Fill Out Your Dtsfunctional Additude Scale Form

If I don’t set the highest standards for

1.myself, I am likely to end up a second-rate person.

My value as a person depends greatly on

2.what others think of me.

3.People will probably think less of me if I make a mistake.

4.I am nothing if a person I love doesn’t love me.

5.If other people know what you are really like, they will think less of you.

6.If I fail at my work, then I am a failure as a person.

My happiness depends more on other

7.people than it does me.

8.I cannot be happy unless most people I know admire me.

9.It is best to give up your own interests in order to please other people.

If I am to be a worthwhile person, I must

1.be truly outstanding in at least one major respect.

If you don’t have other people to lean on,

2.you are bound to be sad.

3.I do not need the approval of other people in order to be happy.

4.If you cannot do something well, there is little point in doing it at all.

5.If I do not do well all the time, people will not respect me.

6.If others dislike you, you cannot be happy.

7.People who have good ideas are more worthy than those who do not.

8.If I do not do as well as other people, it means I am an inferior human being.

9.If I fail partly, it is as bad as being a complete failure.

A central tenet of cognitive theory of depression is that dys- functional attitudes have a critical etiologic role for vulnerability to depression (Beck, Rush, Shaw, & Emery, 1979). Individuals who endorse dysfunctional attitudes are thought to be at increased risk for depression onset (e.g., Alloy et al., 2006; Segal, Gemar, & Williams, 1999). Further, elevations in dysfunctional attitudes are thought to maintain an episode and are often central targets of

intervention during cognitive– behavioral treatment. Consistent with these ideas, numerous studies have observed high levels of dysfunctional attitudes among people diagnosed with unipolar depression (e.g., Dent & Teasdale, 1988; Norman, Miller, & Dow, 1988).

Dysfunctional attitudes are often assessed with the Dysfunc- tional Attitude Scale (DAS; Weissman, 1979). The DAS was

Christopher G. Beevers, Department of Psychology, University of Texas at Austin; David R. Strong, Department of Psychiatry and Human Behavior, Brown University, and Addictions Research, Butler Hospital, Providence, Rhode Island; Bjo¨rn Meyer, Department of Psychology, City University, London, London, England; Paul A. Pilkonis, Department of Psychiatry, West- ern Psychiatric Institute and Clinic, University of Pittsburgh Medical Center; and Ivan W. Miller, Department of Psychiatry and Human Behavior, Brown University, and Psychosocial Research Program, Butler Hospital.

Research reported in this article was supported in part by several National Institute of Mental Health (NIMH) grants. Research conducted at Butler Hospital and Brown Medical School was supported by NIMH Grant MH43866; Ivan W. Miller was the principal investigator. Research conducted as part of the NIMH Treatment of Depression Collaborative Research Program was part of a multisite program initiated and sponsored by the NIMH Psy- chosocial Treatments Research Branch. The program was funded by cooper- ative agreements to six sites: George Washington University, MH33762; University of Pittsburgh, MH33753; University of Oklahoma, MH33760; Yale University, MH33827; Clarke Institute of Psychiatry, University of Toronto, MH38231; and Rush-Presbyterian-St. Luke’s Medical Center, MH35017. The

principal NIMH collaborators were Irene Elkin, coordinator; Tracie Shea, associate coordinator; John P. Docherty; and Morris B. Parloff. The principal investigators and project coordinators at the three research sites were Stuart M. Sotsky and David Glass (George Washington University), Stanley D. Imber and Paul A. Pilkonis (University of Pittsburgh), and John T. Watkins and William Leber (University of Oklahoma). The principal investigators and project coordinators at the three sites responsible for training therapists were Myrna Weissman, Eve Chevron, and Bruce J. Rounsaville (Yale University); Brian F. Shaw and T. Michael Vallis (Clarke Institute of Psychiatry); and Jan A. Fawcett and Phillip Epstein (Rush-Presbyterian-St. Luke’s Medical Cen- ter). Collaborators in the data management and data analysis aspects of the program were C. James Klett, Joseph F. Collins, and Roderic Gillis of the Veterans Administration Cooperative Studies Program, Perry Point, Maryland.

