Factor Structure and Psychometric Properties of the Spanish Version of the "Dysfunctional Attitude Scale-Revised"

Full reference: Ruiz, F. J., Suárez-Falcón, J. C., Odriozola-González, P., Barbero-Rubio, A., López-López, J. C., Eisenbeck, N., Budziszewska, L., & Gil, E. (2015). Factor structure and psychometric properties of the Spanish version of the "Dysfunctional Attitude Scale-Revised". Behavioral Psychology / Psicología Conductual, 23(2), 287-303.

Study type: Cross-cultural validation and factor structure study

Background and objectives

The Dysfunctional Attitude Scale (DAS) is a 40-item measure of dysfunctional schemas, a key construct in the cognitive model of depression according to cognitive therapy. Dysfunctional or depressogenic schemas are thought to be shaped by early negative life experiences, are relatively stable, and remain dormant until negative events activate them, producing negative automatic thoughts that constitute the cognitive core of depression. Although the DAS has been widely accepted as a measure of dysfunctional schemas, exploratory factor analyses have yielded mixed results with most studies showing between two-factor and four-factor solutions.

In a recent Dutch study, de Graaf et al. (2009) proposed a revised version of the DAS (DAS-R) that retains 17 items from the original 40, organized into two factors: Perfectionism/Performance evaluation (11 items) and Dependency (6 items). This innovative study also suggested the possibility of a higher-order factor structure, though this possibility was not analyzed in their initial investigation. The current study aimed to analyze the factor structure and psychometric properties of the Spanish version of the DAS-R with three independent samples (N total = 629), hypothesizing that the Spanish version would show the same two-factor solution found by de Graaf et al. (2009) and would have similar psychometric properties. Additionally, it was hypothesized that the DAS-R would have a hierarchical factor structure with a higher-order construct reflecting dysfunctional schemas in general, and two first-order factors reflecting specific types of dysfunctional schemas.

Method

Participants

The study used three independent samples with a total of 629 participants:

Sample 1: 210 undergraduate students (age range 18-45 years, M = 20.46, SD = 3.45) from a university in northern Spain. Sixty-four percent were studying Psychology, 15% Speech Therapy, and 21% Teaching. Eighty-four percent were women. Nineteen percent had received psychological or psychiatric treatment at some time, but only 4% were currently in treatment. Four percent were taking psychotropic medication.

Sample 2: 289 participants (59.5% female) with ages ranging from 22 to 69 years (M = 35.38, SD = 8.63). Educational level was: 7.3% primary education, 32.8% mid-level study graduates (high school or vocational training), and 59.9% were college graduates. They were recruited through social media and all were Spanish speakers. Thirty-six percent reported having received psychological or psychiatric treatment at some time, but only 6.6% were currently in treatment. Four point eight percent reported psychotropic medication consumption. No specific information about the type of psychological or psychiatric treatment received was reported.

Sample 3: 130 undergraduate students (age range 18-46 years, M = 22.58, SD = 5.09) from a university in southern Spain. Most of them (77%) were studying Psychology. The remaining 23% were studying Teaching, Law, or English Philology. Fifty-nine percent were women. Nineteen percent reported having received psychological or psychiatric treatment at some time, but only 3.8% were currently in treatment. Three point eight percent were taking psychotropic medication. No specific details about the type of psychological or psychiatric treatment received were reported.

Instrument(s) under study

Dysfunctional Attitude Scale-Revised (DAS-R). The revised version of the DAS (de Graaf et al., 2009) consists of 17 items rated on a 7-point Likert scale (7 = fully agree; 1 = fully disagree). It is proposed to contain two correlated factors: Perfectionism/Performance evaluation (11 items) and Dependency (6 items). The authors suggested the possibility of a second-order factor, although this possibility was not analyzed in their original study. The Spanish version of the DAS-R (Ruiz, Odriozola-González, & Suárez-Falcón, 2014) was used.

Other outcome measures

Seven additional instruments were used to assess convergent, discriminant, and criterion-related validity:

1. Automatic Thoughts Questionnaire-8 (ATQ-8) (Hollon & Kendall, 1980; Spanish version by Cano-García & Rodríguez-Franco, 2002). An 8-item measure of frequency of negative automatic thoughts experienced during the past week on a 5-point Likert scale. Used only in Sample 1.

2. Beck Depression Inventory-II (BDI-II) (Beck, Steer, & Brown, 1996; Spanish version by Sanz, Perdigón, & Vázquez, 2003). A 21-item self-report measure of depressive symptoms during the past two weeks on a scale ranging from 0 (not present) to 3 (severe). Used in Samples 1 and 3.

