Introducing the brief version of the Dysfunctional Attitude Scale (DAS-14) based on a large clinical sample

©


Introduction
Attitudes are one of the core structures of the human cognitive system which help to categorise and interpret experiences in the mind (Beck, 1967). They develop in interaction with the social environment while adjusting to individual needs, and are activated during frustration (Epstein, 1998).
As generally present cognitive constructions in humans, attitudes influence thoughts, emotions and behavior. They may become dysfunctional while maladaptive patterns of reactions occur and perpetuate when encountering environmental stimuli, resulting in incapability to mobilise resources in order to maintain optimal functioning. Dysfunctional attitudes are overgeneralised, inconsistent, unrealistic beliefs organized into a continuum regarding the individual's self, the world and the future (Beck, Rush, Shaw, & Emery, 1979). Appearing in mentally healthy individuals as well as in people with diagnosed mental disorders (i.e. depression and anxiety disorders), they develop during adapting to stressful life events. Extensive research investigating depression found that dysfunctional attitudes are not only maintaining/covarying factors of depressive symptoms, but also predictors of depressive relapse and predictors of depressive episode frequency (Brouwer, Williams, Forand, DeRubeis, & Bockting, 2019;Ingram, Miranda, & Segal, 1998;Scher, Ingram, & Segal, 2005;Scotte, 1995;Thase & Simons, 1992;Theasdal & Dent, 1987). Therefore, measuring dysfunctional attitudes is also a clinically relevant question.
In contrast with the majority of questionnaires assessing dysfunctional attitudes and beliefs that target specific problems -such as sleep (Dysfunctional Attitudes and Beliefs about Sleep; Morin, Vallières, & Ivers, 2007) or sexuality (Sexual Dysfunctional Beliefs Questionnaire;Nobre, Gouveia, & Gomes, 2003) -, dysfunctional attitudes are measured in a wider sense by the Dysfunctional Attitude Scale (DAS). The 40-item DAS, as a widely accepted and applied questionnaire, was developed by Arlene Nancy Weissman (1979) and was completed by graduates and undergraduates. An initial, item-pool version of the DAS contained a hundred, 7-point Likert scale items in order to create two parallel forms that measure the distinctive characteristics of depressive cognitions. Applying principal component analysis with varimax rotation to form the structure of the questionnaire, Weissman found ten factors but used only one global dimension, as the aim of the study was to identify a general vulnerability factor to depression. As a result of empiri cal decisions, the DAS-A and the DAS-B comprising of 40 items and one general factor each, seemed psychometrically relevant, of which DAS form A remained the most widely applied version in clinical practice and depression research (e.g. Fuhr, Reitenbach, Kraemer, Hautzinger, & Meyer 2017;Senormanci et al., 2014). Since the DAS-A has become a common measurement tool to monitor cognitive therapeutic processes, a considerable amount of research has been carried out in several countries that proved the validity and reliability of the DAS in other languages (e.g. Ohrt & Thorell, 1998;Power et al., 1994;Sahin & Sahin, 1992).
Although the 100-item DAS was developed to assess pathogenic features of mental disorders, Weissman's assessments were performed on a normative sample. Therefore, the factor structure of the DAS was revised by Beck, Brown, Steer and Weissman (1991) recruiting a large community of 2023 psychiatric outpatients. As a result, 9 factors were unfolded including 66 items. These factors were Vulnerability, Need for Approval, Success-Perfectionism, Need to Please Others, Imperatives, Need to Impress, Avoidance of Appearing Weak, Control Over Emotions, Disapproval-Dependence, which vary from the unidimensional conception of Weissman (1979).
Synthetizing previous research and practice, Kopp (1994) introduced a 35item version of the questionnaire developed by Burns (1980). Burns (1980) transposed the original functional items into dysfunctional statements, resulting in 35 items collected from the DAS displaying dysfunctional beliefs, whereas Weissman and Beck (1978) also used reversely coded items (e.g. in the 35-items DAS item 6: 'I cannot find happiness without being loved by another person.' is composed in DAS-A in item 40: 'I can find happiness without being loved by another person.'). Based on Burns's theoretical model, compared to the DAS-A, no sum score is calculated; in contrast, seven subscales represent seven dysfunctio nal attitudes (Need for Approval, Love, Achievement, Perfectionism, Entitlement, Omnipotence, Autonomy) with total scores ranging from -10 to 10.
Contributing to the reduction of items, a growing differentiation of psychometric methodology, patient fitting problems and redundancy of items resulted in attempts to shorten the questionnaire.
These mixed results may presumably depend on the features of the obtained population due to cultural diversity (as is the case in the abbreviated versions from Norway, the Netherlands, Iran, Spain, Turkey, Malaysia), sample size and the examination of healthy or clinical samples, the different forms of the questionnaire and the different number of items. Additionally, the use of exploratory and confirmatory factor analyses might have led to different structures of the questionnaire. While exploratory factor analysis gives the possibility for researchers to decide the number of extracted factors, confirmatory factor analysis fit indices are more accurate, but at the same time they rely on a priori findings that limit the number of examined factors. Despite the heterogeneity of the obtained results, Perfectionism, Dependency and DAS total score seemed to remain constant factors, independent of the above mentioned influences.
Psychometric studies of DAS have yielded results of reliability and various forms of validity. Internal consistency of the 40-item general DAS and its subscales were at a range of 0.60-0.92 not only in the original English version but also in other languages (e.g. Ohrt & Thorell, 1998;Power et al., 1994;Weissman, 1979). Test-retest reliability of the 24-item Japanese DAS with intraclass correlation (ICC) was relatively high (0.79, CI 95% = 0.63-0.88) and the concurrent validity was also in the acceptable range with the Irrational Belief Test (r = 0.76, p < 0.001; Tajima et al., 2007). One study tested the predictive validity of the 40-item DAS-A using posttreatment Beck Depression Inventory (BDI) score. Beevers and his coworkers (2007) have found that the predictive value of the test is relatively small regarding depressive symptoms after cognitive therapy (β = 0.18, p = 0.02) (with the control of pretreatment BDI). According to convergent validity studies, the relationship between depressive symptoms and dysfunctional attitudes remains unassured. Medium level correlation has been found between different forms of the DAS and the BDI: DAS-A and BDI (r = 0.47, p < 0.001; Chioqueta & Stiles, 2004); DAS-R and BDI (r = 0.37, p < 0.001; Batmaz & Ozdel, 2016). Strong correlation has been obtained between BDI-II and DAS-24 (non-clinical sample: r = 0.44, p < 0.001; clinical sample: r = 0.63, p < 0.001; Tajima et al., 2007), the Malay revised form of the DAS and BDI on a heterogeneous clinical and non-clinical sample (r = 0.68, p < 0.001; Muhktar & Oei, 2010).
Convergent validity of different forms of the DAS confirmed strong relationship with the Automatic Thoughts Questionnaire (r = 0.51, p < 0.001; Muhktar & Oei, 2010), the General Health Questionnaire (r = 0.56, p < 0.001; Ebrahimi et al., 2012) and the Beck Hopelessness Scale (r = 0.51, p < 0.001; Batmaz & Özdel, 2016), which reflect that dysfunctional attitudes are also in close relationship with the general cognitive system.
To conclude, based on correlational studies, the factor structure of the DAS does not show a consistent picture. In this context, the present study aims to (1) develop a brief and psychometrically reliable version of the DAS and (2) analyze its convergent and construct validity.