We thank Aaron T. Beck for allowing us to reproduce items from the original Dysfunctional Attitude Scale.

Correspondence concerning this article should be addressed to Christo- pher G. Beevers, Department of Psychology, University of Texas at Austin,

1University Station A8000, Austin, TX 78712-0187. E-mail: beevers@psy.utexas.edu

originally a 100-item scale and was subsequently refined into two 40-item forms (Versions A and B) based on factor analyses from a student population. Although Beck and colleagues have recom- mended using an 80-item version (Beck, Brown, Steer, & Weiss- man, 1991), the DAS-A is used more frequently. The DAS-A is a self-report questionnaire that measures a variety of rigid, negative, and perfectionist attitudes. Many items also assess dysfunctional contingencies for self-worth (e.g., “If I do not do well all the time, people will not respect me”). Despite the widespread use of the DAS-A, relatively few investigations have examined its psycho- metric properties. Further, no study has used modern analytic techniques, such as item response theory (IRT), to examine the psychometrics of the DAS-A items.

Studies that have investigated the psychometric properties of the DAS-A typically used evaluation methods based on traditional, or classical, test theory (Nunnally, 1979). This approach emphasizes the internal consistency reliability, test–retest reliability, and factor structure of an assessment. Past research suggests that the DAS-A has good internal consistency reliability, test–retest reliability, and replicable factor structure (e.g., Dobson & Breiter, 1983). For instance, using Treatment of Depression Collaborative Research Program (TDCRP) data, Imber et al. (1990) reported that the DAS-A has two related factors: perfectionism and need for ap- proval. These factors had good internal consistency reliability ( .91 and .82, respectively) and were highly correlated (r

.59) with each other. Using structural equation modeling, Zuroff, Blatt, Sanislow, Bondi, and Pilkonis (1999) reported that the Perfectionism and Need for Approval subscales had high factor loadings (.87 and .85, respectively) on a common latent variable. These findings suggest that the DAS-A is sufficiently unidimen- sional to permit use of a total score to assess dysfunctional atti- tudes.

Although these studies made important contributions to the development of the DAS-A, classic test theory methods have limitations. Specifically, they do not directly assess the adequacy of the response options used in the DAS-A or how well DAS-A items discriminate individuals who differ in their level of dysfunc- tional attitudes (for a review of criticisms of classic test theory, see Embretson, 1996). On the basis of these and other limitations, it has been argued that state-of-the-art scale development should use IRT methods to determine the psychometric performance of scale items (Embretson, 1996). Although IRT methods have been used successfully to develop and refine psychopathology-relevant as- sessments such as gambling attitudes (Strong, Breen, & Lejuez, 2004), alcohol problems (Kahler, Strong, Read, Palfai, & Wood, 2004), personality traits (Gomez, Cooper, & Gomez, 2005), and depressive symptoms (Cole, Rabin, Smith, & Kaufman, 2004), no such study has been conducted with any form of the DAS.

An additional benefit of using IRT to refine the DAS-A is that it could reduce the number of items required to assess dysfunc- tional attitudes. A shortened assessment of dysfunctional attitudes could be beneficial for a number of reasons. First, a shorter form could ensure greater compliance (i.e., fewer skipped or missing items). Indeed, in one study, more than 10% of respondents ter- minated a 20-item depression questionnaire before completion (Kohout, Berkman, Evans, & Cornoni-Huntley, 1993). Similarly, faster administration is important in psychotherapy trials, as par- ticipants are often required to complete a battery of self-report questionnaires throughout the course of treatment. Reducing the

length of questionnaires could reduce subject burden. Alterna- tively, if subject burden is already minimal, briefer questionnaires could allow for more frequent assessments during treatment with- out substantially increasing subject burden. Repeated assessments are often critical for identifying putative mediators (cf. Kraemer, Wilson, Fairburn, & Agras, 2002). Finally, psychopathology re- search may also benefit from a shorter version of the DAS, as level of dysfunctional attitudes measured following a dysphoric mood induction is linked to depression vulnerability (Segal et al., 2006). As mood states induced in the laboratory tend to be brief (Martin, 1990), a dysfunctional attitude scale that can be completed quickly may provide an assessment that is more uniformly influenced by a mood induction.

