the present study (α = .55). Again, this value is acceptable for analysis (Kline, 2000), and removal of Item 5 (Things going your way) because of its low item-to-total correlation would have improved the reliability to α = .62, which was not done.

Beck Hopelessness Scale. The Beck Hopelessness Scale (BHS; Beck, Weissman, Lester, & Trexler, 1974) was designed to measure three major aspects of hopelessness: feelings about the future, loss of motivation, and expectations. The test is designed for adults aged 17–80 years, and consists of a list of 20 statements. The person is asked to decide about each sentence whether it describes his or her attitude for the last week, including the day in question. If the statement is false for him, he should write ‘false’ next to it. If the statement is true for him, he should label it ‘true’. There are seven reversed items: 1, 5, 6, 8, 13, 15, and 19. The BHS has high coefficient alpha reliabilities of .82–.93 (Beck et al., 1974), and displayed a good internal reliability in the present study (α = .69).

Procedure

Students were recruited from undergraduate classes at both the University of Limpopo and the University of Pretoria. The purpose of the research was first explained and then instructions on completing the questionnaire were specified. Once students consented in writing to participate, the instruments were administered to them in group set-ups. Follow-up data were collected after a 2-week time lag.

Ethical considerations

The study was approved by the research and ethics committees of the Universities of Limpopo and Pretoria, respectively. All participants consented (orally and in written form) to participation in the study. Participation was voluntary, while confidentiality and anonymity were assured.

Data analysis

Data analysis was conducted in a series of steps with the SPSS 22.0 and EQS 6.1 (Bentler, 2004) programmes. We first randomly split the data into two independent groups and then tested them using EFA and CFA. With respect to Group 1 (n = 460), the EFA was applied using principal com- ponent analysis with Varimax rotation (see Tabachnick & Fidell, 2007). In addition to parallel analysis (PA; Horn, 1965) and the minimum average partial methods (MAP; Velicer, 1976), factor selection was also determined by a set of standard criteria, including (1) the Kaiser–Guttman rule (i.e., factors with eigenvalues ≥ 1.0), (2) the scree plot, (3) cumulative and unique percent of explained variance, and (4) prior EFA findings. With the three-factor lower-order structure found to be most appropriate for Group 1, we then tested next for the validity of the lower-order factor structure for Group 1 (i.e., the same group) (n = 460) using CFA within the framework of structural equation modelling (SEM) based on maximum likelihood (ML) estimation.

Finally, using CFA once more, the best-fitting model for Group 1 was cross-validated with Group 2 data (n = 459). Criteria used to determine the goodness of fit to the data for the hypoth- esised structure included the Yuan–Bentler scaled Chi-square test (Y-B χ2), Satorra–Bentler scaled Chi-square test (S-B χ2), comparative fit index (CFI), Bentler–Bonett non-normed fit index (B-B NNFI), the standard root mean square residual (SRMR), and the root mean square error of approximation (RMSEA), along with its related 90% confidence interval (CI) (robust

at Apollo Group – UOP on August 11, 2016sap.sagepub.comDownloaded from

http://sap.sagepub.com/
208 South African Journal of Psychology 46(2)

versions of the fit indices were used since skewness and kurtosis analyses detected non-normal- ity in the data).

Moreover, based on the determined best-fitting model, stability of the lower-order factor load- ings was examined using analysis of covariance structures to test for their invariance across the two time points. Internal consistency reliability was also examined at two time points, for the total scale and for each of the three lower-order factors. Finally, associations between depression and theoreti- cally related external criteria were examined. As per theory and empirical research, depression scores were expected to be associated with risk factors and environmental precipitants.

During data collection, there was attrition over time (i.e., Time 1 to Time 2). Only 33% of the Time 2 questionnaires were returned. This kind of missing data is acceptable in longitudinal mod- elling (Wang & McArdle, 2008) because participants are said to contribute to estimation of the model at certain time points even if they do not do so at all time points (Singer & Willett, 2003). ML estimation is a common method for handling missing data in SEM and has demonstrated its advantages in many respects. It is less biased than traditional methods like listwise deletion, pair- wise deletion, and mean imputation in the application of CFA across both missing at random (MAR) and missing not at random (MNAR) mechanisms (Wothke, 2000). It also has benefits in estimation efficiency (i.e., variability in parameter estimate) (Arbuckle, 1996).