Participants and procedures
One clinical, one non-clinical and an adolescent group participated in the research, 1542 subjects altogether. The first group consisted of 1077 adult outpatients (69.5% female, M = 34.2 years, SD = 11.8 years) who were referred for treatment to the mental health center of Semmelweis University Department of Clinical Psychology. Inclusion criteria were 18-65 years of age and at least one of the following mental disorders: major depression, different types of anxiety disorder, personality disorder, obsessive compulsive disorder, eating disorder, sleeping disorder and sexual dysfunction ( Table 2). Exclusion criteria were acute psychosis, acute alcohol or drug use, mental retardation and dementia. Patients took part in a pretreatment diagnostic assessments conducted by a clinical psychologist or intern clinical psychologist. Diagnoses based on the ICD-10 (WHO, 2004) were established by a clinical interview, SCID-I, SCID-II and self-report questionnaires.
The subjects of the non-clinical group were recruited via internet from the general population (n = 270, 77.4% female, M = 29.0 years, SD = 9.5 years). The adolescent sample comprised of 195 secondary school students (84.1% female, M = 15.1 years, SD = 0.90 years), from a secondary school in Pest and Veszprém county.
Data of the clinical sample were collected at the mental health center; subjects from the non-clinical group were assessed via internet, while the adolescents completed the questionnaires during classes after parental consent. The study protocol was approved and reviewed by the regional and institutional ethics board of the collaborating institutions (ethical approval numbers of the study are: SE 194/2012; KRE: 53/2019/P/ET).
Prior to questionnaire completion, informed consent was given by participants providing voluntary and anonymous participation. No monetary reimbursement was given for the assessment ( Table 2). Brief version of the Dysfunctional Attitude Scale

