THINKING, FEELING AND DOING IN DEPRESSION: THE EFFECT OF RUMINATION ON INFORMATION-PR OCESSING AND PHYSIOLOGICAL RESPONSE By PAUL J. SEIGNOUREL A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2007
c 2007 Paul Seignourel
iii TABLE OF CONTENTS page LIST OF TABLES.............................................................................................................vi LIST OF FIGURES..........................................................................................................vii ABSTRACT.....................................................................................................................vi ii CHAPTER 1 GENERAL INTRODUCTION....................................................................................1 2 ENHANCED EXPLICIT MEMORY FOR NEGATIVE WORDS IN DEPRESSION: EFFICIENT PROCE SSING OR RESPONSE BIAS?.......................5 Introduction................................................................................................................... 5 Method........................................................................................................................1 2 Overview.............................................................................................................12 Participants..........................................................................................................12 Materials..............................................................................................................14 Session 1..............................................................................................................15 Semi-structured interview............................................................................15 Sorting of adjectives.....................................................................................15 Session 2..............................................................................................................16 Presentation of test items..............................................................................16 Recognition testing.......................................................................................16 Questionnaires.....................................................................................................17 Computation of SDT Measures...........................................................................18 Results........................................................................................................................ .18 ParticipantsÂ’ Characteristics................................................................................18 Recognition Performance....................................................................................19 Self-descriptive adjectives............................................................................20 Non-descriptiv e adjectives...........................................................................21 Adjective Selection..............................................................................................21 Relationship Between Adjective Characteristics and Recognition.....................22 Relationship Between Recognition Performance and Questionnaires................23 Discussion...................................................................................................................23 Information-Processing and Response Bias........................................................27 Additional Findings.............................................................................................30 Implications for the Cognitive Model and Treatment.........................................32
iv Limitations...........................................................................................................33 Conclusions.........................................................................................................36 3 WHEN GOOD TURNS INTO BAD: DELAYED STARTLE POTENTIATION BY POSITIVE STIMULI IN CLINICAL DEPRESSION.........................................38 Introduction.................................................................................................................38 Method........................................................................................................................4 5 Overview.............................................................................................................45 Participants..........................................................................................................45 Materials..............................................................................................................47 Session 1..............................................................................................................48 Semi-structured interview............................................................................48 Sorting of adjectives.....................................................................................48 Startle Task (Session 2).......................................................................................49 Questionnaires.....................................................................................................50 Data Reduction....................................................................................................51 Data Analyses......................................................................................................51 Results........................................................................................................................ .52 ParticipantsÂ’ Characteristics................................................................................52 Blank Startle........................................................................................................53 Emotion-Modulated Startle.................................................................................53 Number of discarded trials...........................................................................53 Latency to peak............................................................................................54 Amplitude.....................................................................................................54 Relationship with Questionnaires........................................................................54 Effects of Depression Severity, Nu mber of Episodes and Medications..............56 Discussion...................................................................................................................57 4 SUBLIMINAL AND SUPRALIMINAL PRIMING OF EMOTIONAL WORDS IN DEPRESSION.......................................................................................................64 Introduction.................................................................................................................64 Information Processing Biases in Depression.....................................................65 Automatic and Controlled Processing.................................................................66 Method........................................................................................................................6 9 Overview.............................................................................................................69 Participants..........................................................................................................69 Semi-Structured Interview...................................................................................71 Questionnaires.....................................................................................................71 Priming Task.......................................................................................................72 Materials.......................................................................................................72 Lexical decision task....................................................................................73 Awareness checks........................................................................................75 Data Reduction and Analyses..............................................................................76 Results........................................................................................................................ .77 ParticipantsÂ’ Characteristics................................................................................77
v Awareness Check................................................................................................78 Priming Task.......................................................................................................79 Preliminary analyses....................................................................................79 Effects of group and valence on priming.....................................................80 Relationship with Self-Report Measures......................................................80 Discussion...................................................................................................................80 5 GENERAL DISCUSSION.........................................................................................85 LIST OF REFERENCES...................................................................................................88 BIOGRAPHICAL SKETCH.............................................................................................98
vi LIST OF TABLES Table page 2-1. Adjective characteristics by category........................................................................14 2-2. Demographic characteristics an d questionnaire scores by group..............................19 2-3. Recognition hit rates, false positive rates, discriminability and response bias by group and adjective category....................................................................................19 2-4. Characteristics of selected adjectives by group.........................................................22 3-1. Adjective characteristics by category........................................................................47 3-2. Demographic characteristics an d questionnaire scores by group..............................53 3-3. Characteristics of the basic startle response by group................................................53 3-4. Latency and amplitude (T-score) of startle response by group and condition..........54 3-5. Spearman correlations between startle amplitude (T-scores) and scores on the questionnaires within the depressed group...............................................................56 4-1. Demographic characteristics an d questionnaire scores by group..............................77 4-2. Error rates for the awareness checks by group..........................................................78 4-3. RT by group and stimulus category...........................................................................79 4-4. Subliminal and supraliminal priming by group and category...................................80 4-5. Effect sizes and confidence intervals of the difference between the depressed and the control groups in Bradley et al. (1995)...............................................................82
vii LIST OF FIGURES Figure page 2-1. Recognition discriminability and response bias for the self-descriptive adjectives as a function of group...............................................................................................20 4-1. Trial structure in th e lexical decision task..................................................................74
viii Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy THINKING, FEELING AND DOING IN DEPRESSION: THE EFFECT OF RUMINATION ON INFORMATION-PR OCESSING AND PHYSIOLOGICAL RESPONSE By Paul J. Seignourel August 2007 Chair: Russell M. Bauer Major: Psychology Since the advent of the cognitive mo del of mood disorders, research on information-processing biases in depressi on has flourished, yieldi ng important insights into the cognitive biases shown by depressed individuals. In the following studies, we explore several questions pertaining to the na ture of information-pr ocessing biases and physiological responsiveness in depression. Our sample consisted of 26 clinically depressed and 22 never-depre ssed individuals who particip ated in two experimental sessions. One of our goals was to determine whether the explicit memory biases for negative, self-descriptive information observe d in depression could be better accounted for by better processing, response bias or sel ection bias. Using rec ognition testing and a signal detection theory approach, we observe d an interaction betw een group and valence only for response bias, such that depressed individuals showed greater endorsement of negative, self-descriptive adjectives, while the two groups did not differ for nondescriptive adjectives. A s econd goal was to examine physio logical reactivity using the
ix startle eyeblink methodology. We found that de pressed individuals showed potentiation rather than inhibition of startle after viewi ng positive adjectives. This effect was observed only after a delay, and only within the depressed group. Within the depressed group, startle potentiation at the long delay was related to rumination and the number of depressive episodes. A third goal of the st udy was to replicate previous findings of increased subliminal and supraliminal priming for negative words in depression. Although the effects we observed were in the predicted direction, they failed to reach significance. Taken together, these findings s uggest that depression is characterized by several abnormalities in the processing of and physiological responsiveness to negative and positive material. In particular, they s uggest that depressed individuals may show preferential processing of negative material, co mbined with delayed negative reactions to positive material. Such biases are likely to contribute to the development, maintenance and worsening of depressive episodes.
1 CHAPTER 1 GENERAL INTRODUCTION Depression, one of the most common psychol ogical disorders in the United States and other countries, is a complex condition that manifests in a variety of physiological, cognitive, and behavioral symptoms. Alt hough dysphoria or anhedonia may represent its most prominent features (at l east one of these two symptoms is required to establish a diagnosis), increased or decr eased appetite, increased or decreased sleep, low energy, poor concentration, psychomotor retardat ion or agitation, feelings of guilt or worthlessness and suicidal ideation are other common symptoms experienced by depressed individuals (American Psychiat ric Association, 2000). Major depression, which requires the presence of at least four of these symptoms for a period of at least two weeks, has been recognized as a serious heal th problem in the United States, associated with decreased functioning and satisfaction (Wells et al., 1989), increased rates of medical illness and increased utilization of medical services (Katon & Sullivan, 1990). Life-time prevalence ranges from 10% to 20% (Patten, 2003) and wo men are at greater risk than men, a finding that appears to generalize to most developed and some developing countries (Culbertson, 1997). Corre sponding to the great variety of symptoms that can contribute to depression, theories a nd treatments of depre ssion have originated from a broad range of distinct approaches, from neurobiological conceptualizations to cognitive, behavioral or psyc hosocial theories. Biological th eories of depression have begun to delineate abnormalities in brain neurotransmitter systems and localized brain regions (Davidson, Pizzagalli, Nitschke & Putnam, 2002; Drevets, 2000), while
2 behavioral and cognitive theories have emphasized changes in overt behavior and cognition, respectively. In parallel to thes e theories, independent pharmacological and psychosocial treatments of depression ha ve been developed. Pharmacological and psychosocial approaches appear similarly e ffective in treating depression, and their combination is more effective than either treatment alone (Casacalenda, Perry, & Looper, 2002; Keller et al., 2000). One of the most well-validated models of depression is the cognitive model, developed by Aaron Beck (1967). According to this model, depressed individuals display systematic negative biases in their perception and interpretation of life events. In particular, they are more likely to engage in sustained processing of negative events, and to interpret life events negatively or pessimistically. The cognitive model has been closely associated with the development of cognitive therapy (CT), which has as its main goals to challenge and rectify the negativ e biases and interpretations displayed by depressed individuals. Within th e last 20 years, CT has been firmly established as an effective and powerful treatment of depr ession, comparable in its efficacy to pharmacological treatment and other empiri cally-supported psychosocial treatments of depression such as interpersonal therapy and behavior therapy (Dobson, 1989; Gloaguen, Cottraux, Cucherat, & Blackburn, 1998). Mo reover, the addition of CT to pharmacological treatment has been shown to reduce relapse rates (Gloaguen et al., 1998), and, in at least one large, multi-cente r treatment study of chronically depressed individuals, the combination of pharmacologi cal treatment with the cognitive-behavioral system of psychotherapy, which includes elements of CT, problem-solving and
3 interpersonal therapy, was more effective th an pharmacotherapy alone (Keller et al., 2000). The cognitive model of depression pr edicts that dysfunctional beliefs and attributions interact with life events to precipitate depressi ve episodes (Beck, 1976). Thus, dysfunctional beliefs and attributions constitute a vulnerability to depression, which can remain dormant as long as life even ts do not contribute to increased stress. As a prototypical example of these dysfunctional at titudes, derived from learned helplessness theory, it has been shown that depressed indivi duals tend to interpret negative events in a more stable (i.e., which will remain constant over time), global (which applies to many situations) and internal (whi ch constitutes a flaw within the person) manner than do healthy individuals (Sweeney, Anderson, & Bailey, 1986). There is also support for the view that depression is associated with nega tive beliefs about the self, the world and the future (Beck, 1976). In terms of information-processing bias es, numerous studies have reported that depressed individuals show greater recall of negative adjectives under self-referent conditions (Mineka & Sutton, 1992; Willia ms, Watts, MacLeod, & Mathews, 1988). These findings are consistent with the hypothe sis of a general processing bias towards negative and schema-congruent information in depression. However, not all measures of information-processing have suggested robust ne gative biases in depression. For instance, most studies failed to show any bias for m easures of implicit memory in clinical and subclinical depression (but see Watkins, Vache, Verney, Muller, & Mathews, 1996, for an exception and conceptual discussion). Si milarly, most studies failed to show any attentional biases towards negative information (Williams et al., 1988). Thus, the exact
4 nature of information-processing biases in depression has been the subject of much controversy. In the following chapters, we explore three unanswere d questions pertaining to this domain. The first question concerns the nature of explicit memory biases in depression, which have generally been view ed as evidence of better processing of negative, self-descriptive information in that population. We examine whether these biases are better accounted for by better proces sing, response bias, or selection bias. The second question pertains to phys iological responsiveness to emotional stimuli. Using the methodology of the startle eyeblink reflex, pr evious studies have found abnormal startle modulation by affective stimuli in severely de pressed individuals, but not in mildly or moderately depressed individuals. We examin e whether the failure to find differential physiological responsiveness in mild and m oderate depression might be overcome by using depression-relevant stimuli and allowi ng sufficient processing time after stimulus offset. The third question concerns the pr esence of negative biases in subliminal repetition priming. We attempt to replicate previous findings of greater priming for negative adjectives, while addr essing some of the methodolog ical limitations of previous studies. In all three experiments presented be low, all the subjects belong to the same subject pool. The number of subjects for each ex periment varies due to the presence of missing or unusable data, and the fact that so me of the subjects did not return for the second session.
5 CHAPTER 2 ENHANCED EXPLICIT MEMORY FOR NEGATIVE WORDS IN DEPRESSION: EFFICIENT PROCESSING OR RESPONSE BIAS? Previous research using r ecall paradigms suggests that depressed individuals show better memory for negative, self-descriptive information. We examined other potential explanations for these findings, including re sponse bias and selection bias. Twenty-four depressed and 22 never-depressed participants completed a recognition memory test with negative and positive adjectives individually rated as self-descriptive. We used signal detection theory to compute discriminability (dÂ’) and response bias (c). There was a significant interaction between grou p and valence for response bias, F (1, 42) = 4.37, p < .05, with depressed participants showing great er bias than control participants for negative adjectives, while the two groups did not differ for positive adjectives. Depressed participants showed lower discrimina bility than control participants, F (1, 42) = 7.29, p < .05, regardless of valence. Th ese results could not be accounted for by differential selection of self-descriptive adjectives. Ou r results question previous assertions that depressed individuals show be tter memory for negative, self-descriptive information, and suggest response bias as an alternate explanation. Therap ists working with depressed individuals should be alert to the possibi lity of distorted or inaccurate memory, particularly regarding negative information. Introduction One of the most robust findings on informa tion-processing biases in depression is that depressed individuals show better recal l of negative, self-descriptive adjectives,
6 while non-depressed individuals show better recall of positive, self-descriptive adjectives (Mineka & Sutton, 1992; Williams, Watts MacLeod, & Mathews, 1988). This phenomenon was first established by Derry a nd Kuiper (1981). In their study, they asked depressed and non-depressed participants to decide whether negative and positive adjectives described them, and later submitte d them to unexpected recall testing. Their findings have been interpreted as evidence of better memory for negative, self-descriptive words in depressed individuals, and have played an important role in the development of an experimental model of inform ation-processing in depression. Many researchers have used BeckÂ’s cognitive model as a framework for interpreting such results. Beck (1967; 1976) contended that depressed individuals possess a negative self-schema, which serves as a powerful filter for organizing their perceptions, interpretations, beliefs, and memory. Bowe rÂ’s network model (1981) of mood influences on information-processing has provided furt her support for this idea. According to network theory, the experience of a certain mood automatically activates stored representations of stimu li that have been prev iously associated with that particular mood. For instance, a person who has been attacked by a dog will be more likely to think of a dog if he or she subsequently experiences fear. In the case of depression, the negative mood experienced by clinically depressed individuals makes them more likely to remember negative events or negative words. Teasdale ( 1983) emphasized the potential implications of the negative memory bias in depression. He noted that, since depressed individuals show a bias towards rememb ering negative events, they may develop negative expectations for adap tive coping behaviors (e.g., going to a social gathering). As
7 a result, they may tend to avoid such adaptiv e behaviors, leading to the maintenance of the depressed state. Another way to conceptualize findings of better recall for nega tive information in depression is to place them in the more genera l context of the Â“self-r eference effect.Â” The self-reference effect was first described by Rogers, Kuiper and Kirker (1977) as the improvement in memory performance resulti ng from processing information in reference to the self (i.e., Â“Does this word describe you?Â”), as compared to general semantic processing (e.g., Â“Does this word mean the same as X?Â”), or processing in reference to someone else (e.g., Â“Does this word describe Y?Â”). Since the arti cle by Rogers et al., several studies have replicat ed the self-reference effect and described some of its moderators. A meta-analysis (Symons & Johns on, 1997) suggests th at, in the general population, the self-reference effect is mo re robust for positive than for negative information. These findings are consistent with the view that most people possess a robust positive self-concept, which facilita tes the encoding and retrieval of positive information stored in reference to the se lf. On the contrary, BeckÂ’s theory (1967) suggests that clinically depressed individuals possess a negative self-c oncept. Thus, it is not surprising that previous research found th e self-reference effect to be reversed in depression, with better processing of negativ e, self-descriptive than positive, selfdescriptive information. One problem with the above-mentioned concep tual frameworks is that they rely on the assumption that the greater recall for negative, self-des criptive information observed in depressed individuals means better memor y. In general, however, greater recall and better memory cannot be automatically equa ted. According to th e generate-recognize
8 model, free recall proceeds in two distinct stages: first, the generation of potential candidates for recall, and, second, the evaluati on of each of these candidates as belonging to the actual learning list (Anderson & Bowe r, 1972). Although the traditional generaterecognize model has been criticized for its in ability to account for encoding specificity effects (Thomson & Tulving, 1970; Tulving & Thomson, 1973), many modern theories of recall also assume distinct stages of retrieval and monitoring for recall (Koriat & Goldsmith, 1996; Whittlesea, 1997). Thus, great er recall could result from better generation of test items (i.e., better memo ry), or from a more liberal approach to monitoring. One way to distinguish Â“trueÂ” memory e ffects from a more liberal approach to monitoring is to apply signal detection theo ry (SDT). In SDT, yes-no judgments are assumed to be a combination of (a) the am ount of information favoring a particular response, and (b) the criterion used to decide how much evidence is needed to make a particular response. The former is referred to as discriminability, while the latter is referred to as response bias (McNicol, 1972). In memory testing, discriminability refers to the amount of evidence suggesting that a particular item belonge d to the original learning situation, while response bias is th e more or less liberal criterion used by a participant for deciding whether or not to endorse each item. In general memory research, SDT has become a standard methodological re quirement (Fiedler, Ni ckel, Muehlfriedel, & Unkelbach, 2001). Its use in research on the effects of mood or depression on memory for emotional stimuli, however, has been very limited. One exception is the study by Zuroff, Colussy and Wielgus (1983), who examin ed memory for emotional adjectives in depressed and non-depressed female college students using SDT. They found that the
9 increased recognition for negative words obs erved in the depressed group corresponded to greater response bias rather than greater discriminability, calling into question the view that depression is associated with better memory for negative words. Their study, however, suffered from an important limitatio n: during recognition testing, test items were adjectives that had been individually selected as self-descriptive by each participant, but foils were additional adjec tives that had not been select ed for self-descriptiveness. This important difference between the two se ts of adjectives makes it impossible to determine whether findings in Zuroff et al. are specific to self-descriptive adjectives. Moreover, it raises the possibility that partic ipants may have used self-descriptiveness as a source of information when completing r ecognition testing. Group differences in the use of such a strategy, rather than memory biases, may have been responsible for the observed results. It should also be noted that participants in Zuroff et al. were female college students who scored above a certain cutoff on a self-report measure of depression symptomatology. It is unclear whether their results would generalize to clinically depressed individuals. Dunbar and Lishman (1984) also used a SDT approach to memory biases in depression. Contrary to Zuroff et al. (1983), they found an interaction between valence and discriminability, such that the depressed group showed greater discriminability for negative words, while the control group s howed greater discriminability for positive words. The depressed group also showed a response bias against endorsing positive words. The study by Dunbar and Lishman, howev er, also had important limitations. In their sample, which consisted entirely of inpatients, the procedure used to diagnose depression was not specified, and no attemp t was made to control for potentially
10 etiologically-related comorbid anxiety disorder s. Moreover, the stimuli they used were not especially relevant to depression. For in stance, Â“cheeseÂ” and Â“lawnÂ” belonged to the set of positive words, and Â“snakeÂ” and Â“mobÂ” belonged to the set of negative words. Dunbar and LishmanÂ’s choice of stimuli may have had an influence on their results, as numerous studies have shown the importan ce of using depression -relevant words when assessing information-processing biases in th at population (for a review, see Williams et al., 1988). Fiedler et al. (2001), who administered a recognition test for emotional words while manipulating participantsÂ’ mood using emotional films, also reported results contrary to those of Zuroff et al. (1983). They found robust mood congruency effects, and showed that these effects were due to incr eased mood-congruent disc riminability, rather than response bias. This result was replic ated in two experiments, and confirmed by analyses of reaction times. Fiedler et al. concluded from these findings that moodcongruent effects reflect a Â“trueÂ” memory advantage for mood-congruent information, rather than the use of heuristic judgments independent of memory performance per se. Given these contradictory results, the present study was designed to determine whether memory biases in clinical depressi on are due to a true memory advantage for negative, self-descriptive information, or to the adoption of a more liberal criterion for reporting such information. In SDT terms, th e two possibilities co rrespond to greater discriminability and lower response bias, respectively. Of course, the SDT framework does not require that the respons e be one or the other. Rath er, both could be present to varying degrees.