The measurement error introduced by attrition subsequently threatens not only the external validity of the study but the reliability of the measure(s) used. (The issue of reliability will be touched on in the discussion.) There are numerous reasons why attrition may occur in a study, including changes in personal attributes and attitudes of respondents. In the case of the present study, the novelty of the study and sheer curiosity are some of the reasons students may have par- ticipated. By the second administration, these could have waned. We are not able to delineate all the reasons for attrition, yet we can assess some of its effects. In instances where substantial attri- tion has occurred, it is best to compare the demographics of respondents who remained in the study and those who dropped out. Individuals who responded to both surveys in this current study did not differ significantly from attritors in terms of SES, education, gender, race, and age (e.g., remaining group: Mage = 21.66 years, SD = 3.48 years; attritors: Mage = 21.00 years; SD = 2.61 years). Similarity between the two groups on demographic characteristics means that to some extent, the remaining group still retains characteristics of the original sample, and there is a possibility that there is no distortion of parameter estimates.

Results

EFA and CFAs for the hypothesised model

EFA of the BDI-II. EFA performed on Group 1 (n = 460) data generated a three-factor solution (see Table 1). Eigenvalues were 3.25 (15.46% of variance), 2.75 (13.10% of variance), and 2.38 (11.33% of variance) for the first, second, and third factors, respectively. All items loaded signifi- cantly on three factors that could be appropriately labelled as NA, PD, and S (consistent with Byrne et al., 2004, 2007; Osman et al., 2008). All 21 items demonstrated acceptable factor loadings (≥.30) on a given factor following rotation, with the vast majority loading at .40 or higher.

Only one item cross-loaded on more than one factor: Item 1 (sadness) with loadings of .44 on Factor 1 (NA) and .42 on Factor 3 (S). The italicised factor loadings in Table 1 represent the item considered to cross-load on two factors. Thus, this item was considered not specific to any domain of depression in our sample. As noted by Ward (2006), in existing factor analytic studies of the BDI-II, the so-called cognitive or somatic factors also contain affective or emotional content and maybe this phenomenon accounts for the occurrence of these cross-loadings. Otherwise, all items

at Apollo Group – UOP on August 11, 2016sap.sagepub.comDownloaded from

http://sap.sagepub.com/
Makhubela and Mashegoane 209

loaded distinctively and without cross-loadings. Heeding Costello and Osborne’s (2005) recom- mendations on factor analysis, that after rotation and comparison of the item loading, the model with the ‘cleanest’ factor structure – item loadings above .30, few or no item cross-loadings, no factors with fewer than three items – has the best fit to the data, we considered our three-factor model to have the best fit for the student data.

CFA 1. The validity of the BDI-II structure derived empirically with the EFA and schematically portrayed in Figure 1 was tested with data from Group 1 (n = 460), the calibration sample. The results showed a well-fitting model to the data. Fit statistics for the test of this hypothesised model are shown in Table 2. The SEM testing was based on robust ML estimation.

The first step in SEM analysis was to test for the hypothesised model (i.e., a three-factor lower- order model generated from the data using the EFA). For the structural model, the significant χ2 value and its ratio to df is less than 1.5 (S-B χ2 = 275.477, df = 185, p < .0001, χ2/df=1.49), along with goodness-of-fit indices, namely, the B-B NNFI = .92 and CFI = .93, which were indicative of a well-fitting model (applying criteria used by Byrne [2006], and Dozois et al. [1998]), and an acceptable SRMR = .05 and RMSEA = .03 with a 90% CI for RMSEA of .02–.04, implied that the model fits the data well. The model for the study is presented in Figure 1.

All path coefficient estimates have the expected signs (information on parameter estimates is available from the first author). The magnitudes of the standardised path coefficient estimates, for the measurement components of the model, suggest that the items BDI-2, BDI-3, BDI-7, BDI-14, and BDI-17 have a stronger effect on Factor 1: NA than BDI-1, BDI-9, and BDI-11; and BDI-13, BDI-15, BDI-19, and BDI-20 have a stronger effect on Factor 2: PD than BDI-12 and BDI-4;

Table 1. Exploratory factor analysis three-factor solution.