Measurements
Sociodemographic data. As for the clinical sample, age, sex, socioeconomic status, highest level of education and marital status were assessed during the first clinical interview. Questions regarding sociodemographic data of the non-clinical sample were administered via internet. Sociodemographic data of the adolescent sample was collected by paper-and-pencil assessment.
Dysfunctional Attitude Scale (DAS). In the current study, the 35-item version of the DAS was administered in all samples (Kopp, 1994;Weissman & Beck, 1978). Items were scored on a 5-point Likert scale (-2 = Absolutely agree to 2 = Absolutely disagree) resulting in a total score between -70 and 70. Seven subscales are assumed to exist, with 5 items in each subscale (Need for Approval, Seeking Love, Performance Evaluation, Perfectionism, Entitlement, Omnipotence, Autonomy), with results ranging from -10 to 10 each. A higher total score represents more frequent activation of the dysfunctional attitudes. Internal consistency of the subscales ranged from 0.57 to 0.79 in previous studies (Mészáros et al., 2014).
Beck Depression Inventory (BDI). The Beck Depression Inventory (BDI, Beck, Ward, Mendelson, Mock, & Erbaugh, 1961;Kopp & Fóris, 1993) contains 21 items measuring the severity of emotional, motivational, cognitive and somatic symptoms of depression. Each item consists of four statements with varying severity of one particular symptom. Total scores range from 0 to 63 with higher scores reflecting more severe depression. In the present study, Cronbach's α of the BDI confirmed previously assessed excellent reliability (0.90) by Mészáros et al. (2014).

Statistical analysis
To evaluate the factor structure of the DAS-R, an exploratory factor analysis (EFA) was first performed to explore the factor structure on the clinical sample. Prior to conducting the EFA, the Kaiser-Meyer-Olkin measure of sampling adequacy (KMO) and Bartlett's test of sphericity were used to check for the factorability of the data. We decided on the number of factors using parallel analysis based on a polychoric correlation matrix when performing the EFA (Timmerman & Lorenzo-Seva, 2011). The CFA was applied to confirm the factor structure we extracted in the EFA with maximum likelihood estimation with the clinical, non-clinical and adolescent samples. The following indices were used to evaluate how well the data fit the model: Chi-Square value, degrees of freedom (DF), root mean square error of approximation (RMSEA < 0.08), comparative fit index (CFI < 0.90), and Tucker-Lewis index (TLI > 0.90) (Bentler, 1990).
We also calculated the percent of Explained Common Variance (ECV; Reise, Moore, & Haviland, 2010), an index of unidimensionality, attributable to the general factor and each of the three group factors. When Percent of Uncontaminated Correlations (PUC) values are higher than 0.80, general ECV values are less important in predicting bias; when PUC values are lower than 0.80, general ECV values greater than 0.60 and Coefficient Omega Hierarchical values greater than 0.70 suggest that the presence of multidimensionality is not severe enough to disqualify the interpretation of the instrument as primarily unidimensional (Reise, Scheines, Widaman, & Haviland 2013). In turn, group factor ECVs establish the uniqueness of each factor, with a low group ECV indicating little unique variability due to that subscale factor.
We also tested convergent validity with Beck Depression Inventory. Parson correlation was used to test the convergence.
We used Amos, SPSS 27.0 and FACTOR 10.10.03. for statistical analyses.