3. Depression Anxiety Stress Scales-21 (DASS-21) (Antony, Bieling, Cox, Enns, & Swinson, 1998; Spanish version by Sanz, Perdígón, & Vázquez, 2003). A 21-item, 4-point Likert scale consisting of sentences describing negative emotional states. It contains three subscales (Depression, Anxiety, and Stress) and has shown good internal consistency and convergent and discriminant validity. Used in Sample 2.

4. Metacognitions Questionnaire-30 (MCQ-30) (Wells & Cartwright-Hatton, 2004; Spanish version by Odriozola-González, 2011). A short version of the MCQ-65 consisting of 30 items on a 4-point Likert scale that measures dysfunctional metacognitive beliefs. It contains the following five factors: Positive beliefs about worry, Negative beliefs about uncontrollability and danger of worry, Beliefs about the need to control thoughts, Cognitive confidence, and Cognitive self-consciousness. Used in Sample 2.

5. Acceptance and Action Questionnaire-II (AAQ-II) (Bond et al., 2011; Spanish version by Ruiz, Langer, Luciano, Cangas, & Beltrán, 2013). A general measure of psychological inflexibility consisting of 7 items rated on a 7-point Likert scale. The items reflect unwillingness to experience unwanted emotions and thoughts and the inability to be in the present moment and behave according to valued actions when experiencing psychological events that could undermine them. Used in all three samples.

6. Believability of Anxious Feelings and Thoughts Questionnaire (BAFT) (Herzberg et al., 2012; Spanish version by Ruiz & Odriozola-González, 2014). A self-report measure of cognitive fusion with anxious thoughts and feelings. It consists of 16 items representing different thoughts that are rated on a 7-point Likert scale ranging from 1 (not at all believable) to 7 (completely believable) according to the extent that the individual believes in them. The BAFT has excellent internal consistency and a hierarchical factor structure with three lower-order factors and one hierarchical factor. Used in Sample 2.

Data analysis

Confirmatory factor analysis (CFA) was conducted using LISREL, version 8.71 (Jöreskog & Sörbom, 1999) and adopting a maximum-likelihood estimation method. The robustness of the two-factor model found by de Graaf et al. (2009) and an alternative one-factor model was assessed by conducting confirmatory factor analyses on the overall sample. The one-factor model was analyzed because most of the studies use the total score on the DAS without separating the scores by their factors. Goodness-of-fit was examined computing the following fit indexes: (a) the root mean square error of approximation (RMSEA); (b) the comparative fit index (CFI); (c) the non-normed fit index (NNFI); (d) the goodness-of-fit index (GFI); and (e) the expected cross-validation index (ECVI). According to Kelloway (1998), RMSEA values of .10 or less represent a good fit to the data. With respect to the CFI, NNFI, and GFI, values above .90 indicate well-fitting models, and above .95 represent a very good fit to the data. The ECVI was computed to compare the goodness-of-fit of the two-factor versus the one-factor model.

An additional CFA was conducted to test for the presence of a second-order factor, as was suggested by de Graaf et al. (2009). The fit of this model was compared to the one of the two-correlated-factor model. Following the recommendations by Gignac (2007), the Schmid-Leiman transformation (Schmid & Leiman, 1957) was conducted as an alternative to the nested factor modeling to explore the factor loadings of the items and the extracted variance accounted for by the general factor. This procedure performs a secondary exploratory factor analysis on the latent factor intercorrelations obtained from a previous EFA and facilitates interpretation of primary factors (items) relative to higher-order factors by computing direct relations between primary variables and second-order factors. Likewise, the proportion of the general factor accounts for the extracted variance is indicative of the presence of a general factor (range= 40%-50%; Gorsuch, 1983). This analysis was computed using Factor 9.2. An exploratory unweighted least squares factor analysis with direct oblimin rotation and the Schmid-Leiman transformation (Schmid & Leiman, 1957) was conducted. Additionally, the syntax developed by Wolff and Preising (2005) for SPSS was used to compute the total extracted variance accounted for by the higher order factor. Lastly, additional confirmatory factor analyses were performed with the model that showed the best fit to the data in order to explore whether the factor structure was the same for the samples that responded on paper-and-pencil (samples 1 and 3) and the sample that responded via internet sample 2).

