Depression and Implicit Memory: Understanding Mood Congruent Memory Bias

Transcription

1 Cognitive Therapy and Research, Vol. 28, No. 3, June 2004 ( C 2004), pp Depression and Implicit Memory: Understanding Mood Congruent Memory Bias Elaine S. Barry, 1,2,3 Mary J. Naus, 1 and Lynn P. Rehm 1 The present paper reviews the depression and implicit memory literature, emphasizing studies addressing possible mood congruent implicit memory biases in depression. Although some of these studies seem to indicate the presence of mood congruent biases in implicit memory, others fail to show this effect. Although the studies differ on a variety of dimensions (participant population, sample size, implicit memory task, depressive status, etc.), a thorough review of the literature suggests that these are not the most important considerations in understanding the presence or absence of mood congruent memory biases in depression. Rather, the cognitive framework of Transfer Appropriate Processing is used as a tool to organize and explain these findings. In particular, the role of perceptual and conceptual cognitive processes by depressed participants performing implicit memory tasks are examined in the context of perceptual and conceptual task demands. Examining unconscious influences on emotion could have important implications for understanding and treating depression. KEY WORDS: depression; implicit memory; mood congruent bias. In the past two decades, clinicians attempting to understand and treat depression have increasingly turned to research and theories from cognitive psychology, and especially memory, to advance their efforts (e.g., Eysenck, 1991; Ingram, Miranda, & Segal, 1998; Rehm & Naus, 1990; Teasdale, Lloyd, & Hutton, 1998; Williams, Watts, MacLeod, & Mathews, 1997). Similarly, cognitive psychologists are increasingly using depressed individuals in research studies as an empirically based method to better understand the workings of a normative cognitive system (e.g., Bradley, Mogg, & Millar, 1996; Bradley, Mogg, & Williams, 1994; Rathus, Reber, Manza, & Kushner, 1994). Indeed, cognitive researchers such as Colin MacLeod and Andrew Mathews 1 Department of Psychology, University of Houston, Texas. 2 Present address: The Commonwealth College of The Pennsylvania State University, Division of Health and Human Development, Uniontown, Pennsylvania. 3 Correspondence should be directed to Elaine S. Barry, Human Development and Family Studies, Penn State Fayette, One University Drive, Rt. 119 N., P. O. Box 519, Uniontown, Pennsylvania 15401; esb12@psu.edu /04/ /0 C 2004 Plenum Publishing Corporation

2 388 Barry, Naus, and Rehm have strongly encouraged the union of laboratory work on cognition with clinical research on psychopathology (e.g., MacLeod & Mathews, 1991). Within traditional laboratory work on cognition, some cognitive psychologists chose to focus on the distinction between implicit memory and explicit memory (e.g., Bowers & Schacter, 1990; Graf & Schacter, 1985; MacLeod, 1989; Richardson- Klavehn & Bjork, 1988; Roediger, 1990; Schacter, 1987), even though historically, most cognitive memory research concentrated on explicit, declarative memory. It was in the context of explicit memory research that several cognitive models of emotion were developed during the 1990s. Although a comprehensive review of all of these models is beyond the scope of this paper, excellent reviews can be found in Eysenck (1991) and Ingram et al. (1998). However, for illustrative purposes, we will briefly describe several different cognitive models of emotion based on memory (e.g., Beck, 1987; Rehm & Naus, 1990; Riskind, 1989; Williams et al., 1997; Williams, Watts, MacLeod & Mathews, 1988). The Bower (1981, 1987) model would also be included in this category, and it is reviewed in the section below called Priming. Beck s cognitive model of depression (Beck, 1987) proposed that depression is based in depressive schemas that result in distorted cognitions that result in automatic (negative) thoughts. According to his model, individuals prone to developing depression have depressogenic schemas that remain latent until activated by stressful life events. After activation, these negative schemas provide the individual with access to negative cognitive processing that results in depression. The Rehm and Naus (1990) memory model of emotion attempted to incorporate then current models of memory into a comprehensive theory of depression. Central to this model is the idea that the information processing of the depressed individual is emotionally biased. In addition, the emotional quality of experience is an important feature that is stored in semantic memory schemas and also influences how new and old information is interpreted. Riskind s model (Riskind, 1989) was based on the centrality of cognitive priming (discussed in the next section of this paper) and redefining mood as a cognitive state rather than a subjective feeling. Thus, cognitive phenomena are the nucleus of the depressed mood (Riskind, 1989, p. 182). In this model, self-referent effects ( interactive encoding ) take a special place, as negative schemas are primed by negative views of the self and the environment. In the literature review provided in the current paper, self-referent effects are noted when they appear in the research. Finally, Williams et al. (1988, 1997) present a model of depression and anxiety which takes a processing approach. In their model, initial priming and subsequent elaboration are central considerations. Additionally, Williams et al. attempted to specify why processing biases may differ for different emotional disorders. Further, their model assumes that emotions arise to serve biological and social functions that need to be addressed in treating emotional disorders. At the time these models were proposed, studies of the relationship between depressed emotion and implicit memory were not available for consideration. In the intervening decade, however, research concerning depression and implicit memory has proliferated, and the growing numbers of studies with empirical data now demand that a comprehensive model of memory and depression incorporate implicit memory findings. The purpose of this paper is to review the literature on depression and implicit memory to lead to a better understanding of mood congruent memory bias.