Exploratory factor analyses (EFA) with clinical sample
The EFA was first used to analyze the data and identify the underlying factors of all 35 items for clinical sample. The KMO value was 0.92, which was higher than the recommended value of 0.60. The Bartlett's test of sphericity (χ 2 = 9623.41, p < 0.001) was adequate, which implied that the data of the clinical sample were suitable for factor analysis. Parallel analysis confirmed a three-factor model of DAS-R. The EFA with varimax rotation extracted a three-factor solution that accounted for 32.3% of the total variance. Several items did not load onto any of the extracted factors, and few of them demonstrated high cross-loading for another factor (Table 3). Factors 1, 2, and 3 (F1, F2 and F3) were named Performance Evaluation and Perfectionism, Entitlement, and Seeking Love, respectively.

Confirmatory factor analyses (CFA) with clinical, non-clinical and adolescent samples
The internal structure of the 35-item DAS and DAS-14 was tested via a series of CFAs (with MLR method, and oblique rotation) specifically. For the 35-item original version, the one-and seven-factor second order versions were tested, respectively. The abbreviated 14-item version was tested in a one-factor form, and also the three-factor, and bifactor form based on previous exploratory factor analysis.
Regarding the DAS-14, both the bifactor and the three-factor solutions demonstrated adequate model fit for every sample ( Table 4). The original DAS-R unidimensional and seven-factor solutions had inadequate model fit.  Table 5 summarises the factor loadings for the unidimensional and bifactor solutions for the DAS-14. Most general factor item loadings were similar to the group factor loadings and to the item loadings from the unidimensional solution. The ECV coefficients for the DAS-14 were low. Eight of the 14 items had Individual Explained Common Variance (IECV) coefficients below 0.50, which indicated that most items were better measures of the group factors than the general factor. Hierarchical (ωH;McDonald, 1999) measures the proportion of total score variance that can be attributed to a single general factor after accounting for group (i.e., subscale) factors. Coefficient Omega Subscale (ωS) is a version of ωH that measures the proportion of subscale score variance that is uniquely due to that group (i.e., subscale) factor after controlling for the general factor. Thus, ωH = 0.60 would indicate that the DAS-14's total score predominantly reflects a single general factor despite the presence of multidimensionality across items, which in turn would permit researchers to interpret the total score as a sufficiently reliable and appropriate measure of the general construct of dysfunctional attitudes.

Validity of the DAS-14
To provide data on the convergent validity of the brief DAS-14, correlations between the DAS-14 scores and BDI were investigated on the clinical sample (n = 1077). As predicted, the 3 subscales and global DAS14 score were positively correlated with the BDI total score. We have found medium positive correlation with the global DAS-14 score (r = 0.36, p < 0.001), Performance Evaluation and Perfectionism subscale (r = 0.41, p < 0.001), and Entitlement subscale (r = 0.39, p < 0.001). Low correlation was found with Seeking Love subscale (r = 0.27, p < 0.001).