Given the widespread influence of cognitive theory on etiologic and treatment studies of depression (Beck, 2005), the prevalent use of the DAS-A, and the general benefit of using brief and efficient assessments to increase compliance and reduce subject burden, we used IRT methods to examine, refine, and shorten the DAS-A. In doing so, our goal was to develop a briefer version of the DAS-A that (a) has an optimal response format, (b) has items that discrim- inate individuals along the dysfunctional attitudes continuum, (c) is efficient, (d) is highly consistent with the original set of DAS items, (e) has adequate internal and test–retest reliability, and (f) has adequate concurrent and predictive validity. Because parallel forms of the same questionnaire can be useful (e.g., assessing dysfunctional attitudes before and after a mood induction), a final goal was to develop two parallel short forms of the DAS.

To achieve these goals, we pooled data from two treatment studies of unipolar depression: a randomized clinical trial (RCT) comparing the efficacy of several treatments among depressed outpatients (TDCRP; Elkin, 1994) and an RCT comparing the efficacy of several depression treatments in the posthospital care of depressed inpatients (Miller et al., 2005). Using IRT methods, we examined the DAS-A responses prior to treatment to evaluate, refine, and shorten the DAS-A. We then used approaches from traditional test theory to examine the reliability and validity of the short versions of the DAS-A.

Method

Participants

Data were pooled from two RCTs for unipolar depression (N 367). The first RCT was the TDCRP. The design and procedures of the TDCRP have been described in detail elsewhere (e.g., Elkin, 1994). A total of 250 patients met study entry criteria and were randomly assigned to treatment; of these, pre- and posttreatment DAS-A data were obtained from 246 and 191 participants, respec- tively. Participants were predominantly female (78%), White (89%), and in their 30s (M 35.05, SD 8.5). The design and procedures of the second RCT from which we obtained data have also been described elsewhere (e.g., Miller et al., 2005). A total of 121 patients met study entry criteria and were randomly assigned to treatment; pretreatment and posttreatment DAS-A data were available from 121 and 99 participants, respectively. Participants were predominately female (74%), White (93%), and in their late 30s (M 37.97, SD 11.83). Both RCTs required informed written consent and were approved by institutional review boards. Using pretreatment data from depressed outpatients and recently

discharged depressed inpatients should allow us to evaluate the performance of the DAS-A items across a broad range of dysfunc- tional attitudes typically endorsed by depressed individuals.

Measures

DAS (Weissman, 1979). The DAS-A has 40 statements to which participants respond on a 7-point scale (i.e., totally agree, agree very much, agree slightly, neutral, disagree slightly, dis- agree very much, totally disagree). The DAS-A assesses dysfunc- tional beliefs that are thought to reflect a person’s self-evaluation. DAS-A items measure concerns about approval from others, pre- requisites for happiness, and perfectionist standards. The DAS-A has been used widely in depressed and psychiatric control popu- lations (Oliver & Baumgart, 1985). Other studies have docu- mented that the DAS-A has good test–retest reliability (correlation of .84 over an 8-week period; Weissman, 1979). Internal consis- tency reliability measured before treatment in the present study was very good ( .93).

Beck Depression Inventory (BDI; Beck & Steer, 1993). The BDI is a widely used self-report questionnaire that assesses de- pression severity. The BDI consists of 21 items and measures the presence and severity of cognitive, motivational, affective, and somatic symptoms of depression. Past reports indicate that test– retest reliability is adequate (Beck, Steer, & Garbin, 1988). The BDI has been found to be valid among psychiatric inpatient and outpatient samples (Beck et al., 1988). Internal consistency reli- ability measured before treatment in the present study was good ( .84).