Item descriptor Factor 1 – Negative attitude

Factor 2 – Performance difficulty

Factor 3 – Somatic complaints

1. Sadness .44 .42 2. Pessimism .73 3. Past failure .57 4. Loss of pleasure .48 5. Guilty feelings .43 6. Punishment feelings .49 7. Self-dislike .67 8. Self-criticalness .32 9. Suicidal thoughts .50 10. Crying .63 11. Agitation .42 12. Loss of interest .43 13. Indecisiveness .40 14. Worthlessness .62 15. Loss of energy .64 16. Changes in sleeping pattern .68 17. Irritability .56 18. Changes in appetite .53 19. Concentration difficulty .51 20. Fatigue .64 21. Loss of interest in sex .66

at Apollo Group – UOP on August 11, 2016sap.sagepub.comDownloaded from

http://sap.sagepub.com/
210 South African Journal of Psychology 46(2)

BDI-5, BDI-6, and BDI-8 have a stronger effect on Factor 3: S than BDI-10 and BDI-21. Moreover, the model also contains correlations between residual variances.

With respect to the latent factor components of the model (Figure 1), the results show that all the three factors have statistically significant associations between them (NA and PD are associ- ated at r = .76, p < .001; NA and S are associated at r = .78, p < .001; and PD and S are associated at r = .83, p < .001). The high correlation between the latent factors is suggestive of the presence of a higher/second-order general factor (‘Depression’) (Byrne et al., 1995).

CFA 2. Testing of the hypothesised model for Group 2 (i.e., the validation sample) also generated a well-fitting model, and all parameters were viable and statistically significant (see Table 2). This suggests that the hypothesised lower-order model represented the data adequately.

Figure 1. Hypothesised first-order three-factor structure.

at Apollo Group – UOP on August 11, 2016sap.sagepub.comDownloaded from

http://sap.sagepub.com/
Makhubela and Mashegoane 211

Higher-order CFA. Psychometric theory recommends that second-order factors should be extracted whenever factors are correlated (Gorsuch, 1983). In our case, the correlations between the latent factors (i.e., r = .76, .78 and .83, respectively) are indicative of a higher level conceptualisation, which should correspond to a second-order general factor of Depression. As such, a hypothesised a priori hierarchical model (see Figure 2) was tested for factorial validity with the calibration sample. The goodness-of-fit statistics of the second-order model show that the model fits the data well, CFI = .91 and RMSEA = .05 with a 90% CI for RMSEA of .04–.05. Accordingly, we can conclude that the second-order model is the most optimal representative BDI-II structure for non-clinical South African students.

Longitudinal measurement invariance of the hypothesised three-factor model

The stability of hierarchically structured factor loadings was tested across a time lag of 2 weeks. The present study is based on the analysis of means and covariance structures (MACS), which allowed us to address the issue of scalar equivalence in testing for differences in the means of the lower order factors.

The overall fit of the lower-order three-factor model at Time 1 and again at Time 2, with no equality constraints imposed, was first tested. The fit of this model (Table 3) was indicative of an adequately fitting model that included two time points of data. Following this, a test for the invari- ance of the lower-order factor loadings was conducted. As such, the three-factor model (Time 1 and Time 2) was again estimated, but this time with equality constraints placed on all lower-order fac- tor loadings across Time 1 and Time 2.

An overall adequate fit of .94 and no deterioration in the overall fit (i.e., when we assume equal variance in the latent factors and intercepts) between models 1, 2, 3, and 4 (Δ*CFI = .00 and Δ*RMSEA = .00) suggested that the lower order factor loadings were invariant across time (Cheung & Rensvold, 2002). In Model 3, extremely stringent assessment of invariance, we tested for the equality of intercepts across time; this test measures scalar invariance of the BDI-II. Once again, results yielded a Δ*CFI of .00 and Δ*RMSEA of .00, thereby providing credible evidence of invar- iance across time.

Internal consistency reliability for the total scale and the three lower-order factors

Internal consistency reliability coefficients, as computed for Cronbach’s coefficient alpha, are reported in Table 4 for both Time 1 and Time 2. Internal consistency of the total scale score for overall depression was high at Time 1 (α = .84), and even slightly higher at Time 2 (α = .90). Although internal consistency for the S subscale was somewhat weaker than for the NA and PD subscales, it nonetheless exhibited adequate reliability. Byrne et al. (2004) explain the occurrence

Table 2. Three-factor model of the BDI-II structure: goodness-of-fit statistics.