Discussion
The present study aims to revise the Dysfunctional Attitude Scale (DAS) in order to clarify its factor structure. Since the rate of dysfunctionality is arranged on a continuum, the current paper aimed to create a measurement tool in order to examine dysfunctional attitudes in mentally healthy population and among mental disorders. Wide range of applicability was taken into account when choosing items. In addition, keeping clinically relevant information served the purpose of facilitating its clinical application by a shortened version. Owing to bifactor solution, total score of the measure reflects the global intensity of dysfunctionality while subscales help to identify different problem areas. Compared to the original 7 and 9 subscale forms of the DAS, in the present study we identified three dysfunctional attitudes, namely, Seeking Love, Performance Evaluation, and Entitlement. These are consistent with Beck's vulnerability theory describing that individuals categorise and structure their experiences by perception (Beck, et al., 1979). As perceptual schemas are damaged, maladaptive, over-generalised structures may distort cognitions by letting a depressive thinking pattern become dominant in cognitive processes. Dysfunctional attitudes evolving on the basis of maladaptive schemas can be measured by the DAS. Although numerous revised versions of the scale were developed, Perfectionism and Dependence/Seeking Love were recurringly confirmed to exist as stable factors (de Graaf et al., 2009;Ruiz et al., 2015a,b).
The current study is partly in accordance with previous findings, mainly by highlighting the significance of Perfectionism and Seeking Love. The depressive cognitive self-system -a core of Beck's concept -provides a basis for negative preconceptions about not only the self but also interpersonal relationships and about the future. These ideations are dividable into well-defined schemas, of which Unlovableness and Self-worthlessness are primal. Self-worthlessness is a core maladaptive schema of depressive self-experience with a drive to prove a person's value through talent and excellence; and Unlovableness operates the level of trust in people as well as the intensity of a desire to be loved (Beck et al., 1979;Tringer, 2007).
In DAS-14 the third dysfunctional attitude (the third factor) named Entitlement represents the expectations and frustrations toward others, such as in item #23 "If I put other people's needs before my own, they should help me when I need something from them." It is in line with the negative view of the world in Beck's negative cognitive triad of depression (e.g. Genuchi & Valdez, 2015;Kopp, Skrabski, & Szedmák, 2000;Rude, Chrisman, Burton Denmark, & Maestas, 2012).
Psychometric analyses considered not only factor content appropriateness, but also clinically relevant features of the DAS. Focusing on Beck's cognitive vulnerability concept, the present study applied three independent samples in order to maintain its theoretical qualities. Since the DAS is primarily a diagnostic measurement tool, another aim was to maintain comprehensibility, conciseness and clarity of the statements. Hence, two psychometrically weaker items #8 and #10 ("If people whom I care about reject me it means there is something wrong with me." and "Being isolated from others is bound to lead to unhappiness.") were kept in the questionnaire in order to strengthen the theoretical coherence of the scale.
The most remarkable result to emerge from the data is the fitting of the bifactor model, showing a clear advantage over balancing the psychometrical parameters and diagnostic properties including multidimensionality of the scale. The bifactor solution allows the use of a general factor in order to gain a holistic picture of the dysfunctionality level of cognitive processes. On the other hand, analyzing the three subscales of Seeking Love, Performance Evaluation and Perfectionism, and Entitlement separately may contribute to obtaining a more detailed and more specific picture about the problem areas of the patient (Brunner, Nagy, & Wilhelm, 2012;Chen, West, & Sousa, 2006).
Another noteworthy finding is the adequate construct validity of the DAS-14. Convergent validity of the brief DAS-14 was also confirmed by finding medium level correlations with the Beck Depression Inventory, supporting a previous study by Batmaz and Ozdel (2016).
A strength of the study, regarding sample size, is that this study has been carried out on the largest clinical sample in the field of psychometric analysis of the DAS so far. Sample heterogeneity and functional aspects required multiphasic analyses, thus, parallel and bifactor modelling were used.
Settings of questionnaire completion differed by subgroups. While clinical subjects went under a diagnostic process with the control of a health professional, the adolescent sample filled in the form by paper-and-pencil testing in a group setting, and healthy subjects via internet. This could cause differences between groups. However, previous research found similar features of data collecting methods when contrasting paper-and-pencil scores with online assessment (i.e. Cronly et al., 2018;Vosylis, Zukauskiene, & Malinauskiene, 2012). Moreover, the considerable gap between subgroup sizes could have influenced the obtained results. Sample size of the clinical population met factor analytic criteria. In contrast, adolescent and normal population sample sizes were not sufficient for conducting exploratory factor analysis properly. However, fitting of the bifactor model was fairly similar in all samples. Thus, our results are confined mainly to a clinical population with the supplementary role of normal and adolescent data.
On the other side, a limitation of the study is that various test batteries were given during assessments. Consequently, test priming effects also differed by subgroups. For instance, the clinical population completed depression and anxiety scales besides the DAS. Previous research supported the finding that monitoring depressive symptoms and anxiety may influence self-reported mood and negative mood priming might increase levels of dysfunctional attitudes (Fresco, Heimberg, Abramowitz, & Bertram, 2006;Mark, Sinclair, & Wellens, 1991). Consequently, an affective priming effect could cause differences in test results. Finally, since our data collecting method was cross-sectional, no test-retest reliability or predictive value was investigated. Research into solving this problem requires longitudinal design. Apart from structural validity, other types of validity should be investigated on normal population.
To summarise, our findings support that DAS-14 is an appropriate, reliable and valid questionnaire to be used in research and clinical practice. It can be used not only for diagnostic purposes but also for monitoring therapeutic effects along multidimensional scaling. In the present study, the brief DAS showed sufficient internal consistency with theoretically proven three factors, thus allowing for reduced administration time. All in all, the brief DAS-14 has several benefits compared to the original, longer forms.