Cognitive Bias Questionnaire (CBQ; Krantz & Hammen, 1979). The CBQ was used to assess negatively biased, self-referent in- formation processing. The CBQ presents four vignettes (8 –12 sentences in each vignette) that are ambiguous in outcome (e.g., employer gives potentially negative feedback on employee’s work). Participants imagine themselves in each situation and then select one of four response options to four questions per vignette. Options for each question represent depressed– distorted, nondepressed– distorted, depressed–nondistorted, and nondepressed–nondistorted cognitive styles. Consistent with pre- vious studies, the present study focused on the depressed– distorted subscale, which has exhibited good reliability and validity (Krantz & Hammen, 1979; Miller & Norman, 1986; Norman, Miller, & Klee, 1983). Internal consistency reliability measured before treat- ment in the present sample was adequate ( .80).

Hopelessness Scale (HS; Beck, Weissman, Lester, & Trexler, 1974). The HS is a 20-item true–false self-report questionnaire that assesses participants’ negative expectations regarding the fu- ture. Scores on the HS range from 0 to 20, with higher scores indicating higher levels of hopelessness. The HS has adequate 1-week and 3-week test–retest reliability (Beck, Steer, & Ranieri, 1988). Internal consistency reliability measured before treatment in the present study was good (Kuder–Richardson 20 .92).

Statistical Model: Overview of Item Response Theory (IRT) Analyses

IRT methods provide a means of scaling both items and persons along a theorized underlying latent continuum of dysfunctional attitudes. These methods assume that individuals vary along a

single latent continuum. Thus, a common factors analysis was conducted prior to IRT modeling. With support for a primary dimension underlying the DAS, we chose to apply a nonparametric IRT modeling strategy to explore the performance of individual DAS items.

Two broad classes of IRT models include parametric (cf. Birn- baum, 1968; Rasch, 1960) and nonparametric approaches (cf. Mokken & Lewis, 1982; Molenaar, 1997; Ramsay, 1991). We chose to use a nonparametric approach to modeling responses to the DAS-A because we had no a priori expectation about the form of response distributions, and we wanted to allow items with nonmonotonic item response functions to be identified. Nonpara- metric methods have been recommended prior to choosing and fitting parametric item response models to personality data and when analyzing smaller data sets, as in the present study (Meijer & Baneke, 2004). Further, parametric and nonparametric approaches often lead to similar item selection (Lei, Dunbar, & Kolen, 2004).

Using a nonparametric approach, we constructed item charac- teristic curves that relate the likelihood of endorsing increasing scores on each item to latent levels of dysfunctional attitudes prior to examining the performance of individual options. We then examined items’ option characteristic curves (OCCs). These OCCs relate the likelihood of endorsing each option on each item to latent levels of dysfunctional attitudes. On the basis of examination of the OCC, items with poor discrimination were identified and dropped from further analysis. Items were identified as having good discrimination if the likelihood of choosing higher options (e.g., “agree very much” vs. “disagree very much”) increased systematically with increasing levels of dysfunctional attitudes. Poor discrimination was identified when higher item options failed to be observed with higher likelihood than lower options despite increases in levels of dysfunctional attitudes. We required that items provide information (e.g., higher options become more likely than lower options) within ranges of the dysfunctional attitudes that would be observed within a significant number of individuals in the present sample (5th–95th percentiles). Therefore, we de- cided that a multicategory item should make at least two levels of discrimination to justify inclusion in a shortened version of the DAS-A. Among items making multiple discriminations, response options that did not contribute to the discrimination effectiveness of an item were collapsed.

Finally, to explore whether improvements in the efficiency of the DAS-A could be achieved by identifying the presence of items with similar measurement properties, we graphed the item re- sponse functions for the retained items. Using estimates from the item characteristic curves analysis, we identified the region of the continuum (i.e., the item severity) where items provided maximum information and examined how well the items discriminated within this region. From this analysis, items with similar severity and discrimination were considered to provide redundant statistical information about level of dysfunctional attitudes.

We used a nonparametric kernel-smoothing method and soft- ware (TESTGRAF) developed by Ramsay (2001). These methods have been used previously in several studies on the performance of scales measuring depression (Santor & Coyne, 1997, 2001), and in-depth reviews of these methods are available (Santor & Ram- say, 1998). This approach estimates OCCs at each evaluation point by using a local average—a method that gives observations in- creased influence in determining the estimated OCC values if they

fall closer to the specific evaluation point. We considered items to have good response properties if (a) the probability of endorsing increasingly severe response options increased with increasing levels of dysfunctional attitudes and (b) if curves for at least some of the response options intersected more than once between the 5th and 95th percentiles of estimated dysfunctional attitudes.