11 Both hit rates and false positive rates are required for the computation of SDT indices. When one needs to di stinguish between different cat egories of adjectives, recall testing makes it difficult to compute false positiv e rates. In the case of memory biases in depression, this difficulty is compounded by the need to consider two distinct dimensions of adjective category: emotiona lity and self-descrip tiveness. Like Zuroff et al. (1983) and Dunbar and Lishman (1984), we therefore used recognition testing for our study. In order to avoid the confound present in Zuroff et al ., we decided to have participants sort adjectives for self-d escriptiveness in a preliminary session, from which both test items and foils were selected. Another important methodological issue th at has rarely been mentioned in the literature on memory biases in depression is the possibility that depressed and nondepressed individuals would show systematic differences in the selection of selfdescriptive adjectives. In pa rticular, one would expect, gi ven the presence of a negative self-schema in depression, that depressed i ndividuals would select as self-descriptive adjectives that are more robus tly negative than those selected by control pa rticipants, and as non-descriptive adjectives th at are more positive. In tur n, these selection biases could be entirely responsible for the observed memory differences between the two groups. Zuroff et al. (1983) mentioned this possibility and noted that, in their study, depressed and non-depressed participants indeed varied in their selection of self-descriptive adjectives. However, they did not examine the influence of this selection bias on memory performance. The goal of our study was to examine the factors contributing to memory biases for negative, self-descriptive adjectives in clin ical depression. Three factors were
12 considered (1) a Â“trueÂ” memory advantage, corresponding to greater discriminability, (2) a more liberal criterion for reporting such adjectives, co rresponding to lower response bias, and (3) a spurious effect of group differences in the selection of self-descriptive adjectives. Given the limitations and inconsiste nt findings of previous research, we did not make any specific predictions regarding th e respective contributions of the first two factors. Regarding the third f actor, we predicted that depre ssed individuals would select negative, self-descriptive adjec tives that are more strongly negative than those selected by control participants (e.g., desperate vs. bored ) and positive, self-descriptive adjectives that are less strongly positive than those select ed by control participants (e.g., thankful vs. enthusiastic). However, we did not expect these selection biases to fully account for differences in memory performance, since previous studies whose validity was not threatened by this possible confound also show ed better recall for ne gative adjectives in depression, particularly unde r self-referent encoding conditions (e.g., see Bradley & Mathews, 1983). Method Overview Depressed and control particip ants took part in two experi mental sessions separated by one to three weeks. During the first se ssion, they selected negative and positive adjectives as self-descrip tive or non-descrip tive of themselves. During the second session, they were presented with test ite ms from each category, and subsequently completed a recognition test involving all test items and an equal number of foils. Participants Participants were recruited through f liers posted within the community and advertisements in local newspapers. All pa rticipants were betw een 18 and 60 years old.
13 Participants within the depr essed group all met diagnostic criteria for a current Major Depression Disorder (MDD), w ithout Psychotic Features, ac cording to the Diagnostic and Statistical Manual of Mental Disorder s, Fourth Edition (American Psychiatric Association, 2000). Exclusion cr iteria for the depressed group included (a) A history of psychotic disorder, bipolar disorder, or obs essive-compulsive disorder (OCD), (b) Any history of neurological disorder such as epilepsy or trauma tic brain injury (c) A recent history of substance abuse or dependence (less than six mont hs), (d) A current diagnosis of post-traumatic stress disorder (PTSD), (f) A history of recent suicide attempt (less than six months), or plans or intent to commit suic ide, (g) Use of medications known to affect cognitive functioning, such as benzodiazepines or certain anti-psychotic medications. For the control group, all exclusion criteria appl ied and, in addition, e ligible subjects were required to have no history of mood disorder or anxiety diso rder, and a score lower than 13 on the Beck Depression Invent ory Â– Second Edition (BDI-II). Depressed individuals who were taking an ti-depressant medications were included in the study as long as they had been on a st able dosage of medicat ion for at least six weeks. Given the high comorbidity between depression and anxiety disorders (Hasin, Goodwin, Stinson, & Grant, 2005), we also incl uded depressed individu als with a current comorbid anxiety disorder (except for OCD a nd PTSD) or with a history of PTSD in full remission, as long as it was determined th at the anxiety disorder was not a major contributor to the current majo r depressive episode. In cases of diagnostic uncertainties, consultation with a licensed psychologist was sought, and a generally conservative approach was taken to part icipantsÂ’ inclusion. The init ial sample consisted of 26 depressed and 26 control participants. Two pa rticipants from each group did not return
14 for the second session. In addition, three of th e control participants were excluded based on a high BDI-II score, and another based on a current diagnosis of specific phobia. The final sample consisted of 24 depressed and 20 control participants. Materials Sixty negative and sixty positive adjectives were selected from a large pool of emotional adjectives collected from prior st udies (Bradley & Math ews, 1983; Bradley & Lang, 1999; Scott, Mogg, & Bradley, 2001). Ten nave judges (5 men, 5 women; 9 Caucasian, 1 African-American) rated each adje ctive on two separate 9-point scales for valence and arousal, respectively. Valence ra nged from Happy (1) to Unhappy (9), and arousal ranged from Calm (1) to Excite d (9) (see Bradley & Lang, 1999, for a complete set of instructions). The negative and positive ad jectives differed significantly in terms of valence, but not arousal (see Table 2-1). Wo rd length (i.e., number of characters) and frequency were also computed for each adje ctive. Frequency was computed using norms from Kucera and Francis (1967), and log-transf ormed to minimize the influence of very large scores. The negative and positive adjectiv es did not differ in terms of length. They did, however, differ in terms of frequency, with the positive adjectives being more frequent than the negative adjectives (see Table 2-1). Table 2-1. Adjective characteristics by category Negative Adjectives Positive Adjectives Comparison Min/Max M (SD) Min/Max M (SD) Length 3/13 7.4 (2.2) 3/13 7.3 (2.5) t (118) = 0.08 Frequency 2/119 19.9 (27.1) 1/313 49.1 (63.3) t (118) = -3.27* Ln(Frequency) 0.69/5.29 2.5 (1.0) 0.00/5.57 3.3 (1.2) t (118) = -3.65** Valence 6.30/8.00 7.1 (0.4) 1.50/3.70 2.7 (0.5) t (118) = 51.33** p < .01; ** p < .001.
15 Session 1 Semi-structured interview Participants completed a semi-structured interview that served as a basis for all Axis I diagnoses. All interviews were conducted by the first author using the Mood Disorders, Psychotic Screen, SubstanceRelated Disorders and Anxiety Disorders modules of the Structured Clinical Inte rview for the DSM-IV-TR (SCID-IV, First, Spitzer, Gibbon, & Williams, 1996). Interviews we re audiotaped, and 10 interviews from the depressed group and 8 interv iews from the control group were randomly selected and recoded by the second author, blinded to orig inal ratings. Inter-rater agreement was 98% for individual SCID-IV items, and 95% for Axis I diagnoses. The less-than-perfect agreement for diagnoses was due to one depr essed participant for whom the second, but not the first author, diagnosed a distant history of alcohol ab use. There was perfect interrater agreement regarding MDD severity (3 mild, 7 moderate). Sorting of adjectives Participants were presented with a set of 60 cards with the negative or positive adjectives printed on top. They were aske d to put each card in one of four groups: The least like me Less like me More like me and The most like me After an initial sorting, they were instructed, if necessary, to adjust their sorting in order to have at least 12 adjectives per group. Each participant sorted negative and positive adjectives separately. Valence order was alternated among participan ts, and the cards within each valence were presented in fixed random order. From each participantÂ’s sorting, 12 adjectiv es were randomly selected from the two most extreme groups ( The least like me and The most like me ) for both negative and positive adjectives. Thus, there were four categories of individually selected adjectives
16 for each participant: Negative Self-descriptive Negative Non-descriptive Positive Selfdescriptive and Positive Non-descriptive For the second session, six adjectives from each category were randomly selected to serve as test items, while the six others served as foils during recognition. Session 2 Presentation of test items Test items included the 6 individually sele cted adjectives from each category and an additional 12 neutral words (household items), for a total of 36 test items. Participants were not told that they would be tested for recognition during this phase of the experiment. Each test item was presented fo r 6000ms in the center of a black computer screen. After a 2000ms interval, a prompting scr een asked participants to rate how often (Rarely, Sometimes, or Often) they used that word or concept. Part icipants provided their responses verbally, and their responses we re manually recorded by the experimenter. Trials were separated by either a short (6s) or a long (26s) delay.1 Two items that were not included in the recognition task served as training items. Each test item was presented twice, and the delays (short or long) after each item were a lternated across presentations. Eyeblink startle responses and skin conductan ce were collected during this phase of the experiment. Physiological recordings are not the focus of the present paper and will be reported in a separate paper. Recognition testing After a 30-minute delay, during which partic ipants completed some questionnaires and neuropsychological instruments, test items and an equal number of foils per category 1 The delay manipulation was introduced to examine physiological reactivity at various stages of processing.
17 were presented verbally by the experimenter. Pa rticipants were asked to indicate whether they had seen each item during the earlier session when physiological recordings took place. They were encouraged to provide a Â“yes Â” or Â“noÂ” response and to give their best guess if they were unsure of the correct response. Questionnaires In each session, participants completed the BDI-II, the Beck Anxiety Inventory (BAI), and the Rumination Responses Scale (R RS). The BDI-II is a widely used measure of depression symptomatology, with internal consistency and test-retest reliability both above .9 (Beck, Steer, & Brown, 1996). The BA I is a 21-item self-re port measure of anxiety symptoms, which was developed to a ddress the problem of the overlap between other measures of anxiety and measures of depression (Beck & St eer, 1990). The scale has good internal consistency ( .92) and test-retest reliability (.75), a nd its convergent and discriminant validity are supporte d by higher scores in patients diagnosed with an anxiety disorder than patients diagnosed with ma jor depression (Beck, Epstein, Brown, & Steer, 1988). The RRS is a 22-item self-report instru ment assessing the frequency of thoughts about oneÂ’s symptoms of depression. Its internal consistency ranges from .89 to .90 (Nolen-Hoeksema & Morrow, 1991; Treynor Gonzalez, & Nolen-Hoeksema, 2003; Wenzlaff & Luxton, 2003) and its test-retest re liability was .67 in a large community sample (Nolen-Hoeksema, Larson, & Grays on, 1999). It has been shown to predict depressive symptomatology prospectively af ter traumatic events (Nolen-Hoeksema & Morrow, 1991) and death of a loved one (Nolen-Hoeksema, Parker, & Larson, 1994). In our sample, the test-retest reliabilities of these three instruments were as follows: for the BDI-II, .67 for the depressed group, and .79 for the control group; for the BAI, .73 for the depressed group and .90 for the cont rol group; and for the RRS, .75 for the
18 depressed group and .73 for the control group (all significant at the .01 level). Since recognition testing took place during the second session, scores obtained during the second session were used for subsequent analyses. Computation of SDT Measures We computed discriminability (dÂ’) and bi as (c) for each part icipant and category using formulas from McNicol (1972). Some pa rticipants from both groups had perfect hit rates (100%) or false positive rates (0%) fo r some of the categories, which creates difficulties for the computation of dÂ’ and c. Following recommendations from McNicol, perfect hit rates and false positive rates were adjusted to a half-item error rate (i.e., 92% hit rate and 8% false positive rate, respectiv ely). Although arbitrary, this method has the advantage of allowing the computation of discriminability and bias in cases of perfect performance. The alternative, which is to di scard participants with perfect hit rates or false positive rates, can result in under-estim ating discriminability and further complicate analyses through unequal discarding of participants between groups. Results ParticipantsÂ’ Characteristics Comorbid disorders included a history of substance abuse or dependence (eight depressed participants and one control pa rticipant), PTSD in full remission (two depressed participants), generalized anxiety disorder (one depressed participant), and panic disorder with agoraphobi a, in partial remission (one depressed participant). Using DSM-IV criteria, 11 of the 24 depressed pa rticipants were categorized as mildly depressed, while the other 13 were categorized as moderately depressed. Demographic variables and scores on the BDI, BAI and RRS for the two groups ar e reported in Table 2-2. The two groups did not differ in terms of age or education. As expected, however,
19 depressed participants had significantly great er BDI-II, BAI and RRS scores than control participants. Table 2-2. Demographic characteris tics and questionnair e scores by group Depressed ( N = 24) Control ( N = 20) Comparison Females Males Females Male Gender 18 6 15 5 2(1, N = 44) = 0 M SD M SD Age 35.83 13.96 32.25 13.45 t (42) = 0.86 Education 14.63 2.67 15.35 2.43 t (42) = -0.93 BAI 16.67 11.62 4.45 4.06 t (42) = 4.47** BDI-II 26.13 8.48 2.89 3.40 t (41)a = 11.22** RRS 57.25 12.68 32.35 8.15 t (42) = 7.58** a One participant did not fully complete the BDI-II; ** p < .01. Recognition Performance Average discriminability and bias for the two groups and the different categories of adjectives are reported in Ta ble 2-3. Although they were not the primary focus of our analyses, hit rates and false positive rates are also reported in Table 2-3. To directly test our hypothesis, we conducted separate anal yses for the self-d escriptive and the nondescriptive adjectives. Table 2-3. Recognition hit rates, false positiv e rates, discriminability and response bias by group and adjective category Depressed Group Control Group Adjective Category Hit rates False Positives Hit rates False Positives Negative/Self-descriptive .88 (.14) .31 (.27) .90 (.14) .13 (.12) Negative/Non-descriptive .95 (.08) .20 (.21) .95 (.10) .19 (.17) Neutral .96 (.06) .10 (.14) .94 (.12) .05 (.07) Positive/Self-descriptive .91 (.14) .23 (.24) .97 (.07) .14 (.22) Positive/Non-descriptive .92 (.12) .14 (.27) .96 (.12) .12 (.17) Adjective Category Discriminability (dÂ’) Bias (c) Discriminability (dÂ’) Bias (c) Negative/Self-descriptive 1.60 (0.76) -0.26 (0.47) 2.16 (0.59) -0.03 (0.21) Negative/Non-descriptive 2.09 ( 0.61) -0.22 (0.32) 2.12 (0.59) -0.19 (0.25) Neutral 2.22 (0.73) -0.27 (0.36) 2.48 (0.40) -0.14 (0.20) Positive/Self-descriptive 1.90 (0.75) -0.19 (0.41) 2.30 (0.62) -0.15 (0.31) Positive/Non-descriptive 2.18 ( 0.87) -0.09 (0.38) 2.34 (0.63) -0.10 (0.25)
20 Self-descriptive adjectives For discriminability, the 2 (Group: Depr essed, Control) x 2 (Valence: Negative, Positive) ANOVA yielded a significant main effect of group, F (1, 42) = 7.29, p < .05, d = 0.82, with depressed participants showi ng lower discriminability than control participants. There was also a nea r-significant effect of valence, F (1, 42) = 3.84, p = .057, corresponding to greater discriminability fo r the positive adjectives, but no interaction between group and valence, F (1, 42) = 0.56, p > .1. Thus, depressed participants showed lower discriminability than control particip ants for the self-descriptive adjectives, regardless of valence (see Figure 2-1). For response bias, the effects of group, F (1, 42) = 1.72, p > .1, and valence, F (1, 42) = 0.42, p > .1, were non-significant. There was, however, a significant interaction between group and valence, F (1, 42) = 4.37, p < .05. Depressed participants showed greater bias for the negative self-descriptive adjectives, t (32.57) = 2.15, p < .05, d = 0.61, but not positive, self-descriptive adjectives, t (42) = 0.37, p > .1, d = 0.11. Thus, depressed participant showed a greater propensity to endorse negative self-descriptive adjectives. Figure 2-1. Recognition discriminability a nd response bias for the self-descriptive adjectives as a function of group. A) Discriminability: The depressed group showed lower discriminability than th e control group, regardless of valence. 1 1.5 2 2.5 Neg/Char Pos/Char -0.4 -0.3 -0.2 -0.1 0 0.1 Neg/Char Pos/Cha r A B
21 B) Response bias: For response bias, th e depressed group showed greater bias for the negative, self-des criptive adjectives only. These results support the hypothesis that, in depression, negative, self-descriptive adjectives are primed and more likely to be produced as output. They do not support the hypothesis that depressed indivi duals show better pr ocessing of negative, self-descriptive adjectives, as discriminability was lower, in the depressed group, for both negative, selfdescriptive adjectives and positive, self-descriptive adjectives. Non-descriptive adjectives For discriminability, none of the effects reached significance, all p Â’s > .1. For bias, there was a significant effect of valence, F (1, 42) = 4.30, p < .05, but no effect of group, F (1, 42) = 0.01, p > .1, and no interaction be tween group and valence, F (1, 42) = 0.12, p > .1. Thus, the pattern observed for self-d escriptive adjectives was not found for nondescriptive adjectives. In particular, the two groups did not differ in terms of discriminability, suggesting that the lower di scriminability reported above was specific to the self-descriptive adjectives. Two additiona l analyses confirmed this observation. First, the two groups did not differ on discriminability for the neutral words, t (42) = -1.4, p > .1. Second, a 2 (Group: Depressed, Control) x 2 (Descriptiveness: Se lf-descriptive, Nondescriptive) x 2 (Val ence: Negative, Positive) ANOVA sh owed an interaction between descriptiveness and group for discriminability, F (1, 42) = 1.57, p < .05, confirming that the difference between groups was greater fo r the self-descriptive than for the nondescriptive adjectives. Adjective Selection For each participant, we computed th e average length, frequency, valence and arousal of the 12 adjectives they selected for each category (see Table 2-4). Consistent
22 with the presence of a negative self-schema, depressed participants chose negative, selfdescriptive adjectives that were more strongl y negative (i.e., higher ra tings), and positive, non-descriptive adjectives that were more st rongly positive (i.e., lower ratings), compared to those selected by control participants. De pressed participants al so chose negative/nondescriptive adjectives with higher arousa l ratings than those selected by control participants. Table 2-4. Characteristics of selected adjectives by group Depressed Control t (42) Negative/Self-descriptive Length 7.55 (0.48) 7.47 (0.48) 0.51 Frequency 2.68 (0.23) 2.65 (0.16) 0.36 Valence 6.99 (0.13) 6.89 (0.11) 2.89** Arousal 5.09 (0.28) 5.07 (0.25) 0.26 Negative/Non-descriptive Length 7.28 (0.46) 7.50 (0.50) -1.44 Frequency 2.48 (0.28) 2.36 (0.34) 1.25 Valence 7.28 (0.08) 7.33 (0.11) -1.73 Arousal 5.72 (0.28) 5.46 (0.32) 2.80** Positive/Self-descriptive Length 7.60 (0.66) 7.48 (0.68) 0.60 Frequency 3.36 (0.25) 3.36 (0.28) -0.01 Valence 2.89 (0.13) 2.82 (0.13) 1.87 Arousal 4.76 (0.36) 4.90 (0.47) -1.14 Positive/Non-descriptive Length 7.16 (0.68) 7.15 (0.73) 0.03 Frequency 2.94 (0.26) 3.00 (0.37) -0.64 Valence 2.52 (0.11) 2.61 (0.14) -2.23* Arousal 5.47 (0.48) 5.41 (0.53) 0.42 p < .05; ** p < .01. Relationship Between Adjective Characteristics and Recognition For each group and adjective category, we examined whether discriminability or bias correlated significantly w ith length, frequency, valence a nd arousal of the adjectives selected by the participants. Within the de pressed group, the only si gnificant association was between length and response bias for th e negative, self-des criptive adjectives, r (24) = .50, p < .05. Within the control group, lengt h correlated sign ificantly with
23 discriminability, r (20) = -.67, p < .01, and response bias, r (20) = .56, p < .05, for the positive, self-descriptive adjectives. Frequency, valence and arousal did not correlate significantly with discriminability or bias in either group for any adjective category, all p Â’s > .1. Thus, selection of adjectives could not acc ount, in this sample, for the differences observed between the two groups in recognition bias and discriminability. Relationship Between Recognition Performance and Questionnaires We examined the relationship between the questionnaires (B DI-II, BAI and RRS) and recognition discriminability and bi as. In the control group, BDI-II score was negatively correlated with discriminability for negative, self-descriptive adjectives, r = .50, p < .05, negative, non-de scriptive adjectives, r = -.48, p < .05, and positive, nondescriptive adjectives, r = -0.61, p < .01. There was also, in the control group, a significant negative correlation between BA I and recognition bias for the positive, descriptive adjectives, r = -.46, p < .05. In the depressed group, the only significant correlation was between BAI sc ore and response bias for positive, self-descriptive adjectives, r = .41, p < .05. With a Bonferroni correcti on, only the correlation between BDI-II and discriminability for positive, non-desc riptive adjectives in the control group remained significant. Discussion Our findings call into question the classi cal view that clinically depressed individuals have better memo ry for negative, self-descrip tive information. In our study, depressed participants showed greater recognition bias, but not discriminability, for negative, self-descriptive adjectives. In other words, they were more likely than control participants to report remembering negative, self-descriptive adjectives. This greater
24 endorsement, however, was not accompanied by greater accuracy. To the contrary, depressed participants showed lower discrimina bility than control participants for selfdescriptive adjectives, re gardless of valence. When research participants are tested on wo rds that they have individually selected, there is a danger that observed differences in memory performance are in fact due to systematic differences in the selection proce ss. Indeed, previous studies (Zuroff et al., 1983) suggest that depressed individuals might differ from non-depressed individuals in the way they select self-descriptive adjectives. Consistent with our predictions, such biases were also observed in our study. Depressed participan ts selected negative, selfdescriptive adjectives with stronger nega tive ratings and posi tive, non-descriptive adjectives with stronger positive ratings than those selected by control participants. These selection biases suggest the presence of a ne gative self-schema in depression, consistent with BeckÂ’s theory (1967). They did not however, account for the memory biases observed in our study, as valen ce and arousal ratings did not correlate significantly with recognition bias or discrimi nability in either group. Our study is the first to use a SDT approach to examine recognition biases for selfdescriptive and non-desc riptive adjectives in clinical depression. Our results challenge the commonly held view that depression is associated with better memory for negative, self-descriptive information. They are consiste nt with those of Zuroff et al. (1983), who found greater recognition bias for negative adjectives in female college students with depressive symptomatology. They are, however inconsistent with those of Dunbar and Lishman (1984), who found an interac tion between group and valence for discriminability. We extended findings from Zuroff et al. in several ways. First, whereas