The remaining statistical analyses were performed on SPSS Statistics 17 (SPSS Inc., 2008). Cronbach's alphas providing confidence intervals according to Duhacheck and Iacobbucci (2004) were computed to explore the internal consistency of the DAS-R. Corrected item-total correlations were obtained to identify items that should be removed because of low discriminative power (i.e., values below .20). Descriptive data were also calculated. To examine discriminant constructs on the DAS-R were compared computing Student's t, between: (a) participants receiving psychological or psychiatric treatment and those who were not receiving them (samples 1 and 2; analyses were not conducted with sample 3 because only 4 participants were receiving treatment in this sample); and (b) participants with scores above and below the cutoffs on the BDI-II (samples 1 and 3) and DASS-21 (sample 2). Temporal stability of the DAS-R scores across 9 months was examined in a subsample of 106 participants belonging to sample 2. Zero-order correlations between the DAS-R and full DAS scale were computed for sample 1. Correlations with other scales were calculated to assess convergent construct validity. According to the cognitive theory of depression, it was expected that the DAS-R would show medium to strong correlations with negative automatic thoughts (the different versions of the ATQ) and depression (BDI-II) and anxiety symptoms (DASS-21). According to the metacognitive model of emotional disorders (e.g., Wells, 2009), it was expected that the DAS-R would show medium to strong correlations with the subscales of the MCQ-30. Lastly, we expected that the DAS-R would show medium to strong correlations with the AAQ-II and BAFT because psychological inflexibility strongly correlated with the DAS in previous research (Cristea, Montgomery, Szamoskozi, & David, 2013; Ruiz & Odriozola-González, in press).

Results

Factor structure

Table 1 presents the results of the CFA comparing the two-factor and one-factor models. The one-factor model obtained an acceptable fit, but inferior to the observed for the two-factor model. The chi-square difference between the two competing models was 370.38 (df = 1, p < .05), indicating that the two-factor model showed a significantly better fit to the data.

Table 1 shows the goodness-of-fit indices of the one-factor and two-factor models of the DAS (N= 629):

Two-factor model: RMSEA = .053 [90% CI .046, .060], CFI = .99, NNFI = .99, GFI = .89, ECVI = .63 [90% CI .55, .72], χ² (df) Sattora-Bentler = 323.46 (118).

One-factor model: RMSEA = .081 [90% CI .074, .087], CFI = .97, NNFI = .97, GFI = .82, ECVI = 1.07 [90% CI .95, 1.20], χ² (df) = 603.79 (119).

The two factors in the two-factor model were strongly correlated (r = .78). Table 2 shows the original items, their translation into Spanish, and factor loadings for the two-factor model. Both factors were strongly related (r = .78). Table 2 shows the original items, their translation into Spanish, and factor loadings for the two-factor model. All items showed strong factor loadings in the expected factors according to the study by de Graaf et al. (2009), and factor loadings for the two-factor model.

The goodness-of-fit indices for the two-factor model with a second-order factor are shown in Table 3: RMSEA = .045 [90% CI .038, .052], CFI = .99, NNFI = .99, GFI = .99, ECVI = .54 [90% CI .47, .62], χ² (df) Sattora-Bentler = 265.93 (117).

According to the Schmid-Leiman transformation, all items of the DAS-R seemed to represent the general factor because they showed loadings above .30 (Tabachnick & Fidell, 2007). The range of factor loadings was between .43 (item 6) and .80 (item 9). The general factor accounted for 65.4% of the extracted variance, a proportion clearly above the range considered as indicative of the presence of a general factor (40%-50%; Gorsuch, 1983).

Lastly, the hierarchical factor structure presented very good and equivalent fits for the samples that completed the DAS-R on paper-and-pencil (samples 1 and 3): RMSEA= .041, CFI, NNFI, and GFI= .99; ECVI= .85; Sattora-Bentler χ²(117)= 173.34, and the sample that responded via internet (samples 2: RMSEA= .041; CFI, NNFI, and GFI= .99; ECVI= .76; Sattora-Bentler χ²(117)= 184.26. Factor loadings also showed the same pattern and were basically the same as in the total sample.

Internal consistency, descriptive data and criterion validity

Table 4 shows that Cronbach's alpha of the overall DAS-R ranged from .87 (sample 1) to .91 (sample 2), with an overall alpha of .90, 95% CI [.89, .91]. With respect to the DAS-R factors, Perfectionism/Performance evaluation showed alpha between .85 (sample 1) and .89 (sample 2), with an overall alpha of .87, 95% CI [.86, .83]. Dependency ranged from .76 (sample 1) and .82 (sample 2), with an overall alpha of .81, 95% CI [.78, .83]. Corrected item-total correlations of the DAS-R ranged from .43 to .69 for the total sample. With respect to the two factors, Perfectionism/Performance evaluation showed item-total correlation between .47 and .72, whereas for Dependency they were between .45 and .63.

Table 5 shows that participants with scores above the cutoffs on the BDI-II and DASS-21 or who were receiving psychological or psychiatric treatment scored statistically significantly higher on the DAS-R and its subscales than those with scores below these cutoffs. Additionally, participants receiving psychological/psychiatric treatment in sample 1, but not those from sample 2, showed higher scores on the DAS-R than those who were not receiving it.

Temporal stability and zero-order correlations with other related constructs

The temporal stability of the DAS-R across 9 months in a subsample of sample 2 (N= 106) was adequate (r= .72). The overall DAS-R score was very strongly correlated with the complete DAS (r= .95) in sample 1. The correlations of the DAS-R were comparable to those shown by the complete DAS in sample 1.