3 Depression, Implicit Memory, and Mood Congruency 389 IMPLICIT MEMORY In their landmark paper, Graf and Schacter (1985) defined the terms implicit and explicit memory in the following way: Implicit memory is revealed when performance on a task is facilitated in the absence of conscious recollection; explicit memory is revealed when performance on a task requires conscious recollection of previous experiences (p. 501). For example, say you see a squirrel run across the road on your drive into work. Later, someone asks you to name the first animal you can think of. Without consciously recalling the squirrel from earlier in the day, you quickly answer squirrel. In this case, your memory of the squirrel would appear to have been produced without your awareness. This is implicit memory at work. In fact, you may not even recall the squirrel that ran across the road if someone asked what animals you had seen during the day. That type of direct question, with its subsequent conscious search of memory, is an example of explicit memory. Thus, explicit memory refers to conscious memory, whereas implicit memory usually refers to memory without awareness. What is unclear, however, is whether explicit and implicit memory refer to the memory task (a specific methodology) or to the memory process (a mental event). Dunn and Kirsner (1989) and Richardson-Klavehn and Bjork (1988) have noted that these terms unfortunately have been used in the cognitive literature to refer both memory tasks and memory processes. Transparency Assumption Traditionally, the task used to measure memory defines the type of memory that is studied. Since Graf and Schacter s classic (1985) paper, tasks that can be performed without conscious awareness of the study episode (such as word stem completion or word fragment completion, described in more detail below) have been used to examine implicit memory. Similarly, traditional episodic tasks requiring conscious reference to the study episode (such as free recall or recognition) have been used to examine explicit memory. Thus, the term Transparency Assumption (TA) was coined by Dunn and Kirsner (1989) to describe the notion that a given memory task (i.e., an implicit task like word stem completion) reflects an identical underlying process (i.e., implicit memory). In their words, the TA holds that tasks function as transparent windows to underlying mental processes (Dunn & Kirsner, 1989, p. 18). This assumption, however, could logically only be true under certain special circumstances. Unless all of these circumstances are met simultaneously (which is highly unlikely), there is no reason to assume that the task mirrors the process. For example, a patient with amnesia who has impaired episodic memory may demonstrate implicit memory when completing a series of word stems (e.g., Graf & Schacter, 1985). In this case, the patient has no conscious recollection of the study episode, and the task is designed in such a way that performance on the task can be enhanced by prior presentation of a list of words. Thus, word stems may be completed with more previously presented words by amnesic patients compared to controls (who have had no study episode), and this occurs without conscious awareness. This scenario represents a situation in which the task (word stem completion) would appear to represent the