25 Zuroff et al. selected depressed college students by using a cutoff score on the Beck Depression Inventory, depresse d participants in our study were selected based on a formal diagnosis of depression, obtained through standardized assessment using the SCID-IV (First et al., 1996). The equivale nce between clinical depression and selfreported symptoms of depression has been questioned (Coyne, 1994; Vredenburg, Flett, & Krames, 1993). In particular, Coyne, T hompson and Racioppo (2001) have recently found systematic differences between clinica lly depressed and non-cl inically depressed individuals with similar leve ls of self-reported depressi ve symptomatology. Second, our participants spanned a wide rang e of ages (18 to 60), whereas participants in Zuroff et al. were all college students. As noted by Fle tt, Vredenburg, and Kram es (1997), there are particularities of the college environm ent (e.g., greater opportunities for social interactions, academic pressure) that might in fluence the nature of depression in this population. Third, our methods allow us to sp ecify that the increased recognition bias observed in the depressed group was specific to negative, self-descr iptive adjectives. In contrast, Zuroff et al. tested recognition w ith test items that had been individually selected as self-descriptive, but foils that ha d not been selected for self-reference. Thus, their findings are ambiguous as to the specific ity of memory biases in depression. Finally, as noted before, we ruled out selection bi as as a likely explanation for the observed differences in memory performance betw een depressed and control participants. There are several potential reasons for the discrepancy between our findings and those of Dunbar and Lishman (1984). In Dunba r and Lishman, all depressed participants were inpatients, and comorbidity with anxiet y disorders was not reported. It is possible that several of them suffered from an a nxiety disorder, such as PTSD and OCD,
26 etiologically related to their depression. Moreover, many of the stimuli used in Dunbar and LishmanÂ’s study appeared anxiety-related rather than depressi on-related (e.g., Â“crash, scream, snake, mobÂ”). Thus, their findi ngs may have resulted from informationprocessing biases related to a nxiety rather than depression. In contrast, comorbid anxiety was rather minimal in our sample, and the use of self-descriptive adjectives ensured that the stimuli were relevant to each participan tÂ’s self-concept. Anothe r potential source of discrepancy between the two studies is de pression severity. Whereas all of our participants were in the mild or moderate range, the inpatie nt sample used in Dunbar and LishmanÂ’s study may have included a number of participants within the severe range of depression severity. Future research is need ed to determine the influences of anxiety comorbidity and depression severity on inform ation-processing using an SDT approach. Our results should be placed in the genera l context of the sel f-reference effect. As already noted, the self-referen ce effect refers to the memory advantage resulting from processing information in reference to the self (Rogers et al., 1977). It is typically stronger for positive than for negative information, except for individuals with depression, for whom this pattern is reverse d. To our knowledge, pr evious studies have not examined the question of wh ether the self-reference effect is due to an actual memory advantage or to report bias. Fiedler et al. (2001), however, suggested that the effects of mood manipulation on memory biases are due to differences in discriminability rather than response bias. From the discrepancy betw een our results and thos e of Fiedler et al., it is interesting to speculate whether the memory biases due to an experimental manipulation of mood, and those due to the presence of clinical depression, reflect different processes that can be clearly distinguished using SDT. This possibility would
27 have important implications for understanding information-processing biases in depression, and for the debate regarding the continuous vs. discrete nature of psychopathology. Information-Processing and Response Bias How should one interpret the presence of a recognition bias for negative, selfdescriptive adjectives in clinically depr essed individuals? Acco rding to Anderson and Bower (1972), subjects who decide whether they recognize an item proceed through two distinct and sequential processes. First, par ticipants determine the number of context cues associated both with the test item and th e encoding situation. Second, subjects compare the number of cues to a criterion that, depending on the subjectsÂ’ motivation or expectations, can be more or less stringent Although recognition bias has typically been assumed to reflect the latter of these tw o processes, we believe that informationprocessing biases occurring at either of thes e two stages could result in the recognition biases observed in our study. We now ex amine these two possi bilities in turn. Recently, the concept of rumination has rece ived increased attention in depression research. Rumination, as defined by Nolen-Ho eksema (1990), refers Â“to the tendency to repetitively focus on oneÂ’s depressive sympto ms as well as the causes and consequences of these symptoms.Â” In a prospective study of depression and posttraumatic stress symptoms after the 1989 earthquake in the San Francisco Bay Area, Nolen-Hoeksema and Morrow (1991) found that rumination, defi ned in this way, predicted depressive symptomatology 10 days and 7 weeks after th e earthquake, even af ter controlling for initial levels of depressive symptomatology. Since then, numerous studies have explored the nature of rumination (Watkins & Baracaia, 2001; Watkins & Mason, 2002), its consequences on problem-solving (Lyubomi rsky & Nolen-Hoeksema, 1995), and its
28 ability to predict depressive symptoms prospectively (Nolen-H oeksema, Morrow, & Fredrickson, 1993; Nolen-Hoeksema et al., 1994). Rumination is relevant to our study, in that it suggests that depression is associat ed with a tendency to dwell over negative events, concepts, and stimuli. Indeed, Si egle et al. (2003) found that depressed individuals showed prolonged pupil dilation af ter exposure to negativ e, self-descriptive adjectives. In our study, the stimuli, at enco ding, were separated by long delays between trials (6 or 36 second), during which participan ts did not have any ta sk to perform. Such encoding conditions were designed to favor rumination in individuals with ruminative tendencies. While ruminating, these particip ants may have generated negative words related to the stimuli that had just been presented to them, thereby creating spurious memory traces. These memory traces were generated during the same period as the actual test items, and were associated with the same context. Thus, they might have easily been confounded with actual test item s during recognition testing. Contrary to this hypothesis, we did not find, in our study, a significant correlation between recognition indices and rumination sc ores on the RRS. These negative findings, however, do not definitely rule out the possi ble contribution of ru mination tendencies to our findings. As noted by Siegle et al. (2003), there exist multiple measures of rumination, and these measures show low to non-significant correlations, suggesting that rumination might be a multi-dimensional and yet poorly understood construct. Moreover, in Siegle et al., only a few of these m easures correlated with prolonged amygdala activity, and the correlations were in the mild to moderate range. Thus, it is possible that ruminative tendencies that are not captur ed by the RRS contributed to increased rumination at encoding and, ultimately, greater response bias for the negative, self-
29 descriptive adjectives in the depressed group. Williams et al. (1988), for instance, argued that depression is characterized by an increa sed elaboration of negative material, defined as the "the activation of a representation in re lation to other associat ed representations to form new relations between them and to activate old relationsh ips" (p. 170). If elaboration, defined in this way, is dis tinct from rumination, it might have been responsible for the memory biases observed in our study. A second explanation for our findings is suggested by the traditional view that bias, in SDT, refers to the use of a more or le ss liberal criterion for deciding whether a given item belonged to the original list. According to this view, differences in recognition bias are not due to differences at encoding. Rath er, they are the result of differences in strategies, motivations or expect ations at retrieval. In the ca se of recognition testing, what might these differences be? One possibility is that, when making judgments about a situation or object, people use their current affective state as an element of their evaluation, whether their affect is relevant to the current situation or not (Schwarz & Clore, 2003). This theory is known as the Â“Affect-as-Informat ionÂ” hypothesis, and it has strong empirical support and a wide range of applications In our study, participants exposed to an item during recognition testing might have been asking themselves, Â“Does this item feel right?Â” before engaging in an extensive search for cues that would relate the item to the learning context. For the depressed participants, negative, self-descriptive adjectives would have felt particularly Â“right ,Â” because they were congruent with their current mood and the type of thought s in which they typically engaged. From our study, it is impossible to determ ine whether elaboration at encoding, the use of heuristics at retrieval, a combinati on of the two, or some other factors were
30 responsible for observed group differences. Se veral experimental manipulations could tease apart these various possibilities. For in stance, one could mani pulate elaboration at encoding by including a distract ing task after some, but not a ll test items. One could also apply the Â“remember-knowÂ” procedure (Tulving 1985) to determine whether the biases observed in our study were due to familiarity (the impression of being familiar with the item) or recollection (which involves remember ing at least some elements of the context during which the item was learned). Additional Findings In addition to a specific recognition bi as for negative adjectives, depressed participants in our study showed lower discriminability for al l self-descriptive adjectives, regardless of valence. These findings are consis tent with a vast lite rature showing that depression is associated with slight impairm ents in verbal memory (see Burt, Zembar, & Niederehe, 1995, for a meta-analysis). Intere stingly, though, we observed this effect only for the self-descriptive adj ectives. For the non-descriptiv e adjectives, depressed and control participants did not differ in term s of discriminability, and a significant interaction between group and descriptiv eness confirmed that the difference in discriminability between the two groups was greater for the self-descriptive than for the non-descriptive adjectives. How are we to interp ret these complex results? In their initial studies of the self-referen ce effect in depression, Davi s (1979) and Davis and Unruh (1981) published findings suggesting that clinic ally depressed indivi dual, particularly individuals with a first-onset depressive episode, show poor cognitive organization of self-descriptive adjectives. They interpreted these findings as sugges ting that first-onset depression is associated with a lack of stable self-schema, while individual s with chronic or recurrent depression do have a stable (presumably negative) self-schema. In a rebuttal
31 of these findings, Derry and Kuiper (1981) show ed that the self-reference effect was in fact present in clinically depr essed individuals, but restricted to adjectives with negative contents. The study by Derry and Kuiper was hi ghly influential, and generated a vast literature on memory biases for emotional words in depression. Paradoxically, though, our findings lend some support to the initial in terpretation of an unstable self-schema in depression: in our study, depressed individuals showed lower discriminability for both negative and positive self-descriptive adjectives, but not for non-descriptive adjectives. To further test the hypotheses of Davis a nd Unruh, we created two sub-samples of our depressed group: individuals with first-onset, non-chronic depression ( N = 6), and individuals with chronic or recurrent depression ( N = 17). We then compared these two groups for discriminability using a Bonferr oni correction for multiple comparisons. The results were exactly in line with the hypothe ses of Davis and Unruh: individuals with first-onset, non-chronic depression showed lowe r discriminability than those with chronic or recurrent depression for ne gative, self-descriptive, ( M = 0.76, SD = 0.70 for the firstonset subgroup; M = 1.91, SD = 0.55 for the chronic/recurrent subgroup), t (21) = -4.10, p < .05, and positive, self-descriptive adjectives ( M = 1.25, SD = 0.89 for the first-onset subgroup; M = 2.16, SD = 0.56 for the chronic/recurrent subgroup), t (21) = -2.94, p < .05, but not for non-descriptive or neutral adjectives (all p Â’s > .1). Despite the small sample sizes of our subgroups, these results are provo cative, and suggest th at the hypothesis of an unstable self-schema in first-onset de pression deserves further consideration. Another finding that deserves discussion is that the response biases observed in the depressed group concerned only the negative, self-descriptive adjectives. In contrast, there were no group differences in response bias for the positive, self-descriptive
32 adjectives. Assuming that depressed indivi duals possess a negative self-schema, while non-depressed individuals possess a positiv e self-schema, we would expect group differences for the positive as well as the negative self-descriptive adjectives. However, previous studies suggest that, rather than a purely negative self-schema, individuals with mild to moderate depression might possess an ambivalent self-schema, with both positive and negative traits (Ross & Mueller, 1986). In addition, Zuroff et al. (1983), who examined recognition bias in college st udents with high BDI scores, also found a significant difference between groups only fo r the negative adjectives. In our study, clinically depressed participants were categorized as mildly ( N = 11) or moderately ( N = 13) depressed. None of them met criteria for severe clinical depression. Thus, our findings suggest that, in indivi duals with mild to moderate clinical depression, the selfschema may include both positive and negative information. Implications for the Cognitive Model and Treatment The cognitive model of depression stat es that depressed individuals show dysfunctional attitudes, biased judgments a nd irrational beliefs th at contribute to the development and maintenance of depre ssion (Beck, 1967, 1976). A ccordingly, cognitive therapy is designed to challe nge maladaptive judgments a nd attributions, and replace them with more adaptive ones. Less has been discussed about memory biases and their potential role in the development and main tenance of depressive states. Our findings suggest that, in addition to overall worse me mory for self-relevant material, depressed individuals might be prone to increased repor ting of negative information, including false positive errors. Therapists do not generally cha llenge the veracity of the facts recounted by their patients, and for good reasons: they we re not present during the original events, and doing so would likely damage rapport and hinder progress in therapy. Therapists
33 working with depressed clients, however, might be well-advised to keep a critical stance regarding negative events reporte d in therapy. Without directly contesting the veracity of the facts, they can probe by asking further questions, and encourage their clients to carefully separate the facts from their interp retations. Doing so might help their clients develop a more realistic and unbiased view of past events, in the same way cognitive therapy consists of challengi ng dysfunctional beliefs and attr ibutions and replacing them with more adaptive ones. One of the unexpected aspects of our findings is that they supported the view, first developed by Davis and Unruh (1981), that first-onset, non-chronic depression is associated with an unstable self-schema, leading to impaired memory for self-descriptive adjectives. These results have important imp lications for a developmental perspective of cognitive style in depression. Possibly, indivi duals who experience a first-onset episode have not had the time to integrate the expe rience of depression into their self-schema. These individuals might be particularly amenab le to cognitive interven tions, as they have not yet developed a highly or ganized negative self-schema. Limitations Several limitations of our study should be noted. First, our findings were based on recognition testing only, whereas most previous studies of information-processing biases in depression have used measures of fr ee recall (Mineka & Sutton, 1992; Williams et al., 1988). Recall and recognition, however, differ in important ways. Free recall, more so than recognition, involves the active search for retrieval cu es and makes the greatest demands on self-initiat ed, strategic processes. Some studies, for instance, have found greater activation of frontal areas during free recall than r ecognition (Cabeza et al., 1997), and a meta-analysis (Wheeler, Stuss, & Tulvi ng, 1995) showed that patients with frontal
34 lobes lesions are more impaired on free r ecall than recognition. Moreover, there are reasons to believe that the task demands asso ciated with the free recall of negative words would be particularly benefi cial to depressed individuals In a study of conceptually driven implicit memory, Watkins, Vache, Verney, Muller and Mathews (1996) found that depressed participants, compared to non-depre ssed control participants, showed stronger associations between negative words of simila r meanings. Other studies, using different methodologies, have also noted the close connectivity between negative words in depressed individuals (Dozois, 2002; Dozo is & Dobson, 2001). If, indeed, depressed individuals have closely inte rconnected networks of negativ e concepts organized around a negative self-schema, the free recall of negative, self-descriptive words should be facilitated by the presence of numerous a nd proximal retrieval cues. Recognition, which does not rely as heavily on the presence of re trieval cues, would not benefit as strongly from this interconnection within the negative self-schema. In other words, we believe it possible that, on recognition, de pressed individuals would onl y show greater bias (but worse discriminability) for negative, self-des criptive adjectives, while showing a true memory advantage on free recall for these same adjectives. At this point, however, we feel that any potentially Â“trueÂ” memory adva ntage needs to be assessed independently of the response bias present in depressed individuals. A second limitation concerns the high hit rates and low false positive rates observed in most conditions. Some participants particularly in the control group, showed perfect performance for certain categories of adjectives, raising the possibility of ceiling effects. If anything, however, these ceiling e ffects would have reduced the differences in discriminability and bias between groups. It is therefore possible that, if ceiling effects
35 had been avoided, the pattern observed in our study would have been more pronounced, and additional effects that we did not observe would have been present. In order to equate the number of test ite ms and foils in each ca tegory, we decided to have participants select self -descriptive and non-descriptiv e adjectives in a separate, preliminary session. Consequently, participants were exposed to test items and foils once before the actual testing sessi on. Rather than rumination or th e use of heuristic strategies, source confabulation might have been respons ible for the high fals e positive rates and recognition bias observed in the depresse d group for the negati ve, self-descriptive adjectives. In our study, we minimized this possi bility in several different ways. First, the testing session took place at least one week after the pr eliminary session. Second, the context in which the adjectives were presen ted varied substantiall y between sessions. In the preliminary session, adje ctives were printed on small paper cards, while they appeared on a computer screen during th e testing session. Moreove r, the two sessions took place in different buildings, and the pres entation of the adjectiv es during the testing session took place in a dimly lit room while pa rticipants were attached to electrodes measuring physiological response Â– a very dis tinct context. Third, the instructions at recognition emphasized the context of the learni ng list: participants were instructed to indicate whether they had seen each word Â“i n the room with the computer while they were attached to the electrodes.Â” Even though we cannot completely ru le out the potential contribution of source confabulation to our results, we feel that these different precautions make this possibility unlikely. Although positive and negative adjectives we re well balanced in terms of length and arousal, they differed in terms of word frequency, with the posit ive adjectives having
36 greater frequency than the negative. It is possible, although unlikely, that the lower frequency of the negative adjectives woul d have made them more susceptible to differences between groups. Contrary to this hypothesis, we did not find any correlation between word frequency and recognition hit rate s, false positive rates, discriminability or bias in any of the adjective categories. Thus the differences in memory performance that we observed between negative, self-descriptive adjectives and positive, self-descriptive adjectives were probably not due to the unbalanced frequency of negative and positive adjectives. In our study, we did not find a significan t correlation between valence and memory performance, or between arousal and memor y. These results may seem surprising, given previous findings of enhanced memory for emotional than for neutral words (Bock, 1986; Kensinger & Corkin, 2003; Talmi & Moscovitc h, 2004). Most previous studies, however, have used free recall as a primary outcome measure, and one study found better recall, but not recognition, for emotional than fo r neutral words (Doerksen & Shimamura, 2001). In addition, our analyses were conducted separately for each category of adjectives, and the range of valence a nd arousal ratings wa s therefore limited. Conclusions Our study challenges the idea that depressi on is associated with better memory for negative, self-descriptive words. Using SD T, we found that clinically depressed individuals showed greater rec ognition bias for negative, se lf-descriptive adjectives, and worse discriminability for all self-descrip tive adjectives. Our findings could not be accounted for by systematic differences betw een the depressed a nd the non-depressed groups in the selection of self-descriptive adjectives. Whether the observed differences were due to increased elaborati on at encoding, or to the use of distinct heuristic strategies
37 at retrieval, is unknown at this point. Moreover, the discre pancy between our findings and studies of the mood-congruency effect sugge sts that the cognitive style of individuals with clinical depression is qualitatively differe nt from that of control participants having undergone a negative mood induction. These resu lts have important implications for the cognitive model of depression, and for th e treatment of depressed individuals.