Group df B-B NNFI

S-B χ2 *CFI *RMSEA 90% *RMSEA CI

SRMR

Calibrationa 185 .92 275.48 .93 .03 .02–.04 .05 Validationb 390 .93 546.89 .93 .03 .02–.04 .05

BDI-II: Beck Depression Inventory–II; df: degrees of freedom; B-B NNFI: Bentler–Bonett non-normed fit index; S-B χ2: Satorra–Bentler scaled Chi-square test; *CFI: robust comparative fit index; *RMSEA: robust root mean square error of approximation and its 90% confidence interval; SRMR: standardised root mean square residual. an = 460. bn = 459.

at Apollo Group – UOP on August 11, 2016sap.sagepub.comDownloaded from

http://sap.sagepub.com/
212 South African Journal of Psychology 46(2)

of this comparatively lower alpha coefficient for the S factor as not necessarily an anomaly, but due to the general relative rarity of somatic symptoms in the normal population.

Correlations between depression and theoretically linked external criteria

Convergent and discriminant validity. The correlation between the BDI-II and HSCL-15 (depression) was expected to yield convergent validity, and discriminant validity with HSCL-10 (anxiety). Cor- relation analysis shows that the BDI-II is statistically significantly and positively associated with both the HSCL-15 (depression) (r = .67, p < .01) and the HSCL-10 (anxiety) (r = .50, p < .01).

Construct validity. Correlation analysis shows that depression correlates statistically signifi- cantly and positively with hopelessness (r = .44, p < .01), perceived stress (r = .28, p < .01), and

Figure 2. Hypothesised hierarchical second-order model.

at Apollo Group – UOP on August 11, 2016sap.sagepub.comDownloaded from

http://sap.sagepub.com/
Makhubela and Mashegoane 213

self-efficacy (r = .39, p < .01), and statistically significantly but negatively with self-esteem (r = −.51, p < .01).

Discussion

The present study validated the use of the BDI-II (Beck et al., 1996) with South African university students. With that goal, the researchers identified, tested, and cross-validated the factor structure of the BDI-II for two sub-samples of randomly split data using EFA and CFA. As with past studies of the BDI with non-clinical university students (e.g., Byrne et al., 2004; Osman et al., 2008), our results revealed that the data for South African university students is best represented by a three lower-order structure consisting of NA, PD, and S factors with an additional overarching second- order general factor of Depression. The hypothesised hierarchical model was adequately well-fit- ting and had no evidence of potential misspecification.

With the evidence of a robust and parsimonious factor structure, we continued next to assess the BDI-II’s internal consistency reliability and its stability across two time points. Findings of the total score’s internal consistency from the current study were consistent with those previously reported (Dozois et al., 1998). All values were adequate, with the comparatively weak findings

Table 3. Goodness-of-fit and comparative statistics for tests for invariance of BDI-II hierarchical structure across time.

Model and constraints

Y-B χ2 df *CFI SRMR *RMSEA 90% *RMSEA CI

Model comparison

Δ*CFI Δ*RMSEA

1. Configural invariance 550.25 365 .94 .04 .03 .02–.03 2. Lower order factor

loadings invariant 567.23 382 .94 .05 .03 .02–.03 2 vs 1 .00 .00

3. Intercepts invariant 625.11 399 .94 .06 .03 .03–.04 3 vs 1 .00 .00 4. Latent factor means

invariant 587.28 396 .94 .05 .03 .02–.03 4 vs 1 .00 .00

BDI-II: Beck Depression Inventory–II; df: degrees of freedom; Y-B χ2: Yuan–Bentler scaled Chi-square test; *CFI: robust comparative fit index; *RMSEA: robust root mean square error of approximation and its 90% confidence interval; SRMR: standardised root mean square residual; Δ*CFI: comparative fit index difference value; Δ*RMSEA: robust root mean square error of approximation difference value. p < .001.

Table 4. Internal consistency of the BDI-II across time.

