0022-3018/92/1802-0077803.00/0THE JOURNAL OF NERVOUS AND MENTAL DISEASE Vol. 180, No. 2Copyright © 1992 by Williams & Wilkins Printed in U.S.A.

Cognitive Bias in the Articulated Thoughts of Depressed andNondepressed Psychiatric Patients

JUDITH WHITE, M.D., PH.D.,' GERALD C. DAVISON, PH.D.,2 DAVID A. F. HAAGA, PH.D.,' AND KERRIN WHITE, M.D.4

Beck's cognitive theory of depression postulates several types of cognitive bias amongdepressed patients. Empirical studies supporting this hypothesis have usually used question-naire "endorsement" measures of cognition, which may suggest responses to subjects. Weused the articulated thoughts during simulated situations (ATSS) method of cognitive assess-ment in comparing cognitive processes of 15 outpatients with major depression with thoseof 15 nondepressed psychiatric outpatients in three simulated situations. Depressed patientsexceeded nondepressed patients in cognitive bias only in the negative (not the neutral orpositive) simulated situation. Discussion centered on the possible utility of ATSS for researchon cognition in stressful situations.

—J New Ment Dis 180:77-81, 1992

Cognitive theory (Beck, 1963, 1987) holds that de-pression is associated with negative thoughts about theself, the world, and the future. According to the theory,this negative thinking is sustained in part by cognitivebiases, such as a tendency to overgeneralize the impli-cations of one failure and conclude that one will neveragain succeed.' Several studies have corroborated thehypothesis that depressed people exceed nondis-tressed people (Dobson and Shaw, 1986; Wilkinson andBlackburn, 1981) and mixed psychopathology controlgroups (Haley et al., 1985; Krantz and Hammen, 1979)on measures of cognitive bias (for a review, see Haagaet al., 1991). We aimed to extend this earlier research,focusing mainly on methodological issues in cognitiveassessment.

First, most of the research suggesting higher levels

' University of Massachusetts School of Medicine, Worcester, Mas-sachusetts.

2 Department of Psychology, Seeley G. Mudd Building, Universityof Southern California, Los Angeles, California 90089-1061. Send re-print requests and correspondence to Dr. Davison.

The American University, Washington, DC.From McLean Hospital, Belmont, Massachusetts.

These data were presented at the 24th Annual Convention of theAssociation for Advancement of Behavior Therapy, San Francisco,November 1990.

' These biases have often been labeled "distortions" (Beck, 1976),but "bias" appears to be a preferable term. Drawing on related workby Coyne and Gotlib (1983), Alloy and Abramson (1988; see alsoKruglanski and Jaffe, 1988) defined bias as "a tendency to makejudgments in a systematic and consistent manner across specifictimes and situations (e.g., a tendency to draw negative conclusionsabout oneself.... )" (p. 227), contrasting it with distortion (a "judg-ment or conclusion that disagrees or is inconsistent with some com-monly accepted measures of objective reality" [p. 2261). In this light,it seems clear that Beck's theory describes biases in depressive cogni-tion (e.g., accurately perceiving a negative event, but then magnifyingits importance and implications), rather than distorted perceptions.Indeed, in many important instances, such as one's hopes for thefuture, there seems to be no acceptable measure of objective realitywith which to compare one ' s thoughts and thereby identify some ofthem as distortions.

of cognitive bias in depressed than in nondepressedsamples has used questionnaire "endorsement" mea-sures of cognitive bias, in which subjects choose fromamong preselected items the thoughts most typical ofthem (e.g., the Cognitive Bias Questionnaire; Krantzand Hammen, 1979). This approach to cognitive assess-ment may be limiting, because responses endorsed onsuch measures might not reflect naturalistic cognitiveprocesses (Weiner, 1985). Indeed, Coyne (1989)claimed that "There is little evidence that such ques-tionnaires tap the cognitive processes of depressed per-sons routinely occurring in the absence of such a struc-tured inquiry" (p. 235). Accordingly, we assessedcognitive biases with a nonendorsement measure.

Second, we considered it important to compare cog-nitive biases among depressed and nondepressed pa-tients in negative, neutral, and positive situations. Someof Beck 's writings suggest that depressed patients en-gage in biased negative thinking even in the face ofseemingly positive input from the environment: "A de-pressed businessman complains that he is on the vergeof bankruptcy, yet examination of his accounts indi-cates that he is completely solvent and, in fact, is pros-pering" (Beck, 1976, p. 218). Other investigators,though, have found that depressed people exceed non-depressed people on measures of cognitive bias mainlyin negative situations (e.g., Krantz & Gallagher-Thomp-son, 1990). To be sure, depressed people frequentlyface such negative, difficult circumstances (Coyne,1989; Krantz, 1985). However, it is important to deter-mine whether they exceed nondepressed people in cog-nitive bias only in negative situations or in a widerrange of situations, as suggested by Beck.