4 390 Barry, Naus, and Rehm underlying process (implicit memory). However, the case becomes more complicated when nonimpaired participants are involved instead of neuropsychological patients. In the case of nonimpaired participants, performance on a word fragment completion task might be contaminated by explicit recollection (e.g., Roediger, Srivinas, & Weldon, 1989). That is, a participant completing a list of word fragments may become aware that some of the words had been presented in the prior list. Thereafter, he or she could enhance performance on the task by purposely recalling the study episode and using explicit strategies to search for previously encountered words that would complete the fragments. In fact, Graf, Squire, and Mandler (1984) and Squire, Shimamura, and Graf (1987) have both demonstrated that explicit remembering by controls could account for their enhanced performance (compared to amnesic patients) under implicit instructions to remember. Although some empirical studies of implicit memory have addressed this potential confound, the majority of studies seem to ignore it at worst, or pay lip service to it at best. Despite this important and substantive theoretical point regarding task and process, however, in the review portion of this paper the terms implicit memory and explicit memory are used to reflect the author of the studies use of these terms. Priming The premiere method for measuring implicit memory is through priming. Tulving and Schacter (1990) state, [p]riming is a type of implicit memory with affiliations to both procedural and semantic memory (p. 301). Traditionally, the study of priming employs the popular cognitive semantic network model of memory, in which memory is conceptualized to have a structure made up of nodes (e.g., Anderson, 1983; Collins & Loftus, 1975). Nodes may represent concepts, events, ideas, or propositions, and memory happens when spreading activation between nodes in this network causes the information contained within the nodes to become conscious (Bower, 1987). In the case of explicit memory, the memory becomes conscious and the person is usually aware of the result. For example, when trying to remember the name of a movie, one might remember that it has a color in the title. One may then begin searching one s memory for the names of colors, and the name of the movie then pops into consciousness. A successful search of memory, and the person is aware of the result. For implicit memory, it happens a little differently. The priming phenomenon allows researchers to determine the effect that spreading activation has had on items within the memory network, even those of which the person may not be aware. The basic premise is that once nodes are activated, their activation spread to other, nearby nodes. Once these nodes are activated, the concepts may become conscious, although there is often no overt realization (or awareness) on the part of the person that the concept was somehow related to the originally activated node. Consider the squirrel example given earlier. A squirrel is an animal, and so the squirrel node is connected in the memory network with the animal node. Later, after the squirrel encounter, when asked to give the name of an animal, the animal node is activated, and that activation spreads to the squirrel node. Thus, squirrel becomes conscious, even though the person is not aware that this response is due to his or her earlier encounter, which may not even be explicitly remembered.

5 Depression, Implicit Memory, and Mood Congruency 391 Why is squirrel activated after hearing animal, rather than one of thousands of other animals that are also connected to the node in the network? According to the network model, it is because the earlier encounter with the squirrel also activated the squirrel node, and that recent experience left some residual activation, lowering the threshold required for activating that concept again. Everything else being equal, then, when the animal node is activated, some amount of activation spreads to the squirrel node as well as to other, nearby nodes. The activation newly spread to the squirrel node combines with the residual activation from the earlier experience. This combined activation surpasses the threshold necessary to activate squirrel and bring it to consciousness. Priming is usually measured by decreased reaction times to respond to stimuli, increased recognition and/or liking of represented stimuli, increased perceptual ability for degraded stimuli, or other responses that indicate an effect of previous exposure of stimuli on the current task. More specific information regarding the use of priming to measure implicit memory will be discussed in the literature review that follows. Bower and Mood Priming In addition to the traditional cognitive priming described above, mood priming is of particular importance to understanding depression. Bower (1981) formally introduced an associative network theory of memory and emotion, based in part on the network theories of Collins and Loftus (1975) and others, as described above. In addition to concept and proposition nodes, Bower proposed emotion nodes as memory units. He later referred to these as special purpose emotion nodes (Bower, 1987, p. 443). In Bower s model, links to emotions and any related concepts, including internal physiological arousal, episodic information, linguistic concepts, and other semantic links, are considered to have bidirectional influence. In other words, emotion nodes will be activated by experiencing the emotions again or by activation of any of their links. Importantly, once these nodes are activated, the categories that are primed are then used to interpret the somewhat ambiguous social events that go on around us (Bower, 1987, p. 444). For example, say you are dining at a restaurant when you receive bad news. The bad news activates negative emotion nodes, which are then linked to the restaurant. The next time you are at the restaurant, these negative emotion nodes may again be activated, even without your awareness (priming). In this way, an ambiguous situation (ex. slow waitstaff, a mistaken drink order, etc.) may be interpreted negatively ( it s because of what I am wearing or I always get treated badly ). Thus, Bower s theory accounts for both cognitive and emotional phenomenon. The research reviewed in this paper is comprised of empirical mood priming studies, which Ingram, Miranda, and Segal (1998) point out are derived from, and test, models of mood and memory. The models most of these studies are testing were described in the introduction to this paper. DEPRESSION AND IMPLICIT MEMORY As mentioned previously, in the past two dozen years there has been enough research interest in the area of depression and implicit memory to warrant a serious