BDI-II Subscales Internal consistency coefficients

Time 1 Time 2

Negative attitude (NA) .73 .82 Performance difficulty (PD) .70 .78 Somatic complaints (S) .62 .75 Depression (total score) .84 .90

BDI-II: Beck Depression Inventory–II. Time 1 N = 919 and Time 2 N = 304. The average performance of the participants on the total BDI-II (M = 11.45; SD = 7.74), NA (M = 2.83; SD = 2.97), PD (M = 5.15; SD = 3.21) and S (M = 3.48; SD = 3.05) is outside the symptomatic range of the BDI-II suggested by Beck et al. (1996).

at Apollo Group – UOP on August 11, 2016sap.sagepub.comDownloaded from

http://sap.sagepub.com/
214 South African Journal of Psychology 46(2)

being related to the S factor. Nevertheless, the fact that somatic symptoms measured by the 6 items comprising this scale are seldom found in the normal population probably explains its rather low reliability.

Similarly, the global depression score stability was considered to be satisfactory, particularly that there is a possibility of extraneous factors such as life events and stress intervening over the 2 weeks interval between the test–retest administration of the BDI-II. Even with that, some stability would still be probable, as people are likely to possess enduring trait-like dispositions in their mood responsiveness (Byrne et al., 2004). Our finding is congruent to Beck, Steer, and Garbin’s (1988) report that test–retest correlations of the BDI ranged between .40 and .86 across retest peri- ods of 1 week to 2–3 months. While this stability of the BDI-II across time is relatively lower than in some findings (e.g., Aasen, 2001; Byrne et al., 2004), it is nevertheless consistent with others (Kuhner, Burger, Keller, & Hautzinger, 2007; Yin & Fan, 2000).

It was also considered prudent to examine correlations between depression and theoretically linked external criteria. Based on theory and empirical research, scores on depression should correlate with risk factors as well as with environmental precipitants (Byrne et al., 2004). As indicated earlier, cor- relations between depression and hopelessness, and perceived stress would be expected to be high. Inversely, its correlations with self-efficacy and self-esteem would be expected to be lower. True to prediction, depression correlated most strongly with hopelessness and perceived stress, and less so, albeit still significantly, with self-esteem. Cross-sectional and longitudinal studies suggest that low self-esteem prospectively predicts depression (Orth et al., 2008). These results were consistent with those of past studies that demonstrated congruity between the correlates of depression reported in Western and Asian studies and those found for African students (Peltzer, Pengpid, Olowu, & Olasupo, 2013; Stewart, Lam, Betson, & Chung, 1999).

Contrary to Western literature in support of self-efficacy as an important protective factor for depression (Bandura, 1997), it correlated strongly and positively with depression in the present study. This finding is consistent with cross-cultural theory that posits that, in allocentric cultures (e.g., African), beliefs that embrace internal sense of personal worth and efficacy may be less sig- nificant than in idiocentric cultures (Markus & Kitayama, 1994). True to expectation, depression as measured by the BDI-II was also strongly and positively correlated with depression as measured by HSCL-15, providing evidence for convergent validity. As discussed elsewhere, there are contra- dictions in the literature on the relationship between depression and anxiety (see Aasen, 2001). However, the present study found a strong and positive correlation between depression and anxi- ety. This result is consistent with the considerable evidence on the comorbidity of the two disorders (Joiner, 1996).

With respect to factorial invariance, evidence of measurement invariance (MI) in the context of the lower-order three-factor structure of the BDI-II across a 2-week time lag was established. MI was established at the level of configural, metric, and scalar invariance. Specifically, across models in which there was increasingly restricted parameterisation on the variance/covariance matrices of the indicators, there was consistent evidence that the three-factor structure provided robust fit with the data. Furthermore, the Δ*CFI and Δ*RMSEA values for comparisons between models 1 (con- figural model), 2, 3, and 4 were all negligible. These findings were consistent with those of earlier studies that demonstrated the stability of the depression scores on the BDI-II over time (e.g., Byrne et al., 2004).

Conclusion

This study was the first to establish the factor structure of the BDI-II among students in South African universities, and going on to assess its invariance over time. A four-factor hierarchical

at Apollo Group – UOP on August 11, 2016sap.sagepub.comDownloaded from

http://sap.sagepub.com/
Makhubela and Mashegoane 215

structure comprising one second-order general factor of Depression and three first-order factors representing NA, PD, and S, commonly found among non-clinical students, was confirmed. This study further demonstrated that the BDI-II yields reliable, internally consistent, and valid (i.e., convergent and discriminate validity) scores for South African university students. Evidence of longitudinal invariance was provided for, indicating that the scale is potentially useful as a measure of depression over time, especially in clinical settings where the instrument is used also for moni- toring of treatment progress.