Results

Unidimensionality

We conducted maximum likelihood common factors analysis of polychoric correlations for the 40 DAS-A items. Results support a primary dimension that was significantly larger than subsequent factors. The primary dimension accounted for 33% of the common variance, with subsequent factors accounting for 4% and 3%, respectively. The eigenvalues for the first three factors were 13.34, 2.41, and 1.71. Thirty-eight out of the 40 items loaded .30 (range .31–.79. Two items (Items 39 and 40) loaded .30 on the primary factor and .30 on the secondary factor. However, with only two items, a secondary dimension was not pursued. These items were retained in the item response analyses for further inspection.

Item Response Analysis

We next submitted all DAS-A items to item response analysis. Analyses are designed to be iterative in that items that fail to meet our criteria for inclusion are dropped and analyses are repeated with the remaining items. In the analysis of the full set of items, inspection of the DAS-A items indicated that 16 items (Items 1, 2, 5, 8, 12, 13, 18, 24, 25, 29, 30, 31, 36, 37, 39, and 40) failed to make multiple discriminations between the 5th and 95th percen- tiles. To illustrate our criteria for inclusion, Figure 1 displays an example of 2 items: one (Item 7) that met inclusion criteria and one (Item 39) that failed to make the required number of discrimina- tions. As is shown in Figure 1, the probability of endorsing increasingly severe response options for Item 7 increases as level of dysfunctional attitudes also increases. In contrast, the probabil- ity of endorsing more severe response options for Item 39 remains relatively stable until approximately the 90th percentile, where the probability of endorsing severe responses quickly increases. Thus, Item 39 discriminates only among participants at the highest end of the dysfunctional attitudes continuum.

After we dropped the 16 items that failed to make multiple dis- criminations, the remaining 24 items were resubmitted to analyses. Although all 24 items appeared to make adequate discriminations, not

all response options were making discriminations. Poorly functioning response options could contribute to decreased reliability in rank ordering individual levels of dysfunctional attitudes.

Examining Utility of Response Options

We examined the seven response options to determine whether poorly performing options might be collapsed, as several of the response options were rarely used and were never more likely to be observed than were other options. The OCC for Option 1 (“totally agree”) and Option 5 (“agree”) performed consistently across items, made distinct discriminations, and were clearly more likely to be endorsed than were other options within specified ranges of the continuum. However, several other options did not perform consis- tently. Option 4 (“neutral”) performed quite poorly. It was the least frequently used option (endorsed in 7% of responses), and it was never more likely to be endorsed than was any other option across all ranges of the continuum. Across all items and all levels of dysfunc- tional attitudes, the probability of endorsing Option 3 (“agree slightly”) was always higher than was Option 4 (“neutral”), suggest- ing a reversal in the order of these response categories. Further, Option 2 (“agree very much”) was not consistently more likely than was Option 3 (“agree slightly”). Whereas Option 7 (“totally dis- agree”) did become consistently more likely than Option 6 (“very much disagree”), this option was used infrequently (endorsed in 9% of responses), and the range of discrimination typically was above the 95th percentile. Therefore, on the basis of inspection of OCCs, we collapsed Responses 2– 4 (“agree very much,” “agree slightly,” “neu- tral”) and Responses 6 and 7 (“disagree very much,” “totally dis- agree”). This resulted in four-level items (i.e., Options 1, 2– 4, 5, 6 –7).

In line with analyses, we labeled the four response options as “totally agree,” “agree,” “disagree,” “totally disagree.”