38 CHAPTER 3 WHEN GOOD TURNS INTO BAD: D ELAYED STARTLE POTENTIATION BY POSITIVE STIMULI IN CLINICAL DEPRESSION Previous research suggests that mildly and moderately depressed individuals show normal startle modulation by emotional stimu li. These findings are somewhat surprising, given the robust information-processing bias es observed in that population. We used clinically relevant stimuli to assess startle reactivity in patients with mild and moderate clinical depression. Fourteen clinically depressed and 15 control participants received white-noise startle probes during and after the viewing of adjectives varying in valence and descriptiveness. While viewing the adjec tives, control but not depressed participants showed startle potentiation by self-descriptive adjectives, F (1, 14) = 7.77, p < .05. After a delay, depressed but not control participants showed startle potentiation by positive stimuli, F (1, 13) = 7.34, p < .05. Within the depressed gr oup, ruminative tendencies and the number of depressive episodes correlated with startle potentia tion after a delay. Our results have important implications for the a ssessment of startle re activity in depression and for an understanding of affective reactions and reinforcement in that population. Introduction One of the most common features of depression is anhedonia, a lack of motivation, interest or pleasure in enjoya ble activities. Anhedoni a is one of two Â“coreÂ” symptoms that additionally include dysphoria. At least one of these is require d to meet diagnostic criteria for a major depressive episode. Among i ndividuals meeting criteria for a Major Depressive Disorder (MDD), a nhedonia is almost always pr esent (American Psychiatric
39 Association, 2000). In these c onditions, it is not surprising th at several theorists have posited that reduced appetitive tendencies a nd exaggerated aversive tendencies may be core features of depression (Eastman, 1976; Fowles, 1988). The term Â“appetitive,Â” as we employ it here, refers to a set of reward-seek ing tendencies favoring behavioral activation and approach towards rewarding stimu li (Lang, Bradley, & Cuthbert, 1990, 1992). Aversive tendencies, on the other hand, are those that favor wit hdrawal from noxious stimuli. Perspectives from different disciplines suggest that depression is associated with reduced appetitive motivations and increased aversive mo tivations. From a cognitive perspective, depressed individua ls tend to adopt a negative vi ew of the self, the world and the future (Beck, 1967, 1976). As suggested by the revised hopelessness theory (Abramson, Metalsky, & Alloy, 1989) their attributions to ne gative events suggest the presence of a pervasive hopelessness about themselves and the world. In addition, information-processing research has shown pr eferential processing of negative, selfreferent information in depression (Min eka & Sutton, 1992; Williams, Watts, MacLeod, & Mathews, 1988). From a neurobiological pers pective, depression has been linked to functional brain asymmetry, particularly hypoa ctivation of the left frontal lobes and abnormalitities of the serotonergic, norepinephrine and dopaminergic systems (Tomarkenand & Keener, 1998). Activity wi thin neural structures including the amygdala, the anterior cingulate cortex (ACC) and the hippocampus have been implicated as well (Davidson, Pizzagalli, Nitschke, & Putnam, 2002; Drevets, 2000). Based on these neurobiological differen ces, Tomarken and Keener posited that depression is associated with a deficiency of the neurobiological substrates of the appetitive system, particularly as it relates to the functions of the frontal lobes such as
40 maintenance of long-term approach-related goals and shifting away from withdrawalrelated goals. These perspectives suggest that depressed individuals may react differently than non-depressed individuals to positive and ne gative stimuli. Relatively few studies, however, have directly examined this hypot hesis. Using facial expression, several researchers have found increased sensitivity to negative stimuli (Sloan, Strauss, Quirk, & Sajatovic, 1997) and decreased sensitivity to positive stimuli (Berenbaum & Oltmanns, 1992; Sloan, Bradley, Dimoulas, & Lang, 2002; Sloan, Strauss, & Wisner, 2001) in depressed individuals. In one study, indices derived from facial expressions showed differential sensitivity despite normal self-reports of experienced emotions (Sloan et al., 1997). Recently, studies using the startle eyebli nk response have examined emotional reactivity in individuals with depression. The startle respons e is a protective reflex following the presentation of abrupt and intens e stimuli. It includes a flexion of the arms and shoulders, a rising of the torso and a rapi d closure (blinking) of the eyes. In humans, the eyeblink response is one of the fastest and most reliable elements of the startle reflex (Hawk & Cook, 1997; Lang et al., 1990). Import antly, the amplitude of the startle eyeblink is modulated by affective or motiva tional states in humans and other mammals: it is potentiated by fear, anxiety and other aversive states, and inhibited by positive affective state (for a review, see Lang et al ., 1992). In the laboratory, emotional states (and the motivational states of approach or avoidance accompanying them) that are induced by viewing emotional pictures, list ening to emotional sounds, thinking about emotional vignettes, perceiving hedonic odors, or awaiting shock have all been shown to
41 modulate startle reactivity (M. M. Bradley, Cuthbert, & Lang, 1993; Ehrlichman, Kuhl, Zhu, & Warrenburg, 1997; Grillon, Ameli, Foot & Davis, 1993; Miltner, Matjak, Braun, Diekmann, & Brody, 1994; Vrana & Lang, 1990). Thus, the methodology of the startle eyebli nk offers an opportunity to examine the emotional response of an individual, or a gr oup of individuals, to positive and negative stimuli. The anatomy of the startle circuitr y is reasonably well characterized in animals and humans. The amygdala, in particular, play s a critical role in startle potentiation by fear (see Davis, Falls, Campeau, & Kim, 1993, for a review), while the nucleus accumbens is important for startle inhibition by pleasant stimuli (Koch, Schmid, & Schnitzler, 1996; Koch & Schnitzler, 1997). Allen, Trinder and Brennan (1999) were the first to examine emo tional reactivity in clinical depression using the startle eyeblink method. They found normal startle modulation (inhibition of star tle by pleasant stimuli and potentiation of startle by unpleasant stimuli) in particip ants with moderate depression. However, a small subgroup of severely depressed individua ls showed potentiation rather than inhibition of startle by positive pictures. Dichter, Tomarken, Shelton and Sutton (2004) assessed startle reactivity to emotional pictures in a group of depressed outpatients before and after treatment with Bupropion. They found an ab sence of affective st artle modulation and no treatment effect. Similarly, Kaviani et al. (2004) found an absence of startle modulation by emotional film clips in a small group of highly depressed and anhedonic inpatients, while relatively less severe patients showed normal startle modulat ion. In another study (Forbes, Miller, Cohn, Fox, & Kovacs, 2005), pa rticipants with a history of childhoodonset depression, compared to normal particip ants, showed greater startle inhibition by
42 pleasant pictures. Overall, however, they s howed a normal pattern of startle modulation, with startle inhibition by pleas ant pictures and startle pote ntiation by unpleasant pictures. Their responses to startle probes after picture offset were also similar to those of normal participants. Grillon et al. (2005), who exam ined startle responses in children and grandchildren of individuals with a hist ory of depression, also found normal startle modulation by threat (i.e., the expectation of an airblast to the larynx). Although the above-mentioned studies used different samp les and methods, and although the results differ to some extent, the overall pattern that emerges from these findings suggests that (a) Individuals with mild or moderate c linical depression show a generally normal modulation of startle amplitude by pleasant and unpleasant stimuli (f or an exception, see Dichter et al.), and (b) Severely depressed individuals show reduced startle modulation, or even a reversal of startle modulation (i.e., potentiation rather than inhibition) by positive stimuli. The absence of differential sensitivity to emotional stimuli in mild and moderate depression may seem surprising, given that robust emotional information-processing biases have been observed in this populati on (e.g., B. P. Bradley, Mogg, & Millar, 1996; P. C. Watkins, Vache, Verney, Muller, & Mathews, 1996). Previous research on the startle response in depression, however, suffe rs from important limitations. First, the stimuli used in these studies have consisted of pictures (Allen et al., 1999; Dichter et al., 2004; Forbes et al., 2005), film clips (Kaviani et al., 2004), and threat of airblast (Grillon et al., 2005), none of which were specifically developed for their relevance to individuals with depression. In the information-processing literature, differences in explicit memory between depressed and non-depressed individu als have been found most consistently for
43 negative and self-descriptive words, or unde r self-referent conditions (Mineka & Sutton, 1992; Williams et al., 1988). In the study by Sieg le, Steinhauer, Carter, Ramel and Thase (2003), for instance, clinically depressed pa rticipants showed pr olonged pupil dilation after exposure to negative words they had individually chosen as descriptive of themselves. Pupil dilation was also increase d, but to a lesser degr ee, for other negative words. A second limitation of previous resear ch concerns the time of delivery of the startle probes. With one exception (Forbes et al., 2005), the probes we re delivered while participants were viewing the pi ctures or films. Current conc eptualizations of depression, however, suggest that the cogniti ve style characteristic of cl inically depressed individuals may consist of an elaboration of negative material lasting beyond exposure to negative stimuli (e.g., Siegle, Steinhauer, Thase, Stenger, & Carter, 20 02). In particular, rumination, defined as Â“the tendency to repe titively focus on oneÂ’s depressive symptoms as well as the causes and consequences of these symptoms,Â” (Nolen-Hoeksema, 1990), has been proposed as a risk factor for th e development and maintenance of depressive episodes (Nolen-Hoeksema, 1991; NolenHoeksema, Morrow, & Fredrickson, 1993; Nolen-Hoeksema, Parker, & Larson, 1994). If rumination, or related constructs, best characterize the information-processing biases of clinically depr essed individuals, one would not necessarily expect differential se nsitivity to emotional tone during stimulus presentation. Rather, one would predict abnor mal startle modulation at later stages of processing, after the stimulus has disappeared a nd is not relevant to the task at hand. With this background in mind, the present study had three objectives (1) We aimed to assess startle modulation with emotional stimuli that have shown robust informationprocessing biases in clinical depression, i. e., negative and positiv e words individually
44 selected by each participant as self-descrip tive, (2) We sought to examine startle modulation during and after exposure to the em otional stimuli, and (3) We wanted to determine whether rumination is related to differential startle modulation in clinical depression. One difficulty associated with our choice of stimuli, however, should be acknowledged: the effects of em otional words on startle modulation have typically been less robust and reliable than the effects of em otional pictures, imagery or film clips. For instance, Larsen, Norton, Walk er and Stein (2002), who examined startle modulation in patients with panic disorder and social phob ia, found an effect size for social threatrelated vs. non-threatening wo rds of d = 0.06. Knost, Flor, and Braun (1997), on the other hand, found a robust potentiation of startle by body-related and pain-related words. Clearly, however, the evidence for the modul ation of startle amplitude by emotional words is much less extensive than it is for emotional pictures or film clips. In the present study, clinically depressed and never-depressed part icipants received startle probes while (no-delay) and after (d elay) viewing words on a computer screen. The words varied in valence (negative, neutral and positive) and descriptiveness (selfdescriptive and non-desc riptive). All depressed particip ants were outpatients, with symptoms ranging from mild to moderate. Fo r all participants, predictions for startle modulation in the no-delay condition include d several possibilities. First, one might expect a typical valence effect, such that negative words would yield greater amplitude than positive ones. Second, startle amplitude might be potentiated by negative, selfdescriptive and positive, non-descriptive adject ives, since both can potentially threaten oneÂ’s self-concept. Third, we might expect greater startle amplitude for self-descriptive than non-descriptive adjectiv es, given previous findings of startle potentiation when
45 attention is turned towards the self (Panay iotou & Vrana, 1998). We did not expect any group differences in the no-delay condition, sinc e previous studies suggest that mildly and moderately depressed individuals do not show any abnormality in affective startle modulation during exposure to emotional stimuli. In the delay condition, we predicted that control participants would not show any modulation of startle amplitude based on th e nature on the word. This prediction was based on the long delay between stimulus offs et and startle probe onset (more than 10 seconds). We reasoned that the affective im pact of the words would probably be shortlived in control participants, since they had no incentive to engage in continued processing after stimulus offset. Within the depressed group, however, we predicted startle potentiation after negative, self -descriptive and positive, non-descriptive adjectives. We also predicted that startl e potentiation in the delay condition would correlate with self-report ed ruminative tendencies. Method Overview Depressed and control particip ants took part in two experi mental sessions separated by one to three weeks. During the first session, they sorted negative and positive adjectives as self-descrip tive or non-descrip tive of themselves. During the second session, their startle eyeblink responses to wh ite noise bursts were recorded while and after they viewed adjectives se lected from the first session. Participants Participants were recruited through f liers posted within the community and advertisements in local newspapers. All pa rticipants were betw een 18 and 60 years old. Participants within the depr essed group all met diagnostic criteria for a current Major
46 Depression Disorder (MDD), w ithout Psychotic Features, ac cording to the Diagnostic and Statistical Manual of Mental Disorder s, Fourth Edition (American Psychiatric Association, 2000). Exclusion cr iteria for the depressed group included (a) A history of psychotic disorder, bipolar disorder, or obs essive-compulsive disorder (OCD), (b) Any history of neurological disorder such as epilepsy or trauma tic brain injury (c) A history of substance abuse or dependence within the la st six months), (d) A current diagnosis of post-traumatic stress disorder (PTSD), (f) A hi story of recent suicide attempt (less than six months), or plans or intent to commit suic ide, (g) Use of medications known to affect cognitive functioning, such as benzodiazepines or certain anti-psychotic medications. For the control group, all exclusion criteria appl ied and, in addition, e ligible subjects were required to have no history of mood disorder or anxiety diso rder, and a score lower than 13 on the Beck Depression Invent ory Â– Second Edition (BDI-II). Depressed individuals who were taking an ti-depressant medications were included in the study as long as they had been on a st able dosage of medicat ion for at least six weeks. Given the high comorbidity between depression and anxiety disorders (Hasin, Goodwin, Stinson, & Grant, 2005), we also incl uded depressed individu als with a current comorbid anxiety disorder (except for OCD a nd PTSD) or with a history of PTSD in full remission, as long as it was determined th at the anxiety disorder was not a major contributor to the current majo r depressive episode. In cases of diagnostic uncertainties, consultation with a board certified psychologi st was sought, and a generally conservative approach was taken to part icipantsÂ’ inclusion. The init ial sample consisted of 24 depressed participants and 24 c ontrol participants. Three of the control participants were excluded based on high BDI-II score, and another based on a current diagnosis of specific
47 phobia. In addition, 10 depressed participants and 5 control participants had unusable startle response data. The final sample consisted of 14 depressed and 15 control participants. Materials Sixty negative and sixty positive adjectives were selected from a large pool of emotional adjectives collected from prior st udies (B. P. Bradley & Mathews, 1983; M. M. Bradley & Lang, 1999; Scott, Mogg, & Brad ley, 2001). Ten nave judges (5 men, 5 women; 9 Caucasian, 1 African -American) rated each adjective on two separate 9-point scales for valence and arousal, respectivel y. Valence ranged from Happy (1) to Unhappy (9), and arousal ranged from Calm (1) to Excited (9) (see M. M. Bradley & Lang, 1999, for a complete set of instructions). The negative and positive adjectives differed significantly in terms of valence, but not arousal (s ee Table 3-1). Word length (i.e., number of characters) and frequency were al so computed for each adjective. Frequency was computed using norms from Kucera and Francis (1967), and log-transformed to minimize the influence of very large scores The negative and positive adjectives did not differ in terms of length. They did, however, differ in terms of frequency, with the positive adjectives being more frequent than the negative adjectives (see Table 3-1). Table 3-1. Adjective characteristics by category Negative Adjectives Positive Adjectives Comparison Min/Max M (SD) Min/Max M (SD) Length 3/13 7.4 (2.2) 3/13 7.3 (2.5) t (118) = 0.08 Frequency 2/119 19.9 (27.1) 1/313 49.1 (63.3) t (118) = -3.27* Ln(Frequency) 0.69/5.29 2.5 (1.0) 0.00/5.57 3.3 (1.2) t (118) = -3.65** Valence 6.30/8.00 7.1 (0.4) 1.50/3.70 2.7 (0.5) t (118) = 51.33** p < .01; ** p < .001.
48 Session 1 Semi-structured interview Participants completed a semi-structured interview that served as a basis for all Axis I diagnoses. All interviews were conducted by the first author using the Mood Disorders, Psychotic Screen, SubstanceRelated Disorders and Anxiety Disorders modules of the Structured Clinical Inte rview for the DSM-IV-TR (SCID-IV, First, Spitzer, Gibbon, & Williams, 1996). Interviews we re audiotaped, and 10 interviews from the depressed group and 8 interv iews from the control group were randomly selected and recoded by the second author, blinded to orig inal ratings. Inter-rater agreement was 98% for individual SCID-IV items, and 95% for Axis I diagnoses. The less-than-perfect agreement for diagnoses was due to one depr essed participant for whom the second, but not the first author, diagnosed a distant history of alcohol ab use. There was perfect interrater agreement regarding MDD severity (3 mild, 7 moderate). Sorting of adjectives Participants were presented with a set of 60 cards with the negative or positive adjectives printed on top. They were aske d to put each card in one of four groups: The least like me Less like me More like me and The most like me After an initial sorting, they were instructed, if necessary, to adjust their sorting in order to have at least 12 adjectives per group. Each participant sorted negative and positive adjectives separately. Valence order was alternated among participan ts, and the cards within each valence were presented in fixed random order. For each participant, six ad jectives were randomly selected from the two most extreme groups ( The least like me and The most like me ). Thus, there were four categories
49 of individually selected adje ctives for each participant: Negative Self-descriptive Negative Non-descriptive Positive Self-descriptive and Positive Non-descriptive Startle Task (Session 2) Upon arrival, surface Ag-AgCl electrodes filled with an isotonic electrolyte were positioned under the participantsÂ’ left and ri ght eyes to record electromyographic (EMG) activity from the orbicularis oculi muscle. Electrode placement followed recommendations by Fridlund and Cacioppo (1986) Participants were then brought into an electrically shielded, soundattenuated room where they we re instructed to sit down and told that they would hear brief noises. Startle eyeblink respons es were elicited by 50 ms, 95 db bursts of white noise (probes) w ith instantaneous rise time produced by a Colbourn S81-02 module and delivered binaur ally through Telephonics (TD-591c) stereo headphones. Twelve probes were delivered in the absence of any other stimulus (blank startles). These initial probes we re designed to ensure appropri ate elicitation of the startle responses, correct any potential problems with electrodesÂ’ functioning or placement, and assess group differences in bl ank startle characteristics. The stimuli used for the emotional startle ta sk were 24 adjectives selected from the first session (6 from each category), plus 12 neutral words and 6 filler adjectives. Each stimulus was presented in white letters fo r 6000ms in the center of a black computer screen. After a 2000ms interval, a prompting scr een asked participants to rate how often ( Rarely Sometimes or Often ) they used that word or conc ept. Participants provided their responses verbally, and their responses we re manually recorded by the experimenter. Trials were separated by either a short dela y (6s) or a long delay (26s), in pseudorandom order.