The study has some limitations which may affect the usefulness of the results. Only students were used, constraining the generalisability of the findings. Future research must extend to non- students and clinical populations, and further develop cut-off scores for the BDI-II among Africans (especially for use in clinical settings). Also, the variable of ethnicity was available, but was not used during analysis because of space limitations. As already suggested, future research must take it into account given that it may also affect the results substantially.

Acknowledgements

The article is an extension of a paper presented by M.S Makhubela and S. Mashegoane as ‘Validity of the Beck Depression Inventory in South Africa’ at the 20th Annual South African Psychology Congress, 16–19 September 2014, in Durban, KwaZulu-Natal.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

References

Aasen, H. (2001). An empirical investigation of depression symptoms: Norms, psychometric characteristics and factor structure of the Beck Depression Inventory-II (Unpublished MA dissertation). University of Bergen, Bergen, Norway.

Al-Turkait, F. A., & Ohaeri, J. U. (2010). Dimensional and hierarchical models of depression using the Beck Depression Inventory-II in an Arab college student sample. BMC Psychiatry, 10, 1471–2440.

Arbuckle, J. L. (1996). Full information estimation in the presence of incomplete data. In G. A. Marcoulides & R. E. Schumacker (Eds.), Advanced structural equation modeling: Issues and techniques (pp. 243–277). Mahwah, NJ: Lawrence Erlbaum.

Bandura, A. (1997). Self-efficacy: The exercise of control. New York, NY: Freeman. Beck, A. T., Rush, A. J., Shaw, B. F., & Emery, G. (1979). Cognitive therapy of depression. New York, NY:

Guilford Press. Beck, A. T., Steer, R. A., & Brown, G. K. (1996). Beck Depression Inventory manual (2nd ed.). San Antonio,

TX: Psychological Corporation. Beck, A. T., Steer, R. A., & Garbin, M. G. (1988). Psychometric properties of the Beck Depression Inventory:

Twenty-five years of evaluation. Clinical Psychology Review, 8, 77–100. Beck, A. T., Weissman, A., Lester, D., & Trexler, L. (1974). The measurement of pessimism: The hopeless-

ness scale. Journal of Consulting & Clinical Psychology, 42, 861–865. Bentler, P. M. (2004). EQS 6.1: Structural equations program manual. Encino, CA: Multivariate Software,

Inc. Bos, S. C., Pereira, A. T., Marques, M., Maia, B., Soares, M. J., Valente, J., . . .Azevedo, M. H. (2009). The

BDI-II factor structure in pregnancy and postpartum: Two or three factors? European Psychiatry, 24, 334–340.

Byrne, B. M. (2006). Structural equation modeling with EQS: Basic concepts, applications, and program- ming (2nd ed.). Mahwah, NJ: Lawrence Erlbaum.

Byrne, B. M., & Baron, P. (1993). The Beck Depression Inventory: Testing and cross-validating a hierarchi- cal structure for nonclinical adolescents. Measurement & Evaluation in Counseling & Development, 26, 164–178.

at Apollo Group – UOP on August 11, 2016sap.sagepub.comDownloaded from

http://sap.sagepub.com/
216 South African Journal of Psychology 46(2)

Byrne, B. M., Baron, P., & Balev, J. (1998). The Beck Depression Inventory: A cross-validated test of facto- rial structure for Bulgarian adolescents. Educational & Psychological Measurement, 58, 241–251.

Byrne, B. M., Baron, P., Larsson, B., & Melin, L. (1995). The Beck Depression Inventory: Testing and cross- validating a second-order factorial structure for Swedish nonclinical adolescents. Behaviour Research & Therapy, 33, 345–356.

Byrne, B. M., Stewart, S. W., Kennard, B. D., & Lee, P. W. H. (2007). The Beck Depression Inventory II: Testing for measurement equivalence and factor mean differences across Hong Kong and American adolescents. International Journal of Testing, 7, 293–309.

Byrne, B. M., Stewart, S. M., & Lee, P. W. H. (2004). Validating the Beck Depression Inventory–II for Hong Kong community adolescents. International Journal of Testing, 4, 199–216.

Chen, G., Gully, S. M., & Eden, D. (2001). Validation of a new General Self-Efficacy Scale. Organizational Research Methods, 4(1), 62–68.