6 392 Barry, Naus, and Rehm examination of the literature. Unfortunately, the literature on the effects of depression on implicit memory has not enabled interested researchers to reach a consensus regarding these effects. In the past dozen years there have been at least that many research studies investigating depression and implicit memory, falling into two diametrically opposed camps. Seven studies showed no significant differences in implicit memory tasks by depressive status (Bazin, Perruchet, & Feline, 1996; Bazin, Perruchet, DeBonis, & Feline, 1994; Danion, Kauffmann-Muller, Grange, Zimmermann, & Greth, 1995; Denny & Hunt, 1992; Ilsley, Moffoot, & O Carroll, 1995; Lang & Craske, 1997; Watkins, Mathews, Williamson, & Fuller, 1992), whereas eight other studies showed better memory performance in mood-congruent implicit memory tasks by participants classified as depressed (Bradley et al., 1994, 1996, Expts. 1 and 2; Bradley, Mogg, & Williams, 1995; Ruiz-Caballero & Gonzalez, 1994, Expts. 1 and 2; Ruiz-Caballero & Gonzalez, 1997; Watkins, Vache, Verney, Muller, & Mathews, 1996). One study actually found that depressed individuals performed worse than controls on a neutral implicit memory task (Elliott & Greene, 1992), three studies showed no performance deficits by depressed participants (Danion et al., 1991; Hertel & Hardin, 1990; Jenkins & McDowall, 2001), and a review of four early studies suggested that mood-congruent memory (MCM) effects may be found in implicit memory tasks, despite the nonsignificant nature of the reported results (Roediger & McDermott, 1992). These 19 studies differ along several different dimensions, including depressive status of participants (clinical depression, dysthymia, elevated Beck Depression Inventory scores, or induced negative mood), hospitalization and medication status (inpatient/outpatient and medicated/nonmedicated in virtually all possible combinations), type of materials used (emotionally valenced words or neutral words), implicit memory task (word stem completion, word fragment completion, lexical decision, or word associations), levels of processing (primarily perceptual or primarily conceptual), and instructions to remember (intentional or incidental encoding). A critical review of the findings is needed to judge what effect, if any, depression has on implicit memory. Tasks Each of the studies reviewed used one of four implicit memory tasks to measure priming. A summary of these tasks appears in Table I. The first task listed in Table I is the lexical decision task. In this paradigm, words are presented briefly to participants, masked, and then presented again (or a nonword is presented, for no trials). The first presentation of the word (the prime ) is too brief to be reliably detected by the participant, and awareness checks are usually performed to ensure that participants perform at chance levels when attempting to determine whether the prime is a word or a nonword. After the second presentation of the word (the target ), the participant is to respond as quickly as possible as to whether or not the target letter string is a legitimate English word ( yes or no ). When the prime word is the same as the target word (or is a related word), and the reaction time to respond yes to the target word is reduced, priming is said to have occurred. Another type of task appearing in Table I is the word fragment completion task. In this task, the participant is presented with a list of words under either intentional or

7 Depression, Implicit Memory, and Mood Congruency 393 Lexical decision (priming) Table I. Implicit Memory Tasks and Measures of Priming Task Example Measure of priming Word fragment completion Word-stem completion Word associations Present banana very briefly, then present banana, participant is to respond as to whether or not the second presentation forms a word. Complete the following word, ba a Complete the following word, ban Name a fruit and participant responds, banana. Quicker to respond yes to words that were briefly presented previously. More likely to complete with banana after being presented with the word banana in an incidental study phase. More likely to complete with banana after being presented with the word banana in an incidental study phase. More likely to respond banana after being presented with the word banana in an incidental study phase. incidental encoding conditions. Some time later (minutes to weeks), participants are given a list of fragmented words where certain key letters (usually consonants) are missing from the word to be completed (ex. a a in, to be completed into assassin). The participant is to complete each fragment with the first word that comes to mind. No mention is made of the earlier study episode, and usually half of the word stems can be completed with words from the previous list. Priming is measured as the extent to which the number of fragments completed using words from the previously studied list exceeds baseline performance (the number of unstudied fragments completed). The most commonly used task in the depression and implicit memory literature reviewed herein is word stem completion, which is the third task described in Table I. Like the word fragment completion task, in this task, the participant is presented with a list of words under either intentional or incidental encoding conditions. Some time later, and again without reference to the study episode, the participant is given a list of three-letter word stems and asked to complete each stem with the first word that comes to mind. Priming is measured in the same way as for word fragment completion. A final implicit memory task listed in Table I and used in this group of experiments is a conceptual implicit memory task. Like word stem completion and word fragment completion, this task requires participants to be presented with a list of words under either intentional or incidental learning conditions. Then, at some later point, participants are asked to provide word associations to experimenter-provided cues. To the extent that words from the previously presented list appear as products of this association, priming has occurred. Compared to the other types of tasks, however, this task seems to require a different kind of processing on the part of the participant. A theoretical distinction has been made (Roediger et al., 1989) between perceptual implicit memory tasks (like the first three tasks described here), and