After recoding, the 24 items were reanalyzed, and OCCs were inspected. As a result of the reanalysis, 6 of the 24 items (Items 6, 17, 23, 26, 27, and 38) failed to make more than one discrimination between the 5th and 95th percentiles and were dropped. The 18 DAS items were retained and again reanalyzed. All 18 items continued to show improved OCCs and continued to make at least two discriminations between the 5th and 95th percentiles. To illustrate the importance of the response format, Figure 2 displays the OCC for Item 3. When allowing all seven options, several of the lower level options (e.g., Options 2– 4) were equally likely within the same range of dysfunctional attitudes and thus could be subsumed within the same option. The OCCs were substantially better when response options were collapsed to form a four-item response format.

DAS-A Item Characteristics

Figure 3 presents the ICC for the 18 remaining DAS-A items with the 4-item response format. Although each of the 18 items adequately discriminated along the dysfunctional attitudes contin- uum, we explored whether efficiency of the DAS-A could be improved by identifying items with similar measurement proper- ties. When two items appeared to provide similar points of dis- crimination along the dysfunctional attitudes continuum, they were considered to be potentially redundant. Although several items did index different levels of the latent trait, there were many redun- dancies among items that operated within adjacent levels of the trait. To evaluate the impact of eliminating items, we rank ordered

the items on the basis of the level within which the item was most discriminating. We then split the 18 items by sorting every other item into a separate 9-item scale. The DAS-SF1 contained DAS-A Items 20, 19, 3, 16, 15, 10, 34, 7, and 33. The DAS-SF2 contained DAS-A Items 21, 28, 35, 11, 4, 32, 22, 9, and 14. The items for each scale are presented in Table 1.

Consistency Within and Between DAS Short Forms

We first examined internal consistency reliability (coefficient alpha) for each short form of the DAS. The alphas were .84 and

.83, respectively, for the DAS-SF1 and DAS-SF2. We also used the continuous estimates of the standard errors to generate reliability estimates across levels of the continuum (see Figure 4) for the DAS-A, DAS-SF1, and DAS-SF2. Although overall internal con- sistency reliability estimates are .80, adopting the shortened DAS forms decreases reliability slightly in the middle ranges of the continuum.

We next examined correlations among these newly formed DAS scales and the original DAS-A scale at two time points: prior to treatment and following treatment. At pretreatment, the original DAS-A was correlated .91 with DAS-SF1 and .92 with DAS-SF2. Both short forms were also highly correlated with each other (r

.85). At posttreatment, similarly high correlations were observed. The DAS-A was correlated .91 with the DAS-SF1 and .93 with the DAS-SF2. The short forms of the DAS were highly correlated (r

.87) with each other at posttreatment.

We also examined whether the means of the short forms were significantly different from each other at each assessment period. At pretreatment, the DAS-SF1 score was significantly higher than the

DAS-SF2 (M 22.37, SD 6.06 vs. M 21.99, SD 5.88), t(362) 2.17, p .03, d .06. Although this mean difference of

0.38points between DAS short forms was statistically significant (in part due to a large sample size), the effect size indicates that this

difference was small. At posttreatment, the DAS-SF1 score was significantly lower than the DAS-SF2 score (M 18.30, SD 5.68 vs. M 18.67, SD 5.62), t(289) 2.17, p .03, d .06. As before, the effect size for this mean difference was quite small.

Change in Dysfunctional Attitudes

We next examined change in dysfunctional attitudes, as assessed by the DAS-A, DAS-SF1, and DAS-SF2, from pretreatment to posttreatment. Significant reductions in dysfunctional attitudes were observed for each of the DAS forms (all ts 9.5, p .001). To determine whether reductions in dysfunctional attitudes dif- fered significantly across forms, we created standardized residu- alized change scores using pretreatment scores to predict its reas- sessment at posttreatment for the DAS-SF1, DAS-SF2, and DAS-A. There were no significant differences in pre- to posttreat- ment residualized change when comparing the DAS-SF1 and DAS-SF2, t(286) .13, p .90, d .00; DAS-A and DAS-SF1, t(288) .09, p .93, d .00; and DAS-A and DAS-SF2,

t(287) .26, p .79, d .01.1 In addition, these change scores were highly correlated with each other (rs ranged from .84 to .91). This suggests that change in dysfunctional attitudes did not sig- nificantly differ across the DAS forms.

1Degrees of freedom vary slightly due to missing data.