Cheung, G. W., & Rensvold, R. B. (2002). Evaluating goodness-of-fit indexes for testing measurement invari- ance. Structural Equation Modeling, 9, 233–255.

Cohen, S., & Williamson, G. (1988). Perceived stress in a probability sample of the United States. In S. Spacapan & S. Oskamp (Eds.), The social psychology of health: Claremont symposium on applied social psychology (pp. 31–67). Newbury Park, CA: SAGE.

Conoley, C. W. (1992). Review of the Beck Depression Inventory [Revised Edition]. In J. J. Kramer & J. C. Conoley (Eds.), Eleventh mental measurements yearbook (pp. 78–79). Lincoln, NE: Buros Institute of Mental Measurements.

Costello, A. B., & Osborne, J. W. (2005). Best practices in exploratory factor analysis: Four recommendations for getting the most from your analysis. Practical Assessment, Research & Evaluation, 10, 1–9.

Coyne, J. C. (1994). Self-reported distress: Analog or ersatz depression? Psychological Bulletin, 116, 29–45. Dozois, D. J. A., Dobson, K. S., & Ahnberg, J. L. (1998). A psychometric evaluation of the Beck Depression

Inventory–II. Psychological Assessment, 10, 83–89. Gorsuch, R. L. (1983). Factor analysis (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum. Grothe, K. B., Dutton, G. R., Jones, G. N., Bodenlos, J., Ancona, M., & Brantley, P. J. (2005). Validation

of the Beck Depression Inventory-II in a low-income African American sample of medical outpatients. Psychological Assessment, 17(1), 110–114.

Hagborg, W. J. (1993). The Rosenberg Self-Esteem Scale and Harter’s Self-Perception Profile for adoles- cents: A concurrent validity study. Psychology in the Schools, 30, 132–136.

Halvorsen, J. O., & Kagee, A. (2010). Predictors of psychological sequelae of torture among South African former political prisoners. Journal of Interpersonal Violence, 25, 989–1005.

Horn, J. L. (1965). A rationale and test for the number of factors in factor analysis. Psychometrika, 30, 179–185.

Joiner, T. E. (1996). A confirmatory factor-analytic investigation of the tripartite model of depression and anxiety in college students. Cognitive Therapy & Research, 20, 521–539.

Kaaya, S. F., Fawzi, M. C., Mbwambo, J. K., Lee, B., Msamanga, G. I., & Fawzi, W. (2002). Validity of the Hopkins Symptom Checklist-25 amongst HIV-positive pregnant women in Tanzania. Acta Psychiatrica Scandinavica, 106(1), 9–19.

Kline, P. (2000). The handbook of psychological testing (2nd ed.). London, England: Routledge. Kuhner, C., Burger, C., Keller, F., & Hautzinger, M. (2007). Reliabilität und Validität des revidierten Beck-

Depressions-inventars (BDI-II). Befunde aus deutschsprachigen Stichproben. Nervenarzt, 78, 651–656. Lovibond, S. H., & Lovibond, P. F. (1993). Manual for the Depression Anxiety Stress Scales (DASS)

(Psychology Foundation Monograph). Sydney, New South Wales, Australia: The Psychology Foundation. Luszczynska, A., Gutiérrez-Donã, B., & Schwarzer, R. (2005). General self-efficacy in various domains of

human functioning: Evidence from five countries. International Journal of Psychology, 40(2), 80–89. Markus, H. R., & Kitayama, S. (1994). The cultural construction of self and emotion: Implications for social

behavior. In S. Kitayama & H. R. Markus (Eds.), Emotion and culture: Empirical studies of mutual influ- ence (pp. 89–130). Washington, DC: American Psychological Association.

Mollica, R. F., Wyshak, G., de Marneffe, D., Khuon, F., & Lavelle, J. (1987). Indochinese versions of the Hopkins Symptom Checklist-25: A screening instrument for the psychiatric care of refugees. American Journal of Psychiatry, 144, 497–500.

at Apollo Group – UOP on August 11, 2016sap.sagepub.comDownloaded from

http://sap.sagepub.com/
Makhubela and Mashegoane 217

Orth, U., Robins, R. W., & Roberts, B. W. (2008). Low self-esteem prospectively predicts depression in ado- lescence and young adulthood. Journal of Personality & Social Psychology, 95, 695–708.