The DAS-R showed correlations with all other assessed constructs in theoretically coherent ways. Specifically, the DAS-R showed positive correlations with negative automatic thoughts, cognitive fusion and psychological inflexibility. Table 6 shows the complete zero-order correlations. Briefly summarizing the main patterns:

• Correlations with the ATQ (Sample 1): DAS-R Total r = .54**
• Correlations with the BDI-II (Samples 1 and 3): DAS-R Total r = .44** to .35**
• Correlations with DASS-21 (Sample 2): Depression r = .47**, Anxiety r = .38**, Stress r = .45**
• Correlations with MCQ-30 (Sample 2): Positive metacognitive beliefs r = .39**, Negative metacognitive beliefs r = .39**, Need to control thoughts r = .35**
• Correlations with AAQ-II (Samples 1-3): DAS-R Total r = .55** to .57**
• Correlations with BAFT (Sample 2): DAS-R Total r = .49**

Discussion and conclusions

The data obtained in this study provide promising evidence that the Spanish version of the DAS-R is a valid and reliable measure of dysfunctional schemas. Overall, the current data are very similar to those obtained by de Graaf et al. (2009). Specifically, the DAS-R showed excellent overall internal consistency (overall α= .90) and good internal consistency for its factors (Perfectionism/Performance evaluation: overall α= .87; Dependency: overall α= .81). The construct convergent validity of the DAS-R was examined by analyzing its correlations with related constructs such as negative automatic thoughts, depressive and anxiety symptoms, dysfunctional metacognitive beliefs, psychological inflexibility, and cognitive fusion. All correlations found were in the expected directions. The DAS-R also presented discriminant validity to the extent that participants experiencing mild levels of emotional symptoms (i.e., participants with scores on the BDI-II and DASS-21 scored statistically significantly higher on the DAS-R and its subscales than those who scored below the cutoffs. Likewise, participants receiving psychological/psychiatric treatment in sample 1, showed higher scores on the DAS-R than those who were not receiving treatment.

The factor analyses conducted in this study deserve detailed attention. The confirmatory factor analysis revealed that the two-factor model of the DAS-R obtained a better fit to the data than the alternative one-factor model. Likewise, all items showed factor loadings in the expected factors according to the study by de Graaf et al. (2009). However, as expected, a model consisting of a hierarchical structure with a general factor and the previous two first-order factors obtained the best fit to the data. This result was statistically significantly better than the fit of the two-factor model without a second-order factor. Lastly, similar fit indices of the factor structure were found for the samples that completed the DAS-R on paper-and-pencil (samples 1 and 3) and the sample that responded via internet (samples 2), demonstrating equivalence of administering the DAS-R in both ways.

The hierarchical factor structure found in this study has several relevant implications. On one hand, the presence of a general factor provides a theoretical justification of using the total score of the DAS-R. This score provides a general measure of dysfunctional schemas and not merely the aggregation of the two types of dysfunctional schemas identified. On the other hand, in some contexts, it may be more advisable to analyze the scores on first-order factors (Perfectionism/Performance evaluation and Dependency). As previously discussed, the possibility of analyzing the presence of specific dysfunctional schemas can be seen as an advance in the study of depression according to cognitive therapy. A limitation of the de Graaf et al.'s study, however, is that the validation of the DAS-R was only conducted with a Dutch version of the DAS. Therefore, replication of the DAS-R factor structure and psychometric properties in other languages and cultures is needed.

In conclusion, the Spanish version of the DAS-R seems to be a reliable and valid measure of dysfunctional schemas, consisting of a hierarchical factor structure with a general factor and two first-order factors. Due to the clear factor structure of the DAS-R, its use should be recommended instead of the full DAS scale. The DAS-R provides the researcher and clinician the possibility to investigate specific types of dysfunctional schemas reliably and provides a theoretically justified reason for the use of the total score as a general measure of dysfunctional thinking. Further research, however, should be conducted to confirm the psychometric properties and hierarchical factor structure of the DAS-R in other languages.

Significance and contribution

This study provides the first comprehensive validation of the Dysfunctional Attitude Scale-Revised in Spanish-speaking populations, demonstrating that the Spanish version retains the hierarchical factor structure and psychometric properties similar to those of the original Dutch version. By establishing that the DAS-R is a valid and reliable measure of dysfunctional schemas in Spanish populations, with a clear structure that allows analysis of both the general factor and specific factors (Perfectionism/Performance evaluation and Dependency), this study significantly contributes to the availability of measurement instruments for the key construct of dysfunctional schemas in the cognitive model of depression. The validation with multiple samples and administration methods, along with evidence of convergent and discriminant validity, establishes the DAS-R as a reliable psychometric tool for research and clinical assessment in Spanish-speaking contexts.

---