8 394 Barry, Naus, and Rehm conceptual implicit memory tasks (like the last one described here). These terms refer to the type of cognitive processing required to complete the task. In short, perceptual implicit memory tasks are more shallow or data-driven tasks, requiring less cognitive effort. Conceptual implicit memory tasks, on the other hand, are more deep or conceptually driven tasks that require at least some cognitive effort. This distinction is further explored in the section entitled Why the Equivocal Results in the Literature?, after the depression and implicit memory research is reviewed. Mood Congruent Memory The research to be reviewed herein primarily centers on the concept of moodcongruent memory. Mood-congruent memory (MCM) occurs when people attend to and learn more about events that match their emotional state (Bower, 1981, p. 147). That is, when the mood of the events or material during the learning episode is matched by the current mood, the result is better memory performance during memory retrieval. For this reason, mood-congruent memory is often confused with mood-state-dependence, which Bower (1981) described as occurring when people recall an event better if they somehow reinstate during recall the original emotion they experienced during learning (p. 147). The distinction between the two concepts is rather fine, and this has implications for research regarding both of these mood and memory effects. The mood-state dependent effect has been found to be difficult to replicate and inconsistent in its appearance (Blaney, 1986; Bower, 1987; Bower & Mayer, 1985; Riskind, 1989; Ucros, 1989; Wetzler, 1985). In his analysis of the literature, Riskind (1989) noted that [i]n general, mood state-dependent memory is not a stable laboratory phenomenon; however, significant state-dependent effects do occur fairly frequently (p. 180). Riskind continued, explaining the conditions under which mood state-dependent effects and mood-congruency effects occur in his cognitive model of emotion described earlier. According to his analysis, mood statedependent effects were more likely to occur with emotional text or story materials and much less likely to occur with list materials, even lists containing positive and negative words. Further, Riskind stated that mood-congruency effects tended to occur under the precise conditions that were often used to create mood state-dependency effects. He explained this using his cognitive-priming formulation, which stated that the cognitive system employs preexisting schemas or categories in learning and retrieving information (Riskind, 1989, p. 178). The use of these schemas places priority on the learning episode, emphasizing interactive encoding (Riskind, 1989, p. 182) and resulting in the pattern of results found in mood and memory research. Mood-congruent memory, compared to mood state dependence, has a strong record of replicability. In fact, mood-congruity effects have even been found in the absence of mood (Perrig & Perrig, 1988)! For the purposes of the current review, it is important to note that the search for any effect of depression on implicit memory centers around Bower s (1981) concept of mood-congruent memory (MCM). In this context, the material to be remembered must have some emotional valence. Further, it should be noted that MCM effects actually require that the pattern of results be reversed for depressed participants and controls, with depressed participants showing enhanced memory for negative words, and controls showing enhanced memory for