50 For each stimulus except the filler adjectiv es, a white noise probe was presented at one of two time points: during stimulus pr esentation (no-delay pr obe: 4200ms, 5000ms or 5800ms after stimulus onset2), or after participants provide d their ratings (delay probe: 10200ms, 11000ms or 11800ms after rating comple tion). For each category, half of the stimuli had a no-delay probe, and the other ha lf had a delay probe. Two initial trials without probes were included at the beginning to familiarize participants with the task. Questionnaires In each session, participants completed the BDI-II, the Beck Anxiety Inventory (BAI), and the Rumination Responses Scale (R RS). The BDI-II is a widely used measure of depression symptomatology, with internal consistency and test-retest reliability both above .9 (Beck, Steer, & Brown, 1996). The BA I is a 21-item self-re port measure of anxiety symptoms, which was developed to a ddress the problem of the overlap between other measures of anxiety and measures of depression (Beck & St eer, 1990). The scale has good internal consistency ( .92) and test-retest reliability (.75), a nd its convergent and discriminant validity are supporte d by higher scores in patients diagnosed with an anxiety disorder than patients diagnosed with ma jor depression (Beck, Epstein, Brown, & Steer, 1988). The RRS is a 22-item self-report instru ment assessing the frequency of thoughts about oneÂ’s symptoms of depression. Its internal consistency ranges from .89 to .90 (Nolen-Hoeksema & Morrow, 1991; Treynor Gonzalez, & Nolen-Hoeksema, 2003; Wenzlaff & Luxton, 2003) and its test-retest re liability was .67 in a large community sample (Nolen-Hoeksema, Larson, & Grays on, 1999). It has been shown to predict 2 The variation in the timing of the probes was introduced so that participants could not predict their occurrence based on picture onset.
51 depressive symptomatology prospectively af ter traumatic events (Nolen-Hoeksema & Morrow, 1991) and the death of a lo ved one (Nolen-Hoeksema et al., 1994). Data Reduction The raw EMG signal was amplified (30,000 gain) and frequencies below 90 Hz and above 1000 Hz were filtered using Colbourn bioamplifiers. The raw signal was then rectified and integrated using a Colbourn Contour Following Integrator with a nominal time constant of 100 milliseconds. Digital sampling at 1000 Hz began 50 ms before presentation of the acoustic startle probe and continued for 250 ms after startle probe offset. Data from each participant were visual ly examined, and trials with clear artifacts (e.g., eyeblink movements before probe onset) were rejected. Subse quent data reduction was completed using a custom software program for data condensing. Latency and amplitude of the peak response within 20 ms to 150 ms after probe onset were determined. We discarded trials with a peak latency outside of the specified latency range, or with a peak amplitude more than 2.5 standard deviations above or below each participantÂ’s mean amplitude. In order to mini mize the effects of in ter-subjects variability in overall amplitude, raw scores were convert ed into T-scores (mean of 50, standard deviation of 10), standardized to the mean of the neutral, no-delay condition for each subjectÂ’s left and right eyes separatel y. For each category, average T-scores were computed using all valid trials from both ey es or, when data from one eye were invalid, from the valid eye. Only subjects who had at least two valid trials for each stimulus category were retained for subsequent analyses. Data Analyses For peak latency and the number of discar ded trials, we did not have any specific predictions and conducted 2 (Group: depressed, control) x 2 (Delay: no-delay, delay) x 5
52 (Stimulus category: negative self-descriptive, negative non-descriptive, neutral, positive self-descriptive and positiv e non-descriptive) analyses of variance (ANOVAs). For amplitude, however, we had specific and dist inct predictions for the no-delay and the delay conditions. Moreover, we were intere sted in the independent contributions of valence and descriptiveness, and their interaction. Thus, we conducted two separate 2 (Group) x 2 (Valence: negative, positive) x 2 (Descriptiven ess: self-descriptive, nondescriptive) ANOVAs for the two delay conditions. We decomposed interactions involving group by examining within-gr oup patterns of startle response. We also examined the effects of depr ession severity, use of antidepressants, number of episodes and scores on the questi onnaires on basic startle characteristics and startle modulation. Given that PearsonÂ’s correl ations are highly sensitive to outliers, particularly with small sample sizes (Siegel & Castellan, 1988), we computed nonparametric, Spearman correlations for all an alyses involving con tinuous variables. Results ParticipantsÂ’ Characteristics Comorbid disorders included a history of substance abuse or dependence (six depressed participants), PTSD in full remissi on (two depressed participants), and social phobia, mild (one depressed participant). Us ing DSM-IV criteria, 7 of the 14 depressed participants were categorized as mildly depr essed, while the other 7 were categorized as moderately depressed. Four depressed partic ipants were on antidepressant medications (Prozac, Paxil, Effexor and Wellbutrin, respec tively). Demographic variables and scores on the BDI, BAI, and RSS for the two groups are reported in Table 3-2. Both groups were predominantly female, and they did not differ in terms of gender, age or education.
53 As expected, depressed participants had significantly greater BDI-II, BAI, and RRS scores than control participants. Table 3-2. Demographic characteris tics and questionnair e scores by group Depressed ( N = 14) Control ( N = 15) Comparison Females Males Females Male Gender 9 5 11 4 2(1) = 0.28 Mean SD Mean SD Age 38.21 14.59 32.13 13.78 t (27) = 1.15 Education 14.43 2.65 15.33 2.41 t (27) = -0.96 BAI 13.64 10.22 4.40 4.29 t (27) = 3.22** BDI-II 25.07 7.93 3.57 3.72 t (26)a = 9.19** RRS 54.86 10.84 31.40 7.73 t (27) = 6.75** a One participant did not fully complete the BDI-II; ** p < .01. Blank Startle For the blank startle, the overa ll rate of discarded tria ls was 2.7% (2.8% for the control group and 2.7% for the depressed group) Trials were discar ded based on visual inspection (1.8%) and latency out of range (0.9%). The two groups did not differ significantly in terms of latency, amplit ude or habituation (see Table 3-3). Table 3-3. Characteristics of th e basic startle response by group Depressed ( N = 14) Control ( N = 15) Comparisona Mean SD Mean SD Latency (ms) 70.9 9.6 67.4 9.6 t (21) = 0.86 Amplitude (V) 18.9 11.4 17.9 10.2 t (21) = -0.21 Habituation (pV/trial) 16.0 24.0 3.5 12.7 t (21) = 1.43 a All comparisons were non-significant, all p Â’s > .1. Emotion-Modulated Startle Number of discarded trials The total percentage of discarded tria ls was 8.6% (9.8% in the depressed group, 7.5% in the control group). Di scarding was based on visual in spection (4.3%), latency out of range (1.8%), and amplitude out of range (2.5%). A 2 (Group) x 2 (Delay) x 5 (Stimulus Category) ANOVA did not yield any si gnificant main effects or interactions, all p Â’s > .1.
54 Latency to peak Latency to peak for each group and c ondition are reported in Table 4-4. A 2 (Group) x 2 (Delay) x 5 (Stimulus Category) ANOVA yielded only a marginal effect of group, F (1, 27) = 3.63, p = 0.067, with the depressed group having generally longer latencies than the control group. Table 3-4. Latency and amplitude (T-score) of startle response by group and condition Depressed Group Control Group Latency Amplitude Latency Amplitude No-Delay Neutrala 69.9 (8.1) 50.00 (0.00) 61.7 (9.1) 50.00 (0.00) Negative, self-descriptive 67.5 (7.8) 51.13 (5.85) 62.7 (9.4) 49.70 (7.02) Negative, non-descriptive 68.2 (5.3) 50.06 (5.82) 66.8 (11.5) 52.72 (8.74) Positive, self-descriptive 68.0 (11.4) 48.56 (7.14) 65.4 (10.4) 50.49 (7.78) Positive, non-descriptive 70.3 (10.4) 50.08 (6.52) 64.4 (9.7) 55.19 (7.51) Delay Neutral 70.4 (4.9) 52.93 (7.15) 64.2 (9.3) 50.13 (5.85) Negative, self-descriptive 69.4 (15.3) 50.10 (8.82) 65.8 (11.7) 51.93 (8.11) Negative, non-descriptive 68.8 (9.6) 52.48 (9.59) 61.6 (11.5) 49.83 (7.46) Positive, self-descriptive 70.9 (15.8) 55.06 (9.89) 65.0 (14.1) 50.94 (7.60) Positive, non-descriptive 70.8 (6.0) 57.01 (8.74) 66.3 (14.1) 51.24 (7.85) a When computing T-scores, we normalized to the mean of the neutral, no-delay condition. Thus, the T-score for this condition was always 50. Amplitude As noted before, T-scores normalized to each participantÂ’s mean startle response in the neutral, no-delay condition we re used to minimize inter-subject variability in startle amplitude. T-scores for each group and condi tion are reported in Table 4-4. In the nodelay condition, a 2 (Group) x 2 (Valence: nega tive, positive) x 2 (D escriptiveness: selfdescriptive, non-descriptive) ANOVA yielded a main effe ct of descriptiveness, F (1, 27) = 4.45, p < .05, and a near-significant interacti on between group and descriptiveness, F (1, 27) = 3.52, p = .072. The other effects were all non-significant, all p Â’s > .1. To decompose the interaction, we compared self-d escriptive and non-descriptive adjectives in the control and depressed groups, separa tely. In the control group, amplitude was
55 greater for the self-descrip tive adjectives than for th e non-descriptive adjectives, t (14) = 2.79, p < .05, while the difference was not si gnificant in the depressed group, t (13) = = 0.17, p > .1. In the delay condition, the only sign ificant effects for the 2 (Group) x 2 (Valence) x 2 (Des criptiveness) ANOVA were for valence, F (1, 27) = 5.40, p < .05, and the interaction betw een valence and group, F (1, 27) = 4.53, p < .05. In the control group, startle amplitude did not differ between the positive and the negative adjectives, t (14) = 0.16, p > .1. In the depressed group, however, th e positive adjectives yielded greater startle amplitude than the negative adjectives, t (13) = 2.71, p < .05.3 Relationship with Questionnaires Spearman correlations between startle ch aracteristics (latency, amplitude and habituation of the basic star tle, and T-scores for the em otion-modulated startle) and scores on the BAI, BDI-II, and RRS were comp uted separately within the control group and the depressed group. Within the contro l group, none of the correlations reached significance, all p Â’s > .1. Within the depressed group, RRS score correlated significantly with startle amplitude after a delay for the negative, non-descriptiv e adjectives and the positive, self-descriptive adjectives (see Table 4-5). In order to determine whether rumination could be responsible for some of the between-group differences in pa tterns of startle amplitude, we compared the depressed group and the control group on startle amplit ude for positive adjectives in the delay condition, before and after correcting for ru mination. A 2 (Group) x 2 (Descriptiveness) ANOVA for the positive adjectives in the delay condition yielded a near-significant 3 Given previous reports of a relationship between valence and arousal ratings and startle amplitude (e.g., M. M. Bradley, Codispoti, Cuthbert, & Lang, 2001), we computed bivariate, non-parametric (Spearman) correlations between T-scores and ratings of valence and amplitude for each adj ective category, separately in each group. We used a Bonferroni correction for mu ltiple comparisons. None of the correlations reached significance, all p Â’s > .1.
56 group effect, F (1, 27) = 3.75, p = .063, d = 0.72. After including RRS score as a covariate, however, the effect of group became non-significant, F (1, 26) = 0.04, p > .1, d = 0.24. Thus, rumination may have played an im portant role in the startle potentiation seen in the depressed group after a delay for the positive adjectives. Table 3-5. Spearman correlations between startle amplitude (T-scores) and scores on the questionnaires within the depressed group BAI BDI-II RRS No-Delaya,b Negative, self-descriptive .08 .50 .34 Negative, non-descriptive -.02 .07 .15 Positive, self-descriptive -.06 .10 .16 Positive, non-descriptive -.02 .09 .29 Delayb Neutral .11 .37 .45 Negative, self-descriptive .20 .50 .53 Negative, non-descriptive .13 .36 .65* Positive, self-descriptive .48 .56 .67* Positive, non-descriptive -.01 .44 .08 a Because the T-score for the neutral, no-delay co ndition was always 50, its correlation with the questionnaires could not be computed; b For each delay and each adjective category, a Bonferroni correction was used to correct for the co mparisons with the three questionnaires; p < .05. Effects of Depression Severity, Number of Episodes and Medications Within the depressed group, we examined the effects of depr ession severity (7 mild, 7 moderate), use of antidepressants (10 unmedicated, 4 medi cated) and number of depressive episodes on startle characteristics. For each of these three variables, we used a Bonferroni correction to account for compar isons with multiple indices of startle responsiveness. For depression severity a nd use of antidepressants, there were no differences between the two subgrou ps on any of the variables, all p Â’s > .1. For the number of depressive episodes, Spearman correlations revealed significant and positive associations with startle T-scores after a de lay for negative, self-descriptive adjectives, r (14) = .75, p < .05, and positive, non-descriptive adjectives, r (14) = .75, p < .05. Because both rumination and number of depr essive episodes correlated with startle
57 amplitude after a delay (albeit for different categories of adjectives), we examined their relationship by computing their Spearman corre lation, which revealed a near-significant positive association, r (14) = .52, p = .07. Discussion This study attempted to understand in formation-processing abnormalities in depression from the point of view of methodol ogy in wide use in the field of affective neuroscience. Results from the present study partly confirmed our predictions. In the nodelay condition, we found startle potentiation by self-descripti ve adjectives within the control group, consistent with previous fi ndings of greater startle amplitude when attention is turned towards the self (Pan ayiotou & Vrana, 1998). Contrary to our predictions, however, the same effects were not observed within the depressed group, where there were no effects of word category in the no-delay condi tion. Consistent with our predictions, the control group did not show any modulation of startle amplitude after a delay. The depressed group, however, showed startle potentiation after positive adjectives, regardless of self-descriptivene ss. Within the depressed group, rumination correlated with startle amplitude after a delay for the negativ e, non-descriptive adjectives and the positive, self-descr iptive adjectives. Moreover, group differences in startle amplitude for the positive adjectives after a delay disappeared after controlling for rumination. Finally, within the depressed group, the number of depressive episodes correlated with startle amplitude after a delay for the negative, self-descriptive and positive, non-descriptive adjectives. We now discuss each of these findings in turn. Panayiotou and Vrana (1998) proposed that the greater startle amplitude observed under conditions of attention towards the se lf is due to greater arousal induced by increased effortful processing. In support of this interpretation, previous research
58 suggests that, in the absence of sensory e ngagement, increased processing demands are associated with startle potentiation (S eljos, Dawson, & Schell, 1995; Vrana & Lang, 1990). In addition, better memory for informati on processed in reference to the self has been shown in numerous studies (e.g. Symons & Johnson, 1997). In our study, instructions asked participants to determine how often they had used or thought of the word or concept. For self-descriptive adjectiv es, these instructions may have resulted in increased processing involvi ng the self and, as a conse quence, increased startle amplitude. An important question regarding the effects of self-descriptiveness of foreground stimuli on startle amplitude concerns its automaticity. One possibility is that preferential processing of self-descriptiv e words is automatic and independent of experimental instructions. Another possibility is that experimental instructions play an essential role in encouraging or discouraging gr eater effort and processing in reference to the self. The latter is suppor ted by research showing that ex perimental manipulations of sentence processing have a predictable e ffect on startle amplitude (Vrana & Lang). Contrary to our predictions, the depre ssed group and the control group did not show a similar pattern of startle modulati on in the no-delay condition. While the control group showed greater startle amplitude while viewing self-descriptive adjectives, the depressed group did not show such an eff ect. Although this result was unexpected, it raises the possibility that differences in self-focus between depr essed and non-depressed participants may have played a role in st artle modulation. Accord ing to Pyczsinski and Greenberg (1987), maladaptive self-focus plays an important role in the development of depressive episodes, and several studies have shown greater self-foc us in clinical and subclinical samples of depressed individuals (Rude, Gortner, & Pennebaker, 2004; Sloan,
59 2005; E. Watkins & Teasdale, 2004). Thus, de pressed participants, in our study, may have processed all the words in reference to themselves, thereby re ducing the impact of descriptiveness. Because we did not assess ba seline self-focus, this possibility cannot be directly examined based on our data. Another possibility is that depressed individuals might take longer to activate relevant motiva tional states after stimulus exposure. Future studies examining the influence of self-foc us over a long delay are required to better understand the relationship between self-focus and physiological activ ation in depression. After a delay, we had expected greater startle amplitude in the depressed group for negative, self-descriptive and positive, non-de scriptive adjectives. Instead, we observed startle potentiation af ter both self-descriptiv e and non-descriptive positive adjectives. These results are consistent with severa l studies showing abnormal reactivity of depressed individuals to pleasant stimu li only (Shestyuk, Deldin, Brand, & Deveney, 2005; Sloan et al., 2002; Sloan et al., 2001). Moreover, Allen et al. (1999), similar to our study, found potentiation rather than inhibition of startle amplitude by pleasant stimuli in a small subgroup of severely depressed indi viduals. Both sets of findings can be interpreted within th e framework of Â“frustrative nonrew ard,Â” which occurs when an appetitive stimulus is withhe ld for an extensive period of time. In both humans and animals, frustrative nonreward can lead to potentiation, rather th an inhibition, of the startle response by the withheld appetitive s timulus (Drobes, Hillman, Bradley, Cuthbert, & J., 1995; Wagner, 1963). In depression, repeat ed lack of reinforcement and a general sense of hopelessness (Abramson et al., 1989) ma y lead to negative reactions to a wide range of positive stimuli. While Allen et al. observed startle potentiation during exposure to pleasant stimuli, this effect occurred in our study only after stimulus offset, at a time
60 when the stimulus was no longer task-releva nt. Another difference between the two studies is that, in Allen et al., this effect was observe d only for severely depressed participants, while our sample consisted exclusiv ely of participants with mild or moderate levels of depression severity. Taken together these findings suggest that the time course of startle modulation by affective stimuli needs to be examined in conjunction with an assessment of depression severity. In partic ular, it is tempting to speculate from our findings and those of Allen et al. that, as th e severity of depression increases, startle potentiation by positive stimuli becomes more automatic and therefore more immediate. Consistent with our predictions, rumination correlated with certain categories of adjectives in the delay cond ition. Conceptually, rumination a nd related constructs consist of continued cognitive proce ssing of negative thoughts beyon d the mere exposure to the aversive stimulus. Our findings are consis tent with this view, and suggest that physiological responsiveness may Â“trackÂ” th e time course of cognitive processing in individuals with depression. Moreover, we found that group differences in startle amplitude for positive adjectives after a de lay disappeared after rumination was entered as a covariate. Thus, our results are consis tent with the hypothesi s that, in depression, rumination and negative thoughts occurring afte r exposure to positive stimuli contribute to delayed negative physiological activation. Un expectedly, however, the categories of adjectives with which rumination correlated in the delay conditi on were negative, nondescriptive and positive, self-descriptive adj ectives. On the contrary, we had expected significant correlations for negative, self-d escriptive and positive, non-descriptive adjectives, which both have th e potential to threaten oneÂ’s self-concept. The number of depressive episodes, rather than rumina tion, correlated signifi cantly with startle
61 amplitude for negative, self-descriptive and positive, non-descriptiv e adjectives after a delay. Thus, individuals who ha ve experienced numerous episodes of depression may be particularly vulnerable to th reats to their self-concept. Our findings have important implications for future psychophysiological research in depression. Although previous studies ha ve generally found normal affective startle modulation in individuals with mild or m oderate symptoms, we found abnormal startle modulation during and after exposure to ad jectives varying in valence and selfdescriptiveness. We believe that the greater sensitivity of our approach to startle modulation in mild and moderate MDD is due to two factors (1) S timuli were selected based on previous literature for their rele vance to clinical depression, and (2) We examined startle modulation both during exposur e to the stimuli, a nd after the stimuli were no longer task-relevant. Considered within the framework of cognitive rumination, our results suggest that physiological activ ation may track cognitive processing in major depression. Moreover, our study, along with fi ndings from Allen et al. (1999), emphasize the importance of examining th e role of frustrative nonrew ard for a better understanding of the behavioral, cognitive and physiologi cal components involved in the development and maintenance of depression. If, due to rep eated lack of reinforcement (or at least perceived reinforcement), depressed individual s come to experience negative affect after exposure to normally appetitive stimuli, they may become likely to avoid positive experiences, which in turn may prevent them from seeking positive experiences and maintain the depressive state. This view is entirely consistent with behavioral models (Eastman, 1976) and with the hopelessness theo ry of depression (Abramson et al., 1989).