Thus, empirical results are consistent with Beck ' spostulate of cognitive biases in depression, but it re-mains important to determine whether cognitive biases:

77

78 WHITE et al.

a) can be detected with methods that do not providethe subject with preselected thoughts to endorse, andb) can distinguish depressed from nondepressedgroups primarily in negative situations, as suggested bysome recent research. To evaluate these questions, weused the articulated thoughts during simulated situa-tions (ATSS; Davison et al., 1983) method of cognitiveassessment. The ATSS method involves the audiotapedpresentation of complex hypothetical situations, whichare interrupted at intervals to allow a subject time tothink aloud. Because ATSS does not constrain re-sponses, it is useful for evaluating whether subjectsreport hypothesized cognitions even if these are notsuggested by the measuring instrument itself (Haaga,1989). Furthermore, ATSS is particularly suitable forour purposes because it permits an examination of cog-nitions in a range of situations; prior research indicatesthat experimental manipulation of simulated situationsreliably affects articulated thoughts (Davison and Zigh-elboim, 1987; Davison et al., 1984).

In brief, we hypothesized that indices of cognitivebias could be coded reliably in articulated thoughts andthat depressed patients would exceed nondepressedpsychiatric patients in cognitive bias in a negative simu-lated situation, but not in a positive or neutral situation.

Methods

Subjects

Subjects were 30 psychiatric outpatients at the LosAngeles County-University of Southern California Med-ical Center's Adult Psychiatric Clinic. Diagnoses weredetermined by the attending psychiatrist at intake tothe clinic and were confirmed by the first author (thenan advanced graduate student in clinical psychology)prior to the subject 's participation in our study. Exclu-sionary criteria for the study as a whole were diagnosesof schizophrenia, bipolar affective disorder, organicmental disorder, or mental retardation. The depressedgroup (N = 15) was diagnosed with major depressionaccording to DSM-III criteria (American Psychiatric As-sociation, 1980). The nondepressed group (N = 15),selected for absence of affective disorder or psychosis,received the following principal diagnoses: dependentpersonality disorder (N = 2), antisocial personality dis-order (N = 1), adjustment reactions (without depressedmood; N = 2), impulse control disorder (N = 1), gener-alized anxiety disorder (N = 1), agoraphobia with panicattacks (N = 3), sex offense (court referred; N = 1),and no mental disorder (interpersonal problems; N =4).

Scores on the Beck Depression Inventory (BDI; Becket al., 1979), a reliable and valid self-report measure ofdepressive symptom severity (Beck et al., 1988), pro-vided convergent validity for the diagnostic distinction

between groups. The depressed group obtained a meanBDI score of 27.8 (SD = 6.9; range, 17-38), significantlyhigher than that of the nondepressed group (6.0 ± 4.9;range, 0–14; t[28] = 9.9, p < .001). Indeed, the twodistributions did not overlap.

The groups were demographically similar. The de-pressed group averaged 38.1 years old (± 12.5), nonsig-nificantly older than the nondepressed group (31.8 ±11.8; t[28] = 1.37, p > .1). Eleven of the 15 depressedsubjects were men, compared with seven of 15 nonde-pressed subjects (x2 [1] = 2.22, p > .1),

Materials

Articulated thoughts were measured in three audio-taped, simulated situations, each consisting of fivescripted segments of 15 to 30 seconds duration. Be-tween segments was a 30-second interval, during whichthe subject thought aloud. The negative situation wasa simulated conversation featuring an acquaintance ex-pressing disappointment about an outdoor party thesubject had planned, which had been disrupted by badweather. The positive situation involved compliments,praise, and gratitude from an acquaintance whom thesubject had helped with a plumbing emergency. Theneutral situation found the acquaintance making politeconversational remarks and discussing innocuous top-ics.

Procedure

Subjects completed the BDI, then were instructed tolisten to each ATSS situation (presented in counterbal-anced order), imagine as clearly as possible being partof the event as it unfolded, and say their thoughts aloudbetween segments (indicated by tones on the audio-tape). Subjects' articulated thoughts were audiotapedand transcribed later.