9 Depression, Implicit Memory, and Mood Congruency 395 positive words. Three studies have delved into depression and implicit memory since 1990, but without using emotionally valenced stimuli (Danion et al., 1991; Hertel & Hardin, 1990; Jenkins & McDowall, 2001). Because they do not meet the criteria necessary to detect MCM bias, they are not included in this review. With this in mind, then, the literature on the effects of depression on implicit memory is examined. Tables II and III present the data to be reviewed in tabular form, noting the important elements and findings of each study. Table II presents studies finding no effect of depression on implicit memory, whereas Table III presents studies finding an effect of depression on implicit memory. The studies presented in these two tables appear in chronological order for ease of reference. Also, as noted earlier, the literature to be reviewed differs across many different dimensions, including participants, tasks, and methodology, further supporting the appropriateness of a chronological presentation for each table. Finally, because participant populations in these studies differ along various dimensions as noted previously, hereafter participant groups will be referred to as participants classified as depressed, or depressed participants and nondepressed participants or controls. Studies Finding No Effect of Depression on Implicit Memory Early experiments often failed to find an effect of depression on implicit memory (Bazin et al., 1994, 1996; Danion et al., 1995; Denny & Hunt, 1992; Ilsley et al., 1995; Lang & Craske, 1997; Watkins et al., 1992). A brief summary of these experiments appearing in Table II will help to understand the methods and subsequent conclusions of each. First, Denny and Hunt (1992) hypothesized that a dissociation between performance on an explicit and an implicit task would provide evidence for an impairment of effort-demanding processing in depressed participants (the explicit free recall task) compared with an automatic implicit memory task. This study appears in Table II. Clinically depressed individuals were matched with controls and both groups were given implicit and explicit memory tests. Each participant was given a list containing positively and negatively valenced words, which they rated according to the words relevance to themselves (i.e., a self-reference orienting task). Denny and Hunt then had participants perform a word fragment completion task and a free recall task (order of tests was counterbalanced across conditions). Although the data seemed to indicate a trend for enhanced priming for depressed participants negative words (what amounts to a depressive bias), the differences were not statistically significant. Depressed participants did recall significantly fewer words in the free recall task, and the pattern of results indicated an MCM bias. Thus, the authors concluded that there was no differential effect of word valence by participant group (no MCM bias) on the implicit memory task. Table II shows that at the same time the Denny and Hunt (1992) paper was published, Watkins et al. (1992) used clinically depressed participants and nondepressed controls, testing them with both implicit and explicit memory tasks. The authors choice of participants was driven by the same underlying reasoning as in the Denny and Hunt (1990) study described above. Word lists consisted of positively and negatively valenced emotional words. Negative words included depression-related

10 396 Barry, Naus, and Rehm Table II. Chronological History of Research on Depression and Implicit Memory ( ), the Case Against the Effect of Depression on Implicit Memory Study Task Materials Participants Encoding instructions Major finding Denny & Hunt (1992) Word fragments Positive and negative Clinically depressed hospitalized, and control Watkins et al. (1992) Word stems Positive, neutral, and negative (depression-related and physical threat words) Clinically depressed or dysthymia (no medication), and control Bazin et al. (1994) Word stems Positive and negative Clinically depressed hospitalized medicated, and control Danion et al. (1995) Word stems Positive, neutral, and Clinically depressed, no negative medication, and control Ilsley et al. (1995) Word stems Positive and negative Clinically depressed (some medicated, some psychotic), and control Bazin et al. (1996) Word stems Positive and negative Clinically depressed, all medicated, and control High anxious & high Lang and Craske, (1997) Word stems Neutral and negative (social and physical threats) depression, high anxious & low depression, low anxious & low depression Self-reference No differences in priming, no MCM bias Imagine themselves in scene with the word No differences in priming, no MCM bias Read words aloud No differences in priming, but MCM trend Rate pleasantness No differences in priming, no MCM bias Self-relevance No differences in priming, no MCM bias Read aloud No differences in priming (no MCM bias) Sentence frames (does the word fit the sentence?) No differences in priming (no MCM bias)

11 Depression, Implicit Memory, and Mood Congruency 397 Table III. Chronological History of Research on Depression and Implicit Memory ( ), the Case For the Effect of Depression on Implicit Memory: In These Studies an Implicit Memory MCM Bias was Found Study Task Materials Participants Encoding instructions Major finding Bradley et al. (1994) Lexical decision Negative (depression- and anxiety-related), positive, categorized neutral and uncategorized neutral Ruiz-Caballero and Gonzalez (1994) Experiment 1 Ruiz-Caballero and Gonzalez (1994) Experiment 2 Students with high/low BDI, no control group Word stems Positive and negative Students with high/low BDI, and control Word stems Positive, neutral, and negative Bradley et al. (1995) Lexical decision Negative (depression- and anxiety-related), positive, categorized neutral, uncategorized neutral Bradley et al. (1996) Experiment 1 Lexical decision Depression-related, categorized neutral, uncategorized neutral Students with high/low BDI, and control Clinically depressed, and control Students dysphoric per BDI, no control group Supraliminal priming (How often do you use the word?) For incidental recall: How relevant to your concerns? No group effect for surpaliminal primes. High negative affect group faster for subliminal priming of depression words compared to neutral, also for positive words Study list More priming in depressed participants for negative words Intentional: study. No effect of incid/intent, MCM bias in priming Incidental-physical (count vowels) Supraliminal priming (How often do you use word?); Incidental recall (how relevant to your concerns?) Suprathreshold (how often do you use word?) Depressed subjects more priming (both types) of negative words than the other 2 groups No differences in suprathreshold priming, dysphoric more subthreshold priming than controls