62 Our findings suggest that the startle methodol ogy is an appropriate method for examining the theory of frustrative nonreward and its ap plications to c linical depression. Several limitations should be acknowledged. First, our sample size was limited, and did not allow us to examine the influence of depression severity, number of depressive episodes and other clinical va riables with sufficient power Our participants, however, were carefully selected and formed a reas onably homogeneous group: all met criteria for MDD of mild or moderate severity; all were outpatients, with no recent history of suicide attempt or plan; and there was only one partic ipant with a current comorbid psychiatric diagnosis (social phobia, mild). Second, the number of stimu li per adjective category was also limited (three per category), with the associ ated risk of loss of si gnal-to-noise ratio in our estimates of average startle amplitude pe r category. Of note, ot her studies using the startle methodology have also included a limite d number of clinica lly-relevant stimuli. Cuthbert et al. (2003), for instance, us ed two individually generated sentences corresponding to each participantÂ’s worst fear in a sample of patients with various anxiety disorders. Clearly, however, future studies using larger samples and a larger number of clinically relevant stimuli will be needed to confirm and extend the present results. A third limitation is that four of th e participants within the depressed group were on antidepressant medications. Several studies have shown that some antidepressants and antianxiety agents can block startle potenti ation by threat-related stimuli (Harmer, Shelley, Cowen, & Goodwin, 2004; Patrick, Berthot, & Moore, 1996). In our study, however, we failed to find an effect of medication status on startle amplitude. Moreover, the depressed group, in our study, showed startle potentiation after posi tive stimuli, rather than exaggerated potentiation by negative stimuli.
63 In conclusion, our study suggests several abnormalities of star tle potentiation and inhibition by affective stimuli in major depression. In the no -delay condition, the depressed group did not show any significant startle modulation, while the control group showed startle potentiation by self-descriptive adjectives. The difference between the two groups may be related to the exaggerated tende ncy to self-focus or other abnormalities in the processing of the self that have been obs erved previously in de pression. In the delay condition, the depressed group showed startle potentiation by positive adjectives. Within the depressed group only, rumination and the nu mber of depressive episodes correlated significantly with startle potentiation after a delay, albeit for different categories of adjectives. Our results point to the need of using clinically relevant stimuli and allowing sufficient processing time after stimulus offset for a better understanding of physiological reactivity in depression. They suggest that, in depression, ru minative tendencies may be paralleled by physiological reactivity long af ter stimulus offset. The negative reaction thus experienced by depressed individuals after viewing positive stimuli may play an important role in the avoidance of rewa rding activities and the maintenance and worsening of depressive episodes.
64 CHAPTER 4 SUBLIMINAL AND SUPRALIMINAL PR IMING OF EMOTIONAL WORDS IN DEPRESSION Introduction One of the most well-validated models of depression is the cognitive model, developed by Aaron Beck (1967). According to this model, depressed individuals display systematic negative biases in their perception and interpretation of life events. In particular, they are more likely to attend t o, and to engage in sustained processing of, negative events, and they are also more likely to interpret life events as negative. The cognitive model has been instrumental in th e development of cognitive therapy (CT). CT has as its main goals to challenge and rectify the negative biases and interpretations displayed by depressed individuals. Within the last 20 years, CT has been firmly established as an effective and powerful tr eatment of depression, comparable in its efficacy to pharmacological treatment and w ith other empirically-supported psychosocial treatments of depression such as interper sonal therapy and behavior therapy (Dobson, 1989; Gloaguen, Cottraux, Cucherat, & Bl ackburn, 1998). The addition of CT to pharmacological treatment has been shown to reduce relapse rates (Gloaguen et al., 1998). Moreover, in at least one large mu lti-center treatment study of chronically depressed individuals, the comb ination of pharmacological tr eatment with the cognitivebehavioral system of psychotherapy, which in cludes elements of CT, problem-solving and interpersonal therapy, was more effectiv e than pharmacotherapy alone (Keller et al., 2000).
65 Information Processing Biases in Depression At its core, the cognitive model is based upon the assumption that depressed individuals show preferential processing of negative information. However, defining Â‘negativityÂ’ and the information processes that are dedicated to it ha s been the subject of much research and debate. Initial research in this area was influenced by BowerÂ’s network model of the effects of mood on information processing (Bower, 1981). This model was largely based on findings of moodcongruent memory in depressed and nondepressed individuals and in normal participants receivi ng pleasant or unpleasant mood induction (Bower, Monteiro, & Gilligan, 1978; Teasdale & Fogarty, 1979). Based on these findings, Bower suggested that m ood states activate cognitive schemas and memories previously associated with this mood state. As it relate s to depression, this model predicts that vulnerability to depr ession could be related to the strength of association between negative mood states and cognitive distortions. Bower suggested that, in depressed individuals, negative sc hemas are highly inte rconnected and easily activated by internal or extern al stimuli. There has been mu ch support for this hypothesis. For example, Segal and Gemar (1997) found increased priming (as measured by increased reaction time in a Stroop task) of self-descriptive negative adjectives by selfdescriptive negative sentences in a sample of depressed individuals. After 20 sessions of cognitive therapy, reduction in depressive sy mptomatology was a significant predictor of the amount of priming for negative, but not positive, adjectives. Ingram et al. (1995) found support for better incidental and inten tional learning of words of negative content in depressed individuals, s uggesting broad advantages in the processing of negative stimuli in depression.
66 Although it has enjoyed broad empirical suppor t, BowerÂ’s model, as pointed out by Teasdale and Barnard (1993), also suffers fr om several limitations. First, Bower and his colleagues, despite numerous attempts, have failed to find support for the mood statedependent hypothesis, which states that congruence between mood at encoding and mood at retrieval should pr oduce better recall. Second, although there are reliable effects of mood on judgments and evaluations, these e ffects can be eliminated by making the source of the mood state salient to participan ts (e.g., W. D. Scott & Cervone, 2002). This finding, which has been widely replicated, s uggests that the eff ects of mood on judgment are at least in part mediated by an Affect-a s-Information mechanism (i.e., when asked to make a judgment or provide an evalua tion, people ask themselves how they feel about it, thereby using their own mood state as a source of information), rather than by automatic association. Third, whereas BowerÂ’s model pr edicts improved attention and memory for stimuli of negative content in depressed indi viduals, findings regardi ng attentional biases in depression have been rather mixed. Fo r example, Bradley and Mathews (1983) found that depressed participants s howed better recall of negative adjectives after self-referent encoding, but not other-referent encoding. However, their reac tion times (RTs) to negative, neutral and positive adjectives were similar. These limitations suggest that the network analogy is insufficient to account fo r the complexity of findings on informationprocessing biases in depression. Automatic and Controlled Processing Williams, Watts, MacLeod and Mathews (1988) and Mineka and Sutton (1992) reviewed findings of attentional and memo ry biases in depression and anxiety and concluded that (a) there is no evidence of automatic at tention bias in depressed individuals, either for threat-related or depression-related words, (b) depressed
67 individuals do not show evidence of faster le xical decision for depression-related words, (c) depressed individuals show better recall of negative than neutral or positive information, especially if the information is self-referent. Based on this review of existing data, Williams et al. (1988) suggested that depressed individuals do not show an automatic bias toward depressed-related stimuli. Rather, information-processing biases in de pression occur at the elaboration stage and involve controlled or strategic, aspects of proces sing (Hasher & Zacks, 1984). They contended that, in contrast, an xiety is associated with automatic as well as controlled information-processing biases. To test th is hypothesis, Mogg, Bradley and Williams (1995) used a visual dot probe task with subliminal or supraliminal presentation of the prime. In the subliminal condition, anxious pa rticipants, but not depressed participants, showed preferential processing of threat-rel ated and depression-related words. In the supraliminal condition, both groups shifted their attention toward depression-related words, and only depressed individuals shifted their attention toward threat-related words. These findings provided strong support for the model developed by Williams et al. (1988). Other studies, however, have challenged this model. In order to examine automatic processing in depression, Bradley, Mogg and W illiams (1994) used subliminal repetition priming in students with high and low scores on the Profile of Mood States (McNair, Lorr, & Droppelman, 1971). In their study, the subliminal priming condition consisted of a brief presentation of a prime word, followed by a mask (a random letter string) and then a target. When they saw the target, participan ts were asked to make a lexical decision (i.e., indicate whether the targ et was a word or not a word). Priming in this condition was
68 called Â“subliminalÂ” because the prime was pr esented for such a brief duration that participants presumably could not identify it. This brevity en sured that any effects due to subliminal priming could safely be consid ered automatic. Using a repetition priming paradigm in which the prime and the target we re the same word, Bradley et al. found that participants with high negativ e affect showed greater sub liminal priming for negative words, while participants with low negative affect showed greater priming for positive words. The same interaction effect was obser ved for supraliminal priming, a condition in which the primes were presented for longer durations. Moreover, increased priming for depression-related words was significantly correlated with depre ssion, but not anxiety symptomatology. Other studies (B. P. Bradley, Mogg, & Millar, 1996; B. P. Bradley, Mogg, & Williams, 1995) later re plicated and extended thes e findings to clinically depressed participants, and Scott, Mogg a nd Bradley (2001) found evidence of increased subliminal semantic priming of depression-re lated words (i.e., when a depression-related target was preceded by a depression-related prime) in students with high depression symptomatology. Taken together these findings have provided evidence for the presence of automatic biases in depressed and dys phoric individuals. They suggest that the distinction between automatic and controlled processing is insuffi cient to explain the distinct information-processing biases observed in depression. The goal of our study was two-fold. First, we sought to replicate the findings obtained by Bradley and colleagues (B. P. Bradle y et al., 1996; B. P. Bradley et al., 1994, 1995), using a different set of words and a slightly different methodology. Given the theoretical and practical importa nce of determining whether de pression is associated with automatic processing biases for negative information, such a replication by an
69 independent research team was clearly wa rranted. Second, we attempted to determine some of the factors contributi ng to subliminal and supralimin al priming in depression. In particular, we hypothesized that the extent of supraliminal priming for negative words shown by an individual may depend on her or his propensity to engage in ruminative thinking. Rumination at encoding may result in strengthened activa tion of the lexical representations, which would then facilitate responding a nd decrease RTs when the stimulus re-appears for the lexical decision ta sk. In contrast, subliminal priming, which occurs very quickly, may represent a dis tinct phenomenon, independent of ruminative tendencies. Method Overview Clinically depressed and never-depressed individuals recruited within the community participated in an experiment i nvolving a semi-structured clinical interview, questionnaires, and a computer-based primi ng task. For the priming task, participants made timed lexical decisions on negative, ne utral and positive words that were unprimed, subliminally primed or supraliminally pr imed. Subliminal and supraliminal priming scores were computed separately for each participant and word category by subtracting the RT of the subliminally or supralimina lly primed words from the RT of the unprimed words. Participants Participants were recruited through f liers posted within the community and advertisements in local newspapers. All pa rticipants were betw een 18 and 60 years old. Participants within the depr essed group all met diagnostic criteria for a current Major Depression Disorder (MDD), w ithout Psychotic Features, ac cording to the Diagnostic
70 and Statistical Manual of Mental Disorder s, Fourth Edition (American Psychiatric Association, 2000). Exclusion cr iteria for the depressed group included (a) A history of psychotic disorder, bipolar disorder, or obs essive-compulsive disorder (OCD), (b) Any history of neurological disorder such as epilepsy or trauma tic brain injury (c) A history of substance abuse or dependence within the la st six months), (d) A current diagnosis of post-traumatic stress disorder (PTSD), (f) A hi story of recent suicide attempt (less than six months), or plans or intent to commit suic ide, (g) Use of medications known to affect cognitive functioning, such as benzodiazepines or certain anti-psychotic medications. For the control group, criteria (b)-(g) were applie d and, in addition, elig ible subjects were required to have no history of mood disorder or anxiety disord er, and had to have a score lower than 13 on the Beck Depression Inventory Â– Second Edition (BDI-II). Depressed individuals who were taking an ti-depressant medications were included in the study as long as they had been on a st able dosage of medicat ion for at least six weeks. Given the high comorbidity between depression and anxiety disorders (Hasin, Goodwin, Stinson, & Grant, 2005), we also incl uded depressed individu als with a current comorbid anxiety disorder (except for OCD a nd PTSD) or with a history of PTSD in full remission, as long as it was determined th at the anxiety disorder was not a major contributor to the current majo r depressive episode. In cases of diagnostic uncertainties, consultation with a board certified clinical psychologist was sought, and a generally conservative approach was take n to participantsÂ’ inclusion. The initial sample consisted of 26 depre ssed and 26 control part icipants. Computer failure resulted in the discarding of two of th e depressed participants In addition, three of the control participants were excluded ba sed on a high BDI-II scor e, and another based
71 on a current diagnosis of speci fic phobia. The final sample consisted of 24 depressed and 22 control participants. Comorb id disorders included a history of substance abuse or dependence (eight depressed pa rticipants and one control participant), PTSD in full remission (two depressed participants), gene ralized anxiety disord er (one depressed participant), social phobia (one depressed participant), and pa nic disorder with agoraphobia, in partial remission (one depres sed participant). Usi ng DSM-IV criteria, 9 of the 24 depressed participants were categor ized as mildly depressed, while the other 15 were categorized as moderately depressed. Semi-Structured Interview Participants completed a semi-structured interview that served as a basis for all Axis I diagnoses. All interviews were conducted by the first author using the Mood Disorders, Psychotic Screen, SubstanceRelated Disorders and Anxiety Disorders modules of the Structured Clinical Inte rview for the DSM-IV-TR (SCID-IV, First, Spitzer, Gibbon, & Williams, 1996). Interviews we re audiotaped, and 10 interviews from the depressed group and 8 interv iews from the control group were randomly selected and recoded by the second author, blinded to orig inal ratings. Inter-rater agreement was 98% for individual SCID-IV items, and 95% for Axis I diagnoses. The less-than-perfect agreement for diagnoses was due to one depr essed participant for whom the second, but not the first author, diagnosed a distant history of alcohol ab use. There was perfect interrater agreement regarding MDD severity (3 mild, 7 moderate). Questionnaires In each session, participants completed a number of questionnaires, including the BDI-II, the Beck Anxiety Inve ntory (BAI), and the Rumination Responses Scale (RRS). The BDI-II is a widely used measure of de pression symptomatology, with internal
72 consistency and test-retest re liability both above .9 (Bec k, Steer, & Brown, 1996). The BAI is a 21-item self-report measure of a nxiety symptoms, which was developed to address the problem of the overlap between ot her measures of anxi ety and measures of depression (Beck & Steer, 1990). The scale has good internal consiste ncy (.92) and testretest reliability (.75), and its convergent and discriminant validity are supported by higher scores in patients diagnosed with an an xiety disorder than patients diagnosed with major depression (Beck, Epstein, Brown, & St eer, 1988). The RRS is a 22-item selfreport instrument assessing the frequenc y of thoughts about oneÂ’s symptoms of depression. Its internal consistency ranges from .89 to .90 (Nolen-Hoeksema & Morrow, 1991; Treynor, Gonzalez, & Nolen-Hoeksema 2003; Wenzlaff & Luxton, 2003) and its test-retest reliability was .67 in a large community sample (Nolen-Hoeksema, Larson, & Grayson, 1999). It has been shown to predic t depressive symptomatology prospectively after traumatic events (Nolen-Hoeksema & Morrow, 1991) and the death of a loved one (Nolen-Hoeksema, Parker, & Larson, 1994). Priming Task Materials Three types of words were se lected for the priming task : depression-relevant words (e.g., depressed, inadequate, useless), categor ized neutral words (household words such as carpet, hallway, basement), and positiv e words (e.g., happy, successful, joy). The words were selected from a large pool of wo rds collected from prior experiments (B. P. Bradley & Mathews, 1983; K. M. Scott et al., 2001) and completed with words from the Affective Norms for English Words (ANEW, M. M. Bradley & Lang, 1999). Ten nave judges (5 men, 5 women; 9 Caucasian, 1 Afri can-American) rated each adjective on two 9-point scales for valence and arousal. For each word category, we selected 4 lists of 16
73 words and 2 lists of 8 words. All lists were matched for length, F (17, 222) = .09, p > .1, and frequency, F (17, 222) = 0.003, p > .1, using Kucera and Fr ancis (1967) norms for ratings of frequency. Within each category, lists were also matched for valence and arousal, all p Â’s > .1, and positive and negative lists were matched for arousal, F (11, 148) < .14, p > .1. Lexical decision task Our task closely followed the methodology of previous studies, which showed increased priming for negative words in de pressed and dysphoric individuals (B. P. Bradley et al., 1996; B. P. Bradley et al ., 1994, 1995). These studies, however, suffered from one important limitation: in the lexical decision task, primes preceding word targets could be either words or non-words, wher eas primes preceding non-word targets were always non-words. As a result, it was possible fo r participants to pred ict the nature of the target (and therefore to make the lexical d ecision) based entirely on the nature of the prime. To avoid this confound, we include d conditions where non-word targets were preceded by word primes. The priming task consisted of two sequen tial, structurally identical parts, each consisting of a block of 24 supraliminal pr imes followed by the lexical decision task proper. For each block, participants first view ed 24 words, presented one at a time on the computer screen in uppercase letters for 5000ms. For each word, participants were asked to indicate Â“how frequently [they thought ] of the word or conceptÂ” (1=rarely, 2=sometimes, 3=often). This initial viewing was designed to pre-expose participants to these words before the lexical task proper. In the following, we label this pre-exposure Â“supraliminal priming.Â” After viewing the s upraliminal primes, participants completed the lexical decision task, which consisted of 144 tr ials. In each trial, participants viewed a
74 fixation cross (800ms), followed by a prime (26ms), a mask (26ms), and a target. ParticipantsÂ’ task was to determine as quickly as possible whether the target was a word or not a word by pressing an appropriate key. The target remained on the screen until the participant responded (see Figure 4-1). Each trial was followed by a 2000 ms inter-trial interval. Fixation cross (800ms) Mask (28ms) Prime (28ms) Target (until response) Figure 4-1. Trial structure in the lexical decision task In each trial, the prime and the mask we re in uppercase letters and of the same length of the target, which was in lowercase letters.4 Moreover, the mask always consisted of a random letter string. Half of the targets were words and the other half were pronounceable, graphemically correct non-words. For target words, there were three conditions for valence (negative, neutral and positive), and three priming conditions (unprimed, subliminally primed, and supralim inally primed). In the unprimed condition, the prime was a random letter string, and th e target did not be long to the set of supraliminal primes presented before the lexical decision task. In the subliminally primed condition, the prime was the same word as the ta rget, and they did not belong to the set of supraliminal primes presented before the le xical decision task. In the supraliminally primed condition, the prime was a random letter string, and the target was the same word as one of the supraliminal primes presente d just before the le xical decision task. With these three conditions, one third of the word targets were preceded by word primes, while the other two thirds were preceded by primes consisting of random letter 4 The uppercase/lowercase manipulation was introduced as a way to favor lexical rather than perceptual priming.