Articulated thoughts coding. Transcripts of articu-lated thoughts were coded by two undergraduate re-search assistants blind to the subjects ' group member-ship. Both coders rated all of the data, scoring eachsegment of articulated thoughts on a 1 (absent) to 5(marked) basis for each cognitive category. Coding cat-egories showed moderate to high interrater reliability,estimated using an alpha model (Novick and Lewis,1967). Reliabilities (in parentheses), definitions, and ex-amples of the categories follow:

a)Arbitrary inference (.64): drawing specific conclu-sions in the absence of relevant evidence (e.g., "Theweatherman was wrong because he was drunk thatday" [which was not stated in the stimulus sce-nario].).b) Selective abstraction (.75): drawing conclusionson the basis of isolated details of an event, even ifthis requires ignoring other contradictory evidence

ARTICULATED THOUGHTS AND DEPRESSION 79

TABLE 1Articulated Thoughts ofDepressed and Nondepressed Patients

NegativeSituationNeutral Positive

ATSS Category Depressed Nondepressed Depressed Nondepressed Depressed NondepressedArbitrary Inference 11.6 ± 2.1 11.1 ± 1.6 11.0 ± 2.1 10.9 ± 1.3 12.4 ± 4.4 10.9 ± 1.6Selective Abstraction 17.1 ± 5.9 14.9 ± 3.5 13.9 ± 4.3 14.6 ± 3.4 18.9 ± 9.3 17.1 ± 5.1Overgeneralization 14.6 ± 5.5 11.5 ± 2.1 12.5 ± 3.2 12.4 ± 2.1 12.3 ± 2.6 10.9 ± 1.4Magnification 15.7 ± 4.8 12.8 ± 2.7 12.4 -t 2.4 12.5 ± 2.6 14.4 ± 3.4 15.1 ± 2.9Personalization 16.2 ± 5.9 13.6 ± 2.7 11.1 ± 1.1 11.9 ± 3.3 12.8 ± 4.4 11.5 ± 1.8Dichotomous Thinking 14.7 ± 6.0 10.9 ± .9 12.1 ± 2.3 11.0 ± 1.2 10.9 ± 1.5 10.7 ± 1.4Total Bias° 89.9 ± 20.1 74.8 ± 5.9 72.9 ± 8.5 73.4 ± 7.3 81.7 ± 10.5 76.3 ± 8.8

Total bias indicates the sum of six previous categories.

(e.g., "I never see that friend myself-I feel soalone.").c) Overgeneralization (.86): holding extreme beliefsbased on specific events and inappropriatelyapplying them to dissimilar occasions or settings(e.g., "The weatherman's always wrong.").d) Magnification (.57): overestimation of the signifi-cance of negative events (e.g., "Society is crum-bling.").e) Personalization (.88): relating events to oneselfeven when there is no connection (e.g., "It was myfault" [that it rained].).f) Dichotomous thinking (.85): thinking in all-or-none terms, categorizing experiences only in one oftwo extremes rather than acknowledging grey areas(e.g., "That' s very bad.").Ratings in each category (1-5) by each rater were

summed across the five segments of a taped situation,then averaged across the two raters, yielding a rangeof possible scores of 5 to 25 for each subject on eachATSS variable in a given situation. Scores on thesesix variables representing types of cognitive bias weresummed to create a total cognitive bias score, rangingin each situation from 30 to 150.

Results

Descriptive statistics for the ATSS categories are pre-sented in Table 1.

Because our hypotheses concerned sets of pairwisecomparisons (depressed expected to exceed nonde-pressed in cognitive bias in the negative situation, notin the other situations), rather than the more globalquestion of whether cognitive bias differs across groupsor across situations, we adopted a multiple comparisonanalytic strategy instead of using a factorial analysis ofvariance. In particular, we used Dunnett's T3 procedure(Dunnett, 1980), cited by Wilcox (1985) as optimal forthe case of relatively small sample sizes and unequalvariances. The T3 procedure involves calculating confi-dence intervals (C.I.) for the difference between twomeans; if the C.I. excludes zero, the null hypothesis is

rejected. The family-wise type I error rate was held at.05 for the family of three comparisons (depressed ver-sus nondepressed in each situation) on total cognitivebias.

Total Cognitive Bias

Depressed subjects significantly exceeded the non-depressed group on total cognitive bias in the negativesituation (95% C.I. for the difference in means was 15.1± 14.3), but not in the neutral (-0.5 ± 7.4) or positive(5.4 ± 9.0) situation.