12 398 Barry, Naus, and Rehm Table III. Continued Study Task Materials Participants Encoding instructions Major finding Bradley et al. (1996). Experiment 2 Lexical decision Depression-related, categorized neutral and uncategorized neutral Watkins, et al. (1996) Word association Positive, neutral, and negative Ruiz-Caballero and Gonzalez (1997) Clinically depressed with depression or dysthymia, no control group Clinically depressed or dysthymic (students who met criterion), and control Word stems Positive and negative Students with high/low BDI, and control Suprathreshold (how often do you use word?) Imagine themselves in scene, principal character?, vividness, pleasantness Semantic: rate pleasantness; Nonsemantic: count closed letters Depressed more priming of depressed words across both prime types, controls no priming of depression-related words MCM bias in priming MCM bias in priming

13 Depression, Implicit Memory, and Mood Congruency 399 and physical-threat words. Each word was presented for 10-s and participants were asked to form images of themselves interacting with the word. There was a significant MCM bias found in free recall for negative words. Interestingly, this was only true for depression-related words control participants recalled more physical-threat words than depressed participants did. Also, similar to the previously discussed paper, there was a trend in the data toward MCM bias in the word stem completion task for negatively valenced words by depressed participants. Again, however, the difference was not reliable and Watkins et al. (1992) concluded that there was no evidence of depressive MCM bias in implicit memory. Bazin et al. (1994) presented implicit and explicit memory tasks to both clinically depressed (hospitalized) patients and non-depressed controls as noted in Table II. These authors sought to provide evidence that the dissociations between explicit memory and implicit memory in depressed patients as reported in previous studies were not due to differences in the available cues for each type of test. Participants were informed that their memory would be tested, and they were presented with emotionally valenced words which they read aloud. An implicit word stem completion task and an explicit cued recall task followed, using the same stems as cues as in the implicit test. Thus, the materials to test implicit and explicit memory were the same, and only the instructions differed in this study. As in the previous two experiments, a nonsignificant trend toward MCM bias in depressed patients for the implicit memory task appeared. There was no MCM bias in the cued recall task, although controls had higher recall scores. Once again, Bazin et al. (1994) concluded that the performance of depressed patients and controls did not differ on the implicit memory task. Interestingly, the authors also tested participants at the end of their hospitalization, when they had made clinical improvement (p. 243). This time, the dissociation in recall between patients and controls disappeared, supporting the notion that it is the depressive disorder itself that is responsible for the previous differences found in explicit memory, and not any confounds resulting from using a hospitalized sample and a nonclinical control group. In the next study to appear in the literature and in Table II, Danion et al. (1995) tested clinically depressed inpatients versus controls on both implicit and explicit memory tasks. They hypothesized that depressed patients would demonstrate a problem with elaboration (a conscious process that requires effort), a deficit that would appear on an explicit memory task. As in the other studies reviewed in this section, these authors predicted no differences in a more automatic (implicit memory) task. Using emotionally valenced words, they instructed participants to try to remember the words. Participants were given 5-s to read each word aloud and rate its pleasantness on an analogue scale. Memory was then tested by a word completion task, followed by a free recall task and a recognition task. No MCM bias was found on the implicit memory task. Results of the two explicit tasks, however, were quite unusual. Depressed participants recalled significantly more negative than neutral words, but also significantly more positive than neutral words. Control participants, on the other hand, recalled significantly more positive than neutral words, and more negative than neutral words although this latter difference was not significant. Overall recall did not differ between the two groups. Results of the recognition task were similar to the recall task. The authors concluded that the affective valence of words influences