75 strings. As noted before, any unbalance betw een word and non-word targets in terms of the nature of the prime would allow participants to guess the nature of the target before it appeared on the screen. In order to eliminat e this possibility, we introduced a similar proportion of word primes (one third) before the non-word targets. T hus, the priming task consisted of 13 different c onditions: non-word targets with non-word primes (1 condition), non-word targets with word primes (3 conditions, depending on the valence of the prime), unprimed word targets (3 conditions ), subliminally primed word targets (3 conditions), and supraliminally prim ed word targets (3 conditions). The 4 lists of 16 words from each category served as targets and/or primes in the unprimed, subliminally primed and supralimin ally primed conditions, and as primes in the non-word target trials. The role and orde r of the lists were counterbalanced across participants, and the order of appearance of the words within list was randomized. Stimulus randomization, timing and presentati on were controlled by E-Prime software (Schneider, Eschman, & Zuccolotto, 2002). Awareness checks The awareness checks were introduced to ensure that participants could not consciously perceive the prime words duri ng the lexical decision task. Participants completed two awareness checks: a lexical de cision awareness check of 48 trials, and a valence decision awareness check of 24 trials. For the lexical decision awareness check, each trial consisted of a fixation cross (800ms), followed by a brief prime (28ms), a mask (28ms) and a row of Xs of the same length as the prime. Half the primes were wo rds, while the other half were random letter strings. Participants were asked to indicate verbally whether the prime was a word or not a word. The experimenter manually record ed their response. The valence decision
76 awareness check had a similar structure, but all the primes were words and participants were asked to indicate (or guess) the nature of the prime. For both awareness check tasks, participants were told that the task was desi gned to test their Â“unc onscious perceptionÂ” of the words, and that they had to guess even if they did not know at all. Two lists of 8 words from each category were used for the awareness checks. The order of the awareness checks and the role of the lists were counterbalanced across participants. Data Reduction and Analyses For the priming task, error trials and tria ls with decision late ncies outside the 2002000 ms range were removed from subsequent an alyses. Median RTs for valid trials were used, given their robustness to outliers and their relatively good power (Ratcliff, 1993). For each participant and word category, a s ubliminal priming score was computed by subtracting the median RT of the sublimina lly primed words from the median RT of the unprimed words. Similarly, a supraliminal priming score was computed by subtracting the median RT of the supraliminally primed words from the RT of the unprimed words. Given the large quantity of data genera ted by the priming task, our analytical strategy focused on specific contrasts rath er than omnibus analyses of variance (ANOVAs). We tested the following predictions: 1) Consistent with prior studies, depresse d participants will have slower overall RTs than control participants. This prediction was tested by examining the main effect of group in a 2 (Group) x 13 (Condition) ANOV A including all the conditions of the priming task. 2) Subliminal and supraliminal priming will result in lower RTs across groups. In order to test this prediction, we computed an overall RT (a veraged across valences) for
77 the unprimed, subliminally primed and supr aliminally primed word targets. For subliminal priming (respectively, supraliminal priming), we then conducted a paired t-test comparing RTs for the unprimed word targets and the subliminally primed (respectively, supraliminally primed) word targets. 3) The main prediction of the study was th at depressed participants would show greater priming for the negative words, while control participants would show greater priming for the positive words. To test th is prediction, we examined the interaction effects in two separate 2 (Group) x 2 (Valence: Negative, Positive) ANOVAs for subliminal and supraliminal priming. We pred icted that both these interactions would be significant, with a pattern of resu lts consistent with our prediction 4) To test the prediction of an associa tion between rumination and supraliminal, but not subliminal priming, for negative words, we examined the Spearman correlation between the RRS and priming scores for negative words. Results ParticipantsÂ’ Characteristics Demographic variables and scores on the BDI, BAI and RRS for the two groups are reported in Table 4-1. Table 4-1. Demographic characteris tics and questionnair e scores by group Depressed Control Comparison Females Males Females Male Gender 19 5 17 5 2(1) = 0.02 Mean SD Mean SD Age 35.54 13.78 32.27 13.36 t (44) = 0.82 Education 14.71 2.66 15.36 2.44 t (44) = -0.87 BAI 22.04 11.71 6.73 5.64 t (44) = 5.57** BDI-II 28.17 8.18 3.95 3.55 t (43)a = 12.77** RRS 60.96 9.32 37.86 11.52 t (44) = 7.50** a One participant did not fully complete the BDI-II; ** p < .01
78 Both groups were predominantly female, and they did not differ in terms of gender, age or education. As expected, however, depres sed participants had significantly greater BDI-II, BAI and RRS scores than control participants. Awareness Check ParticipantsÂ’ error rates for the lexical a nd valence awareness checks are reported in Table 4-2. The overall error rate for the le xical decision awarene ss task was 40.14%, and significantly lower than the chance error rate of 50%, t (44) = -8.93, p < .01. The overall error rate for the valence decision task was 54.08%, and also significantly lower than the 66.67% chance error rate, t (45) = -5.68, p < .01. Thus, participants were able to guess above chance levels for both the lexical de cision and the valence decision awareness checks. In order to examine potential diffe rences between groups and adjectivesÂ’ categories, we conducted two separate 2 (Group) x 3 (Category) ANOVAs with error rates for the lexical decision and valence decisi on awareness tasks, re spectively. None of the main effects or interaction effects reached significance, all p Â’s > .05. Depressed and control participants also did not differ in terms of their error rates for non-words on the lexical decision awareness task, t (43) = 0.64, p > .1. Table 4-2. Error rates for the awareness checks by group Depressed ( N = 24)b Control ( N = 22) Meana SD Meana SD Lexical Decision, Non Words 37.86 13.41 35.23 14.07 Lexical Decision, Negative Words 37.50 13.59 45.45 23.32 Lexical Decision, Neutral Words 48.37 16.56 46.59 21.19 Lexical Decision, Positive Words 45.11 16.32 39.20 22.59 Valence Decision, Negative Words 51.04 13.41 55.11 19.54 Valence Decision, Neutral Words 57.81 20.79 50.57 23.62 Valence Decision, Positive Words 58.33 22.32 51.14 21.45 a Chance error rates are 50% for the lexical deci sion task, and 66.67% for the valence decision task; b For one depressed participant, data from the lexical decision awareness task were not available due to equipment failure.
79 Priming Task Preliminary analyses The overall error rates for the lexical d ecision task were 2.6% for the non-words (3.15% for the depressed group, 1.99% for the control group), and 1.72% for the words (1.36% for the depressed group, 2.11% for the control group). RTs for each group and stimulus category are reported in Table 4-3. Although it was in the predicted direction, the main effect of gr oup was not significant, F (1, 44) = 2.43, p > .1. Table 4-3. RT by group and stimulus category Depressed ( N = 24) Control ( N = 22) Mean SD Mean SD Non-words, unprimed 940.4 220.7 841.4 183.9 Non-word, negative subliminal primes 969.5 244.2 826.3 175.4 Non-words, neutral subliminal primes 952.8 233.8 821.5 169.9 Non-words, positive subliminal primes 929.1 230.8 843.6 212.3 Negative words, unprimed 773.0 137.4 734.4 101.7 Neutral words, unprimed 777.2 110.6 752.1 118.2 Positive words, unprimed 758.6 121.2 720.7 107.8 Negative words, subliminal primes 751.8 127.9 725.5 135.6 Neutral words, subliminal primes 753.2 119.3 724.7 123.0 Positive words, subliminal primes 758.1 141.8 696.4 110.6 Negative words, supraliminal primes 754.9 138.7 708.0 121.1 Neutral words, supraliminal primes 747.8 119.6 701.6 96.5 Positive words, supraliminal primes 709.4 115.6 691.0 103.2 In order to determine the presence of an overall priming effect, we computed the overall RT, averaged across groups and valenc es, for each of the priming conditions. The means were: 753.4 ms ( SD = 110.0 ms) for the unprimed words, 735.8 ms ( SD = 120.1 ms) for the subliminally primed words, and 719.6 ms ( SD = 111.2 ms) for the supraliminally primed words. Both subliminal priming, t (45) = 3.56, p < .01, and supraliminal priming, t (45) = 6.42, p < .01 resulted in reduced RT compared to the unprimed condition.
80 Effects of group and valence on priming The average subliminal and supraliminal priming for each group and word valence are reported in Table 4-4. Contrary to our predictions a 2 (Group) x 2 (Valence: Negative, Positive) for subliminal primi ng did not yield a significant interaction, F (1, 44) = 1.73, p > .1. The main effects of group, F (1, 44) = 0.22, p > .1, and valence, F (1, 44) = 0.04, p > .1, were also non-significant. Simila rly, a 2 (Group) x 2 (Valence: Negative, Positive) for supraliminal priming did not yield a significan t effect of group, F (1, 44) = 0.93, p > .1, valence, F (1, 44) = 0.93, p > .1, or interaction between group and valence, F (1, 44) = 1.42, p > .1. Table 4-4. Subliminal and supral iminal priming by group and category Depressed ( N = 24) Control ( N = 22) Mean SD Mean SD Negative words, subliminal priming 21.1 69.6 8.9 64.1 Neutral words, subliminal priming 24.0 60.9 27.5 67.9 Positive words, subliminal priming 0.4 61.8 24.3 52.7 Negative words, supraliminal priming 18.1 70.5 26.4 79.8 Neutral words, supraliminal priming 29.4 77.1 50.5 69.1 Positive words, supraliminal priming 49.1 46.1 29.7 66.0 Relationship with Self-Report Measures We tested the specific prediction that supr aliminal but not subliminal priming for negative words would be associated with RRS scores within the depressed group. Contrary to our predictions, there was no association between RRS scores and supraliminal priming for negative wo rds in the depressed group, Spearman r (24) = -0.29, p > .1. The association between RRS score and subliminal priming for negative words in the depressed group was also non-significant, Spearman r (24) = -0.19, p > .1. Discussion In our study, we failed to replicate prev ious findings of an interaction between valence and depression in the subliminal a nd supraliminal priming of emotional words
81 (B. P. Bradley et al., 1996; B. P. Bradley et al., 1994, 1995). For subliminal priming, our findings were in the expected direction: depressed participan ts showed slightly greater priming than control particip ants for negative words, wh ile the opposite was true for positive words. For supraliminal priming, th e effect was actually reversed: depressed participants showed less primi ng than control participants for negative words, while the opposite was true for positive words. In both cases, however, the effects were small, with large standard errors, and none of the main effects or interact ion effects reached significance. These non-significan t findings were obtained de spite the fact that the priming manipulations were generally effective in reduc ing RTs. Contrary to our predictions, we also failed to find any rela tionship between self-reported rumination and supraliminal priming for negative words. There are several potential reasons for our negative findings. First, most of the effect sizes in Bradley et al. (1995) were rather modest (see Table 4-5). Except for the supraliminal priming for negative words, our own effect sizes fell within the confidence intervals of correspondi ng effects in Bradley et al. Thus our findings, rather than being anomalous, may fall in the lower end of the c onfidence intervals typica l for these effects. Second, the stimuli used in our study were di fferent from those used in Bradley et al. (1996; 1994; 1995). We included more stimu li than they did per condition, and some of the negative words we selected (e.g., emba rrassed, pain) may not have been relevant to individuals with depression. Third, the two studies in which Bradley a nd colleagues reported subliminal priming biases in depression suffered from a met hodological limitation: in both cases, word targets could be preceded by word or non-wo rd primes, whereas all non-word targets
82 were preceded by non-word primes. Thus, the nature of the prime could inform participants about the nature of the target. If participantsÂ’ perception of a word prime had induced them to assume th at a target was forthcomi ng, group differences in the perception of negative and positive word primes, rather than priming per se, could have been responsible for the eff ects observed in these two studi es. In our study, we avoided this confound by introducing an equal number of word primes (one third) before word and non-word targets. By so doing, we may ha ve reduced the differences between groups for subliminal priming. Table 4-5. Effect sizes and confidence interv als of the difference between the depressed and the control groups in Bradley et al. (1995) Bradley et al. (1995)b Present Study Effect Sizea 95% Confidence Interval Effect Sizea Negative words, subliminal priming 0.49 [-0.17, 1.14] 0.18 Positive words, subliminal priming -0.09 [-0.73, 0.56] -0.41 Negative words, supraliminal priming 0.68 [0.02, 1.34] -0.11 Positive words, supraliminal priming 0.13 [-0.72, 0.77] 0.35 a Uncorrected; b A more efficient strategy for comparing our findings to those of Bradley et al. would have been to compute the effect size of the interaction between word valence and group. Unfortunately, the data in Bradley et al. did no t allow us to compute the effect size of the interaction. Fourth, differences in the na ture and severity of the symptoms experienced by the depressed participants may have contribute d to the different fi ndings. In our study, all depressed participants were in the mild or moderate range of depression severity according to DSM-IV criteria (American Psyc hological Association, 2000). Bradley et al. (1996; 1995) did not specify depression severity according to DSM-IV criteria. However, the average score of the depressed group on the Beck Depression I nventory (Beck, Ward, Mendelson, Mock, & Erbaugh, 1961) in their st udy was 25.1, roughly comparable to the average BDI-II score of 28.2 obtained in our st udy. Thus, it does not appear that the two studies differed dramatically in terms of symptoms severity. Other differences (e.g.,
83 medications, inpatient vs. oupatie nt status, comorbid diagnoses) may have been present, but because Bradley et al. did not report thes e data, we cannot ascerta in this possibility. The fact that participants performed above chance le vels during the awareness checks is clearly a limitation of our study. In Bradley et al. (1996; 1994; 1995), participants performed at chance levels dur ing the awareness checks. The discrepancy between their results and ours is likely due to a difference in methodology: for the lexical decision task, participants in their study had to dis tinguish between words and graphemically-correct, pronounceable non-words, whereas, in our study, participants had to distinguish between words and random lette r strings. The task in our study was much easier, which probably resulted in the above -chance performance. More troubling is the fact that participants pe rformed above chance levels, in our study, for the valence decision awareness check. Clearly, participants we re able to identify some of the words, or at least their emotional t one, even at the very short exposure duration used in our experiment. Given the fact that Bradley et al. replicat ed their findings for clinically depressed individuals in two independent studies (B. P. Bradley et al., 1996; B. P. Bradley et al., 1995), it is likely that their results were not si mply due to chance. For subliminal priming, our results were in the expect ed direction, and within the c onfidence intervals reported in the studies by Bradley et al. Thus, a likely ex planation is that subliminal priming biases are present in depression, but of a small ma gnitude, resulting in low replicability. For supraliminal priming, our results fell outside of the confidence in terval reported in Bradley et al. (1995). Differences in stimulus selection, participantsÂ’ characteristics or other variables were likely responsible fo r this discrepancy, but are unknown at this
84 point. Further research is needed to confirm the existence of supraliminal priming biases in depression, and to determine the moderato rs affecting their pr esence and magnitude.
85 CHAPTER 5 GENERAL DISCUSSION The previous chapters have attempted to answer some unresolved questions pertaining to information-processing bias es in depression. Our sample consisted exclusively of individuals who met criteria for MDD of mild or moderate severity. We found that explicit memory biases, at least recognition memory, were better accounted for by response bias than by increased processing of negative, self-descriptive information. Using the startle methodology, we observed a re versal of startle modulation by positive stimuli after a delay, suggesting that depre ssed individuals may ruminate and experience unpleasant physiological ac tivation after, but not during, expo sure to normally appetitive stimuli. We were, however, unabl e to replicate previous findi ngs of enhanced subliminal and supraliminal priming for negative word s in depression, although the effects in our study were in the predicted direction. Taken together, these findings suggest several conclusions and directions for future research. First, they point to the im portance of stimulus selection for assessing information-processing biases in depressi on. In our experiment on recognition memory, we found differences between depressed and never-depressed participants only for selfdescriptive adjectives. For nondescriptive adjectives, the tw o groups did not differ. In our examination of startle modulation, we found differences between the depressed and the control groups while and after viewing emotional adjectives, despite previous research suggesting that mildly and mode rately depressed indi viduals show normal affective startle modulation. Compared to ot her studies, which used pictures or movie
86 clips, we used self-descriptiv e and non-descriptive adjectives, which have yielded some of the most robust information-processing bi ases in depression (Mineka & Sutton, 1992; Williams et al., 1988). In our third experiment however, we did not find any significant differences between depressed and never-dep ressed individuals in subliminal or supraliminal priming to negative and positive ad jectives. Interestingly, this experiment is the only one for which the adjectives were not selected based on th eir descriptiveness. Had we selected adjectives based on self-des criptiveness, significan t biases between the groups might have emerged. The importance of selecting depression-re levant adjectives has a long history in the lite rature on information-processi ng biases in depression. Derry and Kuiper, for instance, showed enhanced recall only for self-d escriptive negative adjectives (Derry & Kuiper, 1981; Kuiper & Derry, 1982). Moreover, Watkins, Mathews, Williamson and Fuller (1992) showed better re call of depression-related but not threatrelated words in depressed individuals. These findings and ours, suggest that information-processing biases in depre ssion are specific to depression-relevant adjectives, rather than to all negative adjectives. Another question stemming from our rese arch is whether information-processing biases in depression are better characterized by differential processing of negative or positive adjectives. Using a behavioral formula tion, this distinction would be equivalent to describing depression as involving great er sensitivity to aversive stimuli or lower sensitivity to appetitive stimuli. Of course these two possibilities are not mutually exclusive, and behavioral theories of de pression have suggested that both could be present (e.g., see Eastman, 1976). On this ques tion, findings from two of our experiments suggest a different picture: for explicit memory, we found differential response bias
87 between groups only for the negative adjec tives, whereas only the positive adjectives discriminated between depressed and never-dep ressed participants for startle modulation. The discrepancy between these two sets of findi ngs in many ways reflects the state of the literature. Some studies sugge st that depression is best characterized by increased processing of negative information, with normal processing of positive information (e.g., Derry & Kuiper, 1981; Deveney & Deldin, 2004 ; Siegle, Steinhauer, Carter, Ramel, & Thase, 2003; Siegle, Steinhauer, Thase, St enger, & Carter, 2002). Others, however, suggest that reduced processing of positive information is present, with an absence of increased processing of negative informa tion (e.g., Allen, Trinder, & Brennan, 1999; Deldin, Keller, Gergen, & Miller, 2000; Sl oan, Strauss, & Wisner, 2001). An important goal for future research is to better char acterize the circumstances under which depressed individuals exhibit information-processing biases for positive and negative material, respectively.
88 LIST OF REFERENCES Abramson, L. Y., Metalsky, G. I., & Alloy, L. B. (1989). Hopelessness depression: A theory-based subtype of depression. Psychological Review, 96 (2), 358-372. Allen, N. B., Trinder, J., & Brennan, C. (1999) Affective startle modulation in clinical depression: Preliminary findings. Biological Psychiatry, 46 (4), 542-550. American Psychiatric Association. (2000). Diagnostic and statistical manual of mental disorders (4th ed.). Washington, DC: Author. Anderson, J. R., & Bower, G. H. (1972). R ecognition and retrieval processes in cued recall. Psychological Review, 79 97-123. Beck, A. T. (1967). Depression: Clinical, experime ntal, and theoretical aspects. New York: Harper & Row. Beck, A. T. (1976). Cognitive therapy and the emotional disorders Madison, CN: International University Press. Beck, A. T., Epstein, N., Brown, G., & Steer, R. A. (1988). An inventory for measuring clinical anxiety: Psyc hometric properties. Journal of Consulting and Clinical Psychology, 56 (6), 893-897. Beck, A. T., & Steer, R. A. (1990). Manual for the Beck anxiety inventory San Antonio, TX: Psychological Corporation. Beck, A. T., Steer, R. A., & Brown, G. K. (1996). Manual for the Beck depression inventory-II (2nd ed.). San Antonio, TX : Psychological Corporation. Beck, A. T., Ward, C. H., Mendelson, M., Mo ck, J., & Erbaugh, J. (1961). An inventory for measuring depression. Archives of General Psychiatry, 4 561-571. Berenbaum, H., & Oltmanns, T. F. (1992). Emotional experience and expression in schizophrenia and depression. Journal of Abnormal Psychology, 101 (1), 37-44. Bock, M. (1986). The influence of emotiona l meaning on the recall of words processed for form or self-reference. Psychological Research, 48 (2), 107-112. Bower, G. H. (1981). Mood and memory. American Psychologist, 36 129-148.
89 Bower, G. H., Monteiro, K. P., & Gilligan, S. G. (1978). Emotional mood as a context of learning and recall. Journal of Verbal Learning and Verbal Behavior, 17 573585. Bradley, B. P., & Mathews, A. (1983). Nega tive self-schemata in clinical depression. British Journal of Clinical Psychology, 22 (Pt 3) 173-181. Bradley, B. P., Mogg, K., & Mill ar, N. (1996). Implicit memory bias in clinical and nonclinical depression. Behaviour Research and Therapy, 34 (11-12), 865-879. Bradley, B. P., Mogg, K., & Williams, R. (1994). Implicit and explicit memory for emotional information in non-clinical subjects. Behaviour Research and Therapy, 32 (1), 65-78. Bradley, B. P., Mogg, K., & Williams, R. (1995). Implicit and explicit memory for emotion-congruent information in clinical depression and anxiety. Behaviour Research and Therapy, 33 (7), 755-770. Bradley, M. M., Cuthbert, B. N., & Lang, P. J. (1993). Pictures as pr epulse: Attention and emotion in startle modification. Psychophysiology, 30 (5), 541-545. Bradley, M. M., & Lang, P. J. (1999). Affective norms for English words (ANEW): Stimuli, instruction man ual and affective ratings. (No. C-1). Gainesville, FL: The Center for Research in Psychophysiology, University of Florida. Burt, D. B., Zembar, M. J., & Niederehe, G. (1995). Depression and memory impairment: A meta-analysis of the associa tion, its pattern, and specificity. Psychological Bulletin, 117 (2), 285-305. Cabeza, R., Kapur, S., Craik, F. I. M., McIntosh, A. R., Houle, S., & Tulving, E. (1997). Functional neuroanatomy of recall a nd recognition: A PET study of episodic memory. The Journal of Cognitive Neuroscience, 9 254-265. Casacalenda, N., Perry, J. C., & Looper, K. (2002). Remission in major depressive disorder: A comparison of pharmaco therapy, psychotherapy, and control conditions. American Journal of Psychiatry, 159 (8), 1354-1360. Coyne, J. C. (1994). Self-reported dist ress: Analog or ersatz depression. Psychological Bulletin, 116 (29-45). Coyne, J. C., Thompson, R., & Racioppo, M. W. (2001). Validity and efficiency of screening for history of depression by self-report. Psychological Assessment, 13 (2), 163-170. Culbertson, F. M. (1997). Depression a nd gender. An international review. American Psychologist, 52 (1), 25-31.