Specific Cognitive Biases

Given that the depressed group significantly ex-ceeded the nondepressed group on total cognitive biasin the negative situation, we explored the subsidiaryquestion of which specific cognitive bias categoriessignificantly distinguished the groups. For this purpose,we again used Dunnett 's T3 procedure to conduct afamily of six pairwise comparisons (depressed versusnondepressed in negative situation scores on the sixspecific cognitive bias categories), using a family-wisealpha level of .05.

The group difference in arbitrary inference was non-significant (95% C.I. for the difference in means was 0.5± 1.9), as were group differences in selective abstrac-tion (2.2 ± 5.1), overgeneralization (3.1 ± 4.5), magnifi-cation (2.9 ± 4.1), personalization (2.6 ± 4.9), and di-chotomous thinking (3.8 ± 4.7). Thus, the depressedgroup exceeded the nondepressed group in each cate-gory, and the difference in total scores was statisticallysignificant, but none of the differences on specific cate-gories attained significance. It is, therefore, not possi-ble in this sample to differentiate the types of bias anddraw conclusions as to whether some subset of themis particularly strongly associated with depression.

Discussion

Our major hypotheses were supported. Cognitive bi-ases were detected in unstructured, unconstrained re-ports of depressed patients ' cognitions in the ATSS

80 WHITE et al.

procedure. This result conceptually replicates the find-ings of Blackburn and Eunson (1989), who devised reli-able procedures for measuring cognitive bias in the self-monitored thoughts of depressed patients in cognitivetherapy (CT; Beck et al., 1979). The ATSS may be moreflexible than self-monitoring methods for some pur-poses, though, since it allows experimental control overstimulus situations. Self-monitored cognitions, of ne-cessity, involve subjects' responses to idiosyncratic, un-standardized situations.

Experimental control over stimulus situations wasuseful in this study, since it allowed us to corroboratethe view advanced by Krantz (1985; Krantz and Gall-agher-Thompson, 1990) that depression is associatedwith increased cognitive bias particularly in negativesituations, not in neutral or positive ones. The Krantzand Gallagher-Thompson (1990) study differed fromours in type of sample (all subjects were at least 50years old), type of control group (nonclinical), and typeof cognitive bias measure (a modified Cognitive BiasQuestionnaire). The similarity of results in spite ofmethodological variation suggests that situational spec-ificity of depressed-nondepressed differences in cogni-tive bias may be a robust finding.

Several methodological limitations to our studyshould be noted. First, our depressed sample included11 men and four women, whereas a truly representativesample of depressed patients would include a majorityof women (Nolen-Hoeksema, 1987). Replication isneeded to ensure the generalizability of our results, butit is instructive to note the parallel findings of Krantzand Gallagher-Thompson (1990), two-thirds of whosesubjects were women. Second, our sample size was

6 A reviewer of this manuscript argued that obtaining a significantdifference for total cognitive bias, but not for any of the specificcodes, suggests that the groups differed merely in general "negativ-ity," rather than cognitive bias, as depicted by Beck. The constructvalidity of our measures is not conclusively established by one study,but we differ with the critique for two main reasons. First, totalcognitive bias is a sum of the specific bias scores—all of them, andnothing but them. Therefore, we see no reason to ascribe to totalcognitive bias the property of measuring something (e.g., negativity)contrasting with what the specific categories measure (e.g., bias).

Second, it is important to bear in mind the distinction between"nonsignificant difference" and "no difference" (Hunter and Schmidt,1990; Rosnow and Rosenthal, 1989). We did not find that the groupswere equal on specific bias variables, only that group differenceswere not statistically significant. Indeed, all six categories showednonsignificant differences in the predicted direction. Under a nullhypothesis of no group differences on specific bias variables, thisconsistency of directional effect would itself be unlikely. If the sixcomparisons are considered (roughly) independent, then by the bino-mial formula, the p-value for six out of six "

successful" outcomes,each with a probability of .5, would be .5 6 , or < .02.