Osman, A., Barrios, F., Gutierrez, P. M., Williams, J. E., & Bailey, J. (2008). Psychometric properties of the Beck Depression Inventory–II in nonclinical adolescent samples. Journal of Clinical Psychology, 64, 83–102.

Peltzer, K., Pengpid, S., Olowu, S., & Olasupo, M. (2013). Depression and associated factors among univer- sity students in western Nigeria. Journal of Psychology in Africa, 23, 459–466.

Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton University. Shafer, A. B. (2006). Meta-analysis of the factor structures of four depression questionnaires: Beck, CES-D,

Hamilton and Zung. Journal of Clinical Psychology, 62, 123–146. Sherer, M., Maddux, J. E., Mercandante, B., Prentice-Dunn, S., Jacobs, B., & Rogers, R. W. (1982). The Self-

Efficacy Scale: Construction and validation. Psychological Reports, 51, 663–671. Singer, J. D., & Willett, J. B. (2003). Applied longitudinal data analysis: Modeling change and event occur-

rence. New York, NY: Oxford University Press. Steer, R. A., Kumar, G., Ranieri, W., & Beck, A. T. (1998). Use of the Beck Depression Inventory–II with

adolescent psychiatric outpatients. Journal of Psychopathology & Behavioral Assessment, 20, 127–137. Stewart, S. M., Kennard, B. D., Lee, P. W. H., Hughes, C. W., Mayes, T., Emslie, G., & Lewinsohn, P.

(2004). A cross-cultural investigation of cognitions and depressive symptoms in adolescents. Journal of Abnormal Psychology, 113, 248–257.

Stewart, S. M., Lam, T. H., Betson, C., & Chung, S. F. (1999). Suicide ideation and its relationship to depressed mood in a community sample of adolescents in Hong Kong. Suicide & Life-Threatening Behavior, 29, 227–240.

Tabachnick, B. G., & Fidell, L. S. (2007). Using multivariate statistics (5th ed.). Upper Saddle River, NJ: Pearson – Allyn & Bacon.

Van de Vijver, F., & Hambleton, R. K. (1996). Translating tests: Some practical guidelines. European Psychologist, 1, 89–99.

Vanheule, S., Desmet, M., Groenvynck, H., Rosseel, Y., & Fontaine, J. (2008). The factor structure of the Beck Depression Inventory-II: An evaluation. Assessment, 15, 177–187.

Velicer, W. F. (1976). Determining the number of components from the matrix of partial correlations. Psychometrika, 41, 321–327.

Wang, L., & McArdle, J. J. (2008). A simulation study comparison of Bayesian estimation with conventional methods for estimating unknown change points. Structural Equation Modeling: A Multidisciplinary Journal, 15(1), 52–74.

Ward, L. C. (2006). Comparison of factor structure models for the Beck Depression Inventory-II. Psychological Assessment, 18, 81–88.

Whisman, M. A., Judd, C. M., Whiteford, N. T., & Gelhorn, H. L. (2013). Measurement invariance of the Beck Depression Inventory-Second Edition (BDI-II) across gender, race, and ethnicity in college stu- dents. Assessment, 20, 419–428.

Wothke, W. (2000). Longitudinal and multigroup modeling with missing data. In T. D. Little, K. U. Schnabel, & J. Baumert (Eds.), Modeling longitudinal and multiple group data: Practical issues, applied approaches and specific examples (pp. 219–240). Mahwah, NJ: Lawrence Erlbaum.

Wu, P.-C. (2010). Measurement invariance and latent mean differences of the Beck Depression Inventory II across gender groups. Journal of Psychoeducational Assessment, 28, 551–563.

Yin, P., & Fan, X. (2000). Assessing the reliability of Beck Depression Inventory scores: Reliability generali- zation. Educational & Psychological Measurement, 60, 201–223.

at Apollo Group – UOP on August 11, 2016sap.sagepub.comDownloaded from

http://sap.sagepub.com/
Blog ArchiveCopyright © 2019 HomeworkMarket.com Read More
Applied SciencesArchitecture and DesignBiologyBusiness & FinanceChemistryComputer ScienceGeographyGeologyEducationEngineeringEnglishEnvironmental scienceSpanishGovernmentHistoryHuman Resource ManagementInformation SystemsLawLiteratureMathematicsNursingPhysicsPolitical SciencePsychologyReadingScienceSocial Science