90 Cuthbert, B. N., Lang, P. J., Strauss, C., Drobes, D., Patrick, C. J., & Bradley, M. M. (2003). The psychophysiology of anxiety disorder: fear memory imagery. Psychophysiology, 40 (3), 407-422. Davidson, R. J., Pizzagalli, D., Nitschke, J. B., & Putnam, K. (2002). Depression: Perspectives from affective neuroscience. Annual Review of Psychology, 53 545574. Davis, H. (1979). Self-reference and the enco ding of personal information in depression. Cognitive Therapy and Research, 3 (1), 97-110. Davis, H., & Unruh, W. R. (1981). The de velopment of the self-schema in adult depression. Journal of Abnormal Psychology, 90 (2), 125-133. Davis, M., Falls, W. A., Campeau, S., & Ki m, M. (1993). Fear-pot entiated startle: A neural and pharmacological analysis. Behavioural Brain Research, 58 (1-2), 175198. Deldin, P. J., Keller, J., Gergen, J. A., & Miller, G. A. (2000). Right-posterior face processing anomaly in depression. Journal of Abnormal Psychology, 109 (1), 116121. Derry, P. A., & Kuiper, N. A. (1981). Schematic processing and self-reference in clinical depression. Journal of Abnormal Psychology, 90 (4), 286-297. Deveney, C. M., & Deldin, P. J. (2004). Me mory of faces: a slow wave ERP study of major depression. Emotion, 4 (3), 295-304. Dichter, G. S., Tomarken, A. J., Shelton, R. C., & Sutton, S. K. (2004). Earlyand lateonset startle modulation in unipolar depression. Psychophysiology, 41 (3), 433440. Dobson, K. S. (1989). A meta-analysis of the efficacy of cognitive therapy for depression. Journal of Consulting and Clinical Psychology, 57 (3), 414-419. Doerksen, S., & Shimamura, A. P. (2001). Source memory enhancement for emotional words. Emotion, 1 (1), 5-11. Dozois, D. J. (2002). Cognitive organization of self-schematic content in nondysphoric, mildly dysphoric, and moderately-s everely dysphoric individuals. Cognitive Therapy and Research, 26 (3), 417-429. Dozois, D. J., & Dobson, K. S. (2001). Inform ation processing and cognitive organization in unipolar depression: specificity and comorbidity issues. Journal of Abnormal Psychology, 110 (2), 236-246. Drevets, W. C. (2000). Neuroima ging studies of mood disorders. Biological Psychiatry, 48 (8), 813-829.
91 Drobes, D. J., Hillman, C. H., Bradley, M. M., Cuthbert, B. N., & J., L. P. (1995). Effects of food deprivation on affective star tle modulation and eating behavior. Psychophysiology, 32 S28. Dunbar, G. C., & Lishman, W. A. (1984). Depression, recognition memory and hedonic tone: A signal detection analysis. British Journal of Psychiatry, 144 376-382. Eastman, C. (1976). Behavioral formulations of depression. Psychological Review, 83 (4), 277-291. Ehrlichman, H., Kuhl, S. B., Zhu, J., & Wa rrenburg, S. (1997). Startle reflex modulation by pleasant and unpleasant odors in a between-subjects design. Psychophysiology, 34 (6), 726-729. Fiedler, K., Nickel, S., Mueh lfriedel, T., & Unkelbach, C. (2001). Is mood congruency and effect of genuine memory or response bias? Journal of Experimental Social Psychology, 37 201-214. First, M. B., Spitzer, R. L., Gibbon, M., & Williams, J. B. W. (1996). Structured clinical interview for axis I DSM-IV disorders research version psychotic screen New York: New York State Psychiatric Institute, Biometrics Research Department. Flett, G. L., Vredenburg, K., & Krames, L. (1997). The continuity of depression in clinical and nonclinical samples. Psychological Bulletin, 121 395-416. Forbes, E. E., Miller, A., Cohn, J. F., Fox, N. A., & Kovacs, M. ( 2005). Affect-modulated startle in adults with childhood-onset depr ession: relations to bipolar course and number of lifetime depressive episodes. Psychiatry Research, 134 (1), 11-25. Fowles, D. C. (1988). Psychophysiology and ps ychopathology: a motivational approach. Psychophysiology, 25 (4), 373-391. Fridlund, A. J., & Cacioppo, J. T. (1986). Guidelines for human electromyographic research. Psychophysiology, 23 (5), 567-589. Gloaguen, V., Cottraux, J., Cucherat, M., & Bl ackburn, I. M. (1998). A meta-analysis of the effects of cognitive therapy in depressed patients. Journal of Affective Disorders, 49 (1), 59-72. Grillon, C., Ameli, R., Foot, M., & Davis, M. (1993). Fear-potentiated startle: Relationship to the level of state/ trait anxiety in healthy subjects. Biological Psychiatry, 33 (8-9), 566-574. Grillon, C., Warner, V., Hille, J., Merikangas, K. R., Bruder, G. E., Tenke, C. E., et al. (2005). Families at high and low risk fo r depression: A three-generation startle study. Biological Psychiatry, 57 (9), 953-960.
92 Harmer, C. J., Shelley, N. C., Cowen, P. J ., & Goodwin, G. M. (2004). Increased positive versus negative affective perception and memory in healthy volunteers following selective serotonin and norepi nephrine reuptake inhibition. American Journal of Psychiatry, 161 (7), 1256-1263. Hasher, L., & Zacks, R. T. (1984). Automatic processing of fundamental information: The case of frequency of occurrence. American Psychologist, 39 (12), 1372-1388. Hasin, D. S., Goodwin, R. D., Stinson, F. S., & Grant, B. F. (2005). Epidemiology of major depressive disorder: results fro m the National Epidemiologic Survey on Alcoholism and Related Conditions. Archives of General Psychiatry, 62 (10), 1097-1106. Hawk, L. W., & Cook, E. W., 3rd. (1997). A ffective modulation of tactile startle. Psychophysiology, 34 (1), 23-31. Ingram, R. E., Fidaleo, R. A., Friedberg, R., Shenk, J. L., & Bernet, C. Z. (1995). Content and mode of information processi ng in major depressive disorder. Cognitive Therapy and Research, 19 (3), 281-293. Katon, W., & Sullivan, M. D. (1990). Depr ession and chronic medical illness. Journal of Clinical Psychiatry, 51 Suppl 3-11; discussion 12-14. Kaviani, H., Gray, J. A., Checkley, S. A., Raven, P. W., Wilson, G. D., & Kumari, V. (2004). Affective modulation of the startle response in depression: influence of the severity of depressi on, anhedonia, and anxiety. Journal of Affective Disorders, 83 (1), 21-31. Keller, M. B., McCullough, J. P., Klein, D. N., Arnow, B., Dunner, D. L., Gelenberg, A. J., et al. (2000). A comparison of nef azodone, the cognitive behavioral-analysis system of psychotherapy, and their comb ination for the treatment of chronic depression. New England Journal of Medicine, 342 (20), 1462-1470. Kensinger, E. A., & Corkin, S. (2003). Me mory enhancement for emotional words: Are emotional words more vividly remembered than neutral words? Memory and Cognition, 31 (8), 1169-1180. Knost, B., Flor, H., Braun, C., & Birbaumer, N. (1997). Cerebral processing of words and the development of chronic pain. Psychophysiology, 34 (4), 474-481. Koch, M., Schmid, A., & Schnitzler, H. U. (1996). Pleasure-attenua tion of startle is disrupted by lesions of the nucleus accumbens. Neuroreport, 7 (8), 1442-1446. Koch, M., & Schnitzler, H. U. (1997). The acoustic startle respons e in rats--circuits mediating evocation, inhi bition and potentiation. Behavior and Brain Research, 89 (1-2), 35-49.
93 Koriat, A., & Goldsmith, M. (1996). Monitoring and control processe s in the strategic regulation of memory accuracy. Psychological Review, 103 (490-517). Kucera, H., & Francis, W. N. (1967). Computational analysis of present-day American English Providence, RH: Brown University Press. Kuiper, N. A., & Derry, P. A. (1982). Depr essed and nondepressed content self-reference in mild depressives. Journal of Personality, 50 (1), 67-80. Lang, P. J., Bradley, M. M., & Cuthbert, B. N. (1990). Emotion, attention, and the startle reflex. Psychological Review, 97 (3), 377-395. Lang, P. J., Bradley, M. M., & Cuthbert, B. N. (1992). A motivat ional analysis of emotion: Reflex-cortex connections. Psychological Science, 3 44-49. Larsen, D. K., Norton, G. R., Walker, J. R., & Stein, M. B. (2002). Analysis of startle responses in patients with pani c disorder and social phobia. Cognitive Behaviour Therapy, 31 (4), 156-169. Lyubomirsky, S., & Nolen-Hoeksema, S. (1995). Effects of self-focused rumination on negative thinking and interp ersonal problem solving. Journal of Personality and Social Psychology, 69 (1), 176-190. McNair, D., Lorr, M., & Droppelman, L. (1971). Profile of mood states San Diego, CA: Educational and Industrial Testing Service. McNicol, D. (1972). A primer of signal detection theory London: Allen and Unwin. Miltner, W., Matjak, M., Braun, C., Diekmann, H., & Brody, S. (1994). Emotional qualities of odors and their influence on the startle reflex in humans. Psychophysiology, 31 (1), 107-110. Mineka, S., & Sutton, S. K. (1992). Cognitiv e biases and the emotional disorders. Psychological Science, 3 (1), 65-69. Mogg, K., Bradley, B. P., & Williams, R. ( 1995). Attentional bias in anxiety and depression: The role of awareness. British Journal of Clinical Psychology, 34 ( Pt 1) 17-36. Nolen-Hoeksema, S. (1990). Sex differences in depression Stanford, CA: Stanford University Press. Nolen-Hoeksema, S. (1991). Responses to de pression and their eff ects on the duration of depressive episodes. Journal of Abnormal Psychology, 100 (4), 569-582. Nolen-Hoeksema, S., Larson, J., & Grayson, C. (1999). Explaining the gender difference in depressive symptoms. Journal of Personality and Social Psychology, 77 (5), 1061-1072.
94 Nolen-Hoeksema, S., & Morrow, J. (1991). A prospective study of depression and posttraumatic stress symptoms after a na tural disaster: The 1989 Loma Prieta Earthquake. Journal of Personality and Social Psychology, 61 (1), 115-121. Nolen-Hoeksema, S., Morrow, J., & Fredricks on, B. L. (1993). Response styles and the duration of episodes of depressed mood. Journal of Abnormal Psychology, 102 (1), 20-28. Nolen-Hoeksema, S., Parker, L. E., & Larson, J. (1994). Ruminative coping with depressed mood following loss. Journal of Personality and Social Psychology, 67 (1), 92-104. Panayiotou, G., & Vrana, S. R. (1998). Effect of self-focused a ttention on the startle reflex, heart rate, and memory perf ormance among socially anxious and nonanxious individuals. Psychophysiology, 35 (3), 328-336. Patrick, C. J., Berthot, B. D., & Moore, J. D. (1996). Diazepam blocks fear-potentiated startle in humans. Journal of Abnormal Psychology, 105 (1), 89-96. Patten, S. B. (2003). Recall bias and major depression lifetime prevalence. Society of Psychiatry and Psychiatric Epidemiology, 38 (6), 290-296. Pyszczynski, T., & Greenberg, J. (1987). Self-re gulatory perseverati on and the depressive self-focusing style: A self-awarene ss theory of re active depression. Psychological Bulletin, 102 (1), 122-138. Ratcliff, R. (1993). Methods for dealing with reaction time outliers. Psychological Bulletin, 114 (3), 510-532. Rogers, T. B., Kuiper, N. A., & Kirker, W. S. (1977). Self-reference and the encoding of personal information. Journal of Personality and Social Psychology, 35 (9), 677688. Ross, M. J., & Mueller, J. H. (1986). Depre ssion and trait distinct iveness in the selfschema. Journal of Social and Clinical Psychology, 4 (1), 46-59. Rude, S. S., Gortner, E., & Pennebaker, J. W. (2004). Language use of depressed and depression-vulnerable college students. Cognition & Emotion, 18 (8), 1121-1133. Schneider, W., Eschman, A., & Zuccolotto, A. (2002). E-Prime user's guide Pittsburgh, PA: Psychology Software Tools Inc. Schwarz, N., & Clore, G. L. (2003). Mood as information: 20 years later. Psychological Inquiry, 14 296-303. Scott, K. M., Mogg, K., & Bradley, B. P. ( 2001). Masked semantic priming of emotional information in subclinical depression. Cognitive Therapy and Research, 25 (5), 505-524.
95 Scott, W. D., & Cervone, D. (2002). The impact of negative affect on performance standards: Evidence for an affect-as-information mechanism. Cognitive Therapy and Research, 26 (1), 19-37. Segal, Z. V., & Gemar, M. (1997). Changes in cognitive organizati on for negative selfreferent material following cognitive beha vioral therapy for depression: A primed Stroop study. Cognition and Emotion, 11 501-516. Seljos, K. A., Dawson, M. E., & Schell, A. M. (1995). Availability of informationprocessing resources affects startle eyeblink and pre pulse inhibition. Psychophysiology, 32 S68. Shestyuk, A. Y., Deldin, P. J., Brand, J. E ., & Deveney, C. M. (2005). Reduced sustained brain activity during processing of positive emotional stimuli in major depression. Biological Psychiatry, 57 (10), 1089-1096. Siegel, S., & Castellan, N. J., Jr. (1988). Nonparametric statistics for the behavioral sciences (2nd ed.). New York: McGraw-Hill. Siegle, G. J., Steinhauer, S. R., Carter, C. S., Ramel, W., & Thase, M. E. (2003). Do the seconds turn into hours? Relationship be tween sustained pupil dilation in response to emotional information and self-reported rumination. Cognitive Therapy and Research, 27 (3), 365-382. Siegle, G. J., Steinhauer, S. R., Thase, M. E ., Stenger, V. A., & Carter, C. S. (2002). Can't shake that feeling: event-related fMRI assessment of sustained amygdala activity in response to emotional informa tion in depressed individuals. Biological Psychiatry, 51 (9), 693-707. Sloan, D. M. (2005). It's all about me: Se lf-focused attenti on and depressed mood. Cognitive Therapy and Research, 29 (3), 279-288. Sloan, D. M., Bradley, M. M., Dimoulas, E ., & Lang, P. J. (2002). Looking at facial expressions: Dysphoria and facial EMG. Biological Psychology, 60 (2-3), 79-90. Sloan, D. M., Strauss, M. E., Quirk, S. W., & Sajatovic, M. (1997). Subjective and expressive emotional responses in depression. Journal of Affec tive Disorders, 46 (2), 135-141. Sloan, D. M., Strauss, M. E., & Wisner, K. L. (2001). Diminished response to pleasant stimuli by depressed women. Journal of Abnormal Psychology, 110 (3), 488-493. Sweeney, P. D., Anderson, K., & Bailey, S. (1986 ). Attributional styl e in depression: A meta-analytic review. Journal of Personality and Social Psychology, 50 (5), 974991. Symons, C. S., & Johnson, B. T. (1997). The self-reference effect in memory: A metaanalysis. Psychology Bulletin, 121 (3), 371-394.
96 Talmi, D., & Moscovitch, M. (2004). Can sema ntic relatedness explain the enhancement of memory for emotional words? Memory and Cognition, 32 (5), 742-751. Teasdale, J. D. (1983). Negative thinking in depression: Cause, e ffect, or reciprocal relationship? Advances in Behaviour Research and Therapy, 5 3-25. Teasdale, J. D., & Barnard, P. J. (1993). Affect, cognition, and change Hove, UK: Lawrence Erlbaum Associates. Teasdale, J. D., & Fogarty, S. J. (1979). Diffe rential effects of induced mood on retrieval of pleasant and unpleasant events from episodic memory. Journal of Abnormal Psychology, 88 (3), 248-257. Thomson, D. M., & Tulving, E. (1970). Asso ciative encoding and retrieval: Weak and strong cues. Journal of Experime ntal Psychology, 86 (2), 255-262. Tomarkenand, A. J., & Keener, A. D. (1998). Frontal brain asymmetr y and depression: A self-regulatory perspective. Cognition and Emotion, 12 (3), 387-420. Treynor, W., Gonzalez, R., & Nolen-Hoekse ma, S. (2003). Rumination reconsidered: A psychometric analysis. Cognitive Therapy and Research, 27 (3), 247-259. Tulving, E. (1985). Memory and consciousness. Canadian Psychology, 26 (1), 1-12. Tulving, E., & Thomson, D. M. (1973). Encodi ng specificity and retrieval processes in episodic memory. Psychological Review, 80 (5), 352-373. Vrana, S. R., & Lang, P. J. (1990). Fear imagery and the startle-probe reflex. Journal of Abnormal Psychology, 99 (2), 189-197. Vredenburg, K., Flett, G. L., & Krames, L. (1993). Analog versus c linical depression: A clinical reappraisal. Psychological Bulletin, 113 327-344. Wagner, A. R. (1963). Conditioned frustration as a learned drive. Journal of Experimental Psychology, 66 (2), 142-148. Watkins, E., & Baracaia, S. (2001). Why do people ruminate in dysphoric moods? Personality and Individual Differences, 30 723-734. Watkins, E., & Mason, A. (2002). Mood as input and rumination. Personality and Individual Differences, 32 577-587. Watkins, E., & Teasdale, J. D. (2004). Adaptive and maladapt ive self-focus in depression. Journal of Affective Disorders, 82 (1), 1-8. Watkins, P. C., Mathews, A., Williamson, D. A., & Fuller, R. D. (1992). Mood-congruent memory in depression: Emotional priming or elaboration? Journal of Abnormal Psychology, 101 (3), 581-586.
97 Watkins, P. C., Vache, K., Verney, S. P., Muller, S., & Mathews, A. (1996). Unconscious mood-congruent memory bias in depression. Journal of Abnormal Psychology, 105 (1), 34-41. Wells, K. B., Stewart, A., Hays, R. D., Burn am, M. A., Rogers, W., Daniels, M., et al. (1989). The functioning and well-being of depressed patients. Results from the Medical Outcomes Study. JAMA, 262 (7), 914-919. Wenzlaff, R. M., & Luxton, D. D. (2003). The role of thought suppression in depressive rumination. Cognitive Therapy and Research, 27 (3), 293-308. Wheeler, M. A., Stuss, D. T., & Tulving, E. (1995). Frontal lobe damage produces episodic memory impairment. Journal of the International Neuropsychological Society, 1 (6), 525-536. Whittlesea, B. W. A. (1997). Production, ev aluation, and preservation of experiences: Constructive processing in remembering a nd performance tasks. In D. L. Medin (Ed.), The Psychology of learning and motivation (Vol. 37, pp. 211-264). San Diego, CA: Academic Press. Williams, J. M. G., Watts, F. N., MacLeod, C., & Mathews, A. (1988). Cognitive psychology and emotional disorders Chichester, UK: Wiley. Zuroff, D. C., Colussy, S. A., & Wielgus, M. S. (1983). Selective memory and depression: A cautionary note concerning response bias. Cognitive Therapy and Research, 7 (3), 223-232.
98 BIOGRAPHICAL SKETCH I originally studied mathematics and obt ained a Ph.D. in probability from the University Paris VI, France, in 1999. I had ha d a strong interest in clinical psychology since high school, but felt dissatisfied with the dominant psychoanalytical approach adopted in France. Teaching further sensiti zed me to the importance of psychological factors for human performance and well-bei ng and, when I moved to the US in 2000, I decided to retrain as a neurops ychologist. I entered the University of Florida Department of Clinical and Health Psychology in 2001 and conducted my mastersÂ’ thesis on cognitive control in brain injury in 2003 w ith Dr. William M. Perlstein. I have worked on my dissertation on information-processing bi ases in depression with Dr. Bauer since 2003.