Taking these two considerations together, we believe that our inter-pretation (the depressed group showed more cognitive bias in thenegative simulated situation than did the nondepressed group; thiseffect is, by conventional standards, statistically reliable at our smallsample size if a composite measure is taken, but not if each codingcategory is considered separately) is more plausible than the alterna-tive suggested by the reviewer.

relatively small, which limits statistical power; this is-sue is particularly salient in considering our inability tosatisfactorily resolve the secondary question of whichspecific cognitive biases differentiated the groups inthe negative situation.' Third, the specificity of our re-sults to depression could have been more definitivelyevaluated if we had included a pure anxiety-disordercontrol group, rather than a heterogeneous psychiatriccontrol group, given the high comorbidity of depressionand anxiety disorders (e.g., Sanderson et al., 1990). Fi-nally, we did not conduct a formal manipulation checkon the perceived valence of simulated situations usedin the ATSS procedure. The specificity of group differ-ences in cognitive bias to the negative situation pro-vides prima facie evidence that the situations differedas intended, but the identification of each situation as"negative," "neutral," or "positive" remains largelybased on face validity.

Cognitive Assessment

These limitations notwithstanding, our findings wereencouraging from an assessment perspective. TheATSS method appeared to be sensitive to the hypothe-sized situational specificity of depressed-nondepresseddifferences in articulated thoughts. Similarly, an earlierstudy showed that an ATSS index of self-efficacy in astressful simulated situation, but not in a supportiveone, correlated inversely with speech anxiety (Davisonet al., 1991). Taken together, these findings suggest thatstressful simulated scenarios may provide a flexibletool for studying dysfunctional cognition in particulartypes of challenging situations.

If these results are supported by subsequent re-search, ATSS methods could prove very useful for test-ing additional hypotheses regarding cognitive theoryand therapy of depression. Consider one example.Treatment research has suggested that depression re-lapse is lower after CT than after pharmacotherapy(e.g., Simons et al., 1986), yet there is little evidence ofa specific cognitive effect of CT to account for thisdifferential relapse prevention (e.g., Simons et al.,1984). A recent review (Barber and DeRubeis, 1989)concluded that the most plausible model of the long-term effects of CT is that depressed patients learn com-pensatory skills for coping with negative automaticthoughts. Barber and DeRubeis identified several defi-ciencies in current measures of these coping skills, in-cluding possibly biased retrospective reporting, de-mand characteristics of endorsement measures, andthe nonstandard nature of stressors when subjects areallowed to respond to their own idiosyncratic situa-tions.

To circumvent these problems, they suggested thatan ideal measure of compensatory cognitive copingskills should: a) "provide a challenge—that is, some-

ARTICULATED THOUGHTS AND DEPRESSION 81

thing that requires compensation " (Barber and De-Rubeis, 1989, p. 452); b) "approximate an `on-line ' as-sessment of responses to situations and thoughts ratherthan rely on the subject's retrospective account of cop-ing efforts" (p. 453); c) "not ask the subjects to indicatewhich strategies they have been using, but rather sam-ple the thinking and planning of the respondent in thecontext of (hypothetical) stressful events, enabling theresearcher to classify the subject 's responses " (p. 453);and d) allow the researcher to "have some control overthe variety of stressors assessed by the measure" (p.453). We submit that these desiderata are matchedclosely by the ATSS paradigm, which may, therefore,merit attention from CT process-outcome reseachers.

Conclusions

We found that depressed patients exceeded nonde-pressed psychiatric patients in an articulated thoughtsindex of cognitive bias in a negative simulated situation.Thus, our results confirmed that situation-specific cog-nitive biases may be detected in relatively unstructuredsamples of depressives ' cognitions and do not reflectmerely their willingness to endorse preselected nega-tive responses. Finally, this research provided furtherevidence for the utility of the ATSS paradigm and ex-tended its applicability to the measurement of de-pressive thinking, an area in which advances in cogni-tive assessment are needed.

References

Alloy LB, Abramson LY (1988) Depressive realism: Four theoreticalperspectives. In LB Alloy (Ed), Cognitive processes in depression(pp. 223–265). New York: Guilford.

American Psychiatric Association (1980) Diagnostic and statisticalmanual of mental disorders (3rd ed). Washington, DC: Author.

Barber JP, DeRubeis RJ (1989) On second thought: Where the actionis in cognitive therapy for depression. Cognit Ther Res 13:441–457.

Beck AT (1963) Thinking and depression: I. Idiosyncratic content andcognitive distortions. Arch Gen Psychiatry 9:324-333.

Beck AT (1976) Cognitive therapy and the emotional disorders. NewYork: International Universities Press.

Beck AT (1987) Cognitive models of depression. J Cognit Psychother1:5-37.

Beck AT, Rush AJ, Shaw BF, Emery G (1979) Cognitive therapy ofdepression. New York: Guilford.

Beck AT, Steer RA, Garbin MG (1988) Psychometric properties ofthe Beck Depression Inventory: Twenty-five years of evaluation.Clin Psychol Rev 8:77–100.

Blackburn IM, Eunson KM (1989) A content analysis of thoughtsand emotions elicited from depressed patients during cognitivetherapy. Br JMed Psychol 62:23-33.

Coyne JC (1989) Thinking postcognitively about depression. In AFreeman, KM Simon, LE Beutler, H Arkowitz (Eds), Comprehen-

sive handbook of cognitive therapy (pp. 227–244). New York: Ple-num.

Coyne JC, Gotlib IH (1983) The role of cognition in depression: Acritical appraisal. Psychol Bull 94:472–505.

Davison GC, Feldman PM, Osborn CE (1984) Articulated thoughts,irrational beliefs, and fear of negative evaluation. Cognit Ther Res8:349–362.

Davison GC, Haaga DAF, Rosenbaum J, et al (1991) Assessment ofself-efficacy in articulated thoughts: "States of Mind" analysis andassociation with speech anxious behavior. J Cognit Psychother5:83–92

Davison GC, Robins C, Johnson MK (1983) Articulated thoughts dur-ing simulated situations: A paradigm for studying cognition in emo-tion and behavior. Cognit Ther Res 7:17–40.

Davison GC, Zighelboim V (1987) Irrational beliefs in the articulatedthoughts of college students with social anxiety. JRational-Emo-tive Ther 5:238-254.

Dobson KS, Shaw BF (1986) Cognitive assessment with major de-pressive disorders. Cognit Ther Res 10:13–29.

Dunnett CW (1980) Pairwise multiple comparisons in the unequalvariance case. J Am Slat Assoc 75:796–800.

Haaga DAF (1989) Articulated thoughts and endorsement proceduresfor cognitive assessment in the prediction of smoking relapse. JConsult Clin Psychol 1:112–117.

Haaga DAF, Dyck MJ, Ernst D (1991) Empirical status of cognitivetheory of depression. Psychol Bull 110:215-236.

Haley GMT, Fine S, Marriage K, et al (1985) Cognitive bias and depres-sion in psychiatrically disturbed children and adolescents. J Con-sult Clin Psychol 53:535-537.

Hunter JE, Schmidt FL (1990) Methods of meta-analysis: Correctingerror and bias in research findings. Newbury Park, CA: Sage.

Krantz SE (1985) When depressive cognitions reflect negative reali-ties. Cognit Ther Res 9:595-610.

Krantz S, Hammen C (1979) Assessment of cognitive bias in depres-sion. J Abnorm Psychol 88:611–619.

Krantz SE, Gallagher-Thompson D (1990) Depression and informa-tion valence influence depressive cognition. Cognit Ther Res14:95-108.

Kruglanski AW, Jaffe Y (1988) Curing by knowing: The epistemicapproach to cognitive therapy. In LY Abramson (Ed), Social cogni-tion and clinical psychology: A synthesis (pp. 254–291). New York:Guilford.

Nolen-Hoeksema S (1987) Sex differences in unipolar depression:Evidence and theory. Psychol Bull 101:259–282.

Novick MR, Lewis C (1967) Coefficient alpha and the reliability ofcomposite measurements. Psychometrika 32:1–13.

Rosnow RL, Rosenthal R (1989) Statistical procedures and the justifi-cation of knowledge in psychological science. Am Psychol44:1276-1284.

Sanderson WC, Beck AT, Beck J (1990) Syndrome comorbidity inpatients with major depression or dysthymia: Prevalence and tem-poral relationships. Am J Psychiatry 147:1025-1028.

Simons AD, Garfield SL, Murphy GE (1984) The process of changein cognitive therapy and pharmacotherapy for depression. ArchGen Psychiatry 41:45-51.

Simons AD, Murphy GE, Levine JL, Wetzel RD (1986) Cognitive ther-apy and pharmacotherapy for depression: Sustained improvementover one year. Arch Gen Psychiatry 43:43–48.

Weiner B (1985) "Spontaneous" causal thinking. Psychol Bull 97:74-84.

Wilcox RR (1985) Advances in basic statistics: Vol II. Basic hypothe-sis-testing procedures. Los Angeles: UCLA Center for the Study ofEvaluation.

Wilkinson IM, Blackburn IM (1981) Cognitive style in depressed andrecovered depressed patients. Br J Clin Psychol 20:283-292.