14 400 Barry, Naus, and Rehm memory when conscious intentional recollection is required but is devoid of effect on memory when such recollection is not required (p. 233). As shown in Table II, Ilsley et al. (1995) subjected clinically depressed patients and non-depressed controls to a battery of tests, which included implicit and explicit memory measures. Some of their depressed patients were also psychotic. The hypotheses of this study were that semantic memory (tested by an implicit task) would be largely intact in depression (p. 3), whereas these depressed participants would show an impairment of explicit memory. Without knowing their memory would be tested, participants were presented with emotionally valenced words and asked to evaluate the extent to which each of these words described themselves (i.e., a self-referencing task). Participants then performed a word completion test followed by a cued recall test. As in the previously described research, no significant differences were found between participant groups on the word-completion task. Interestingly, there were also no group differences on the cued recall task, failing to support the authors hypothesis regarding explicit memory. The authors offered no insight as to why this might have been the case, they simply speculated that this [MCM bias] is not as robust and replicable a phenomenon as is widely believed (p. 8). Bazin et al. (1996) examined depressed patients, schizophrenic patients, and nondepressed controls, as noted in Table II. They hypothesized that MCM bias would only appear for depressed patients in the explicit recall task, and would disappear when the patients recovered. Using both implicit and explicit memory tasks, they instructed participants to read the words out loud and to expect a memory test. The words were emotionally valenced common words. In the subsequent word stem completion task, Bazin et al. (1996) found no significant differences in the number or types of word stems completed by the three groups. The authors also failed to find a significant difference in cued recall. In fact, both controls and depressed participants recalled significantly more negative words than positive words, failing to demonstrate the MCM bias the authors had hypothesized. The results were the same for depressed recovered patients and controls. Finally, using physically and socially threatening words as stimuli, Lang and Craske (1997) examined nonclinically anxious and depressed participants on explicit and implicit memory tasks. Participant groups were low depression and low anxiety, low depression and high anxiety, and high depression and high anxiety. Lang and Craske instructed participants to read sentences and determine whether or not the target word made meaningful sense. Afterward, participants completed a free recall task and then a word stem completion task. Statistical analysis did not detect any reliable differences in priming of negative words by depressive status. In other words, no MCM bias was found with the socially threatening for individuals characterized with high depression. An important consideration, however, when evaluating this study is the authors use of the free recall task (explicit) prior to the word stem completion task (implicit). Virtually all of the research on implicit memory tests implicit memory first and then explicit memory to avoid contamination. Thus, in addition to the argument promoted by the current paper, it is not surprising that no MCM bias was found in the Lang and Craske (1997) study. Despite this serious design flaw, the study was included in the current review because of its MCM elements.

15 Depression, Implicit Memory, and Mood Congruency 401 Without exception, at least for the earlier studies reviewed above and appearing in Table II, depression had no consistent effect on implicit memory. Each of these studies used a perceptual implicit memory task (word stem completion or word fragment completion). However, as will be seen in the next section, the choice of task alone cannot account for the failure to find significant differences in implicit memory by participants classified as depressed and controls. Studies Finding an Effect of Depression on Implicit Memory More recent research, however, has rather consistently found that depression does affect implicit memory (Bradley et al., 1994, 1995, 1996, Expts. 1 and 2: Ruiz-Caballero & Gonzalez, 1994, Expts. 1 and 2; Ruiz-Caballero & Gonzalez, 1997; Watkins et al., 1996). As in the previous section, a brief summary of these experiments appearing in Table III will help to understand the methods and subsequent conclusions of each. For example, as can be seen in Table III, Bradley et al. (1994) investigated both implicit and explicit memory for emotional and neutral words using a nonclinical research sample. To test two different theories with opposite predictions, they presented participants with both supraliminal (the participant was aware of its presentation) and subliminal (presented too briefly for conscious awareness) primes. They then used a lexical decision task to determine the amount of priming produced under these conditions. Interestingly, when compared to a low negative affect group, participants who scored high on measures of negative affect (the State Trait Anxiety Inventory and the Beck Depression Inventory) showed increased priming for subliminally presented depression-relevant words compared to neutral words. In contrast, there was no enhanced supraliminal priming for high negative affect participants compared to low negative affect participants. Using the automatic/nonautomatic distinction of Mandler (e.g., 1989), Bradley et al. interpreted the depression-congruent priming effect as evidence that subliminal priming is due to a depressive bias in automatic, integrative processes of high negative affect individuals. On the other hand, supraliminal priming allows for strategic processing, which can be used by non-clinical high negative affect participants to compensate for their depressive bias. The authors of this study went so far as to say that the study provides consistent significant evidence of a depression-congruent bias in automatic memory processes, as reflected by subliminal priming effects (pp ). In the same year, and also appearing in Table III, Ruiz-Caballero and Gonzalez (1994) also used depressed versus nondepressed student volunteers and separated the two groups according to their scores on the Beck Depression Inventory (BDI). This particular sample was used to determine if any MCM bias would appear in subclinically depressed Ss (p. 557). Under intentional encoding instructions for Experiment 1, participants were presented with positively and negatively valenced words. Afterward, they performed a word stem completion task, constituting the implicit memory test, and a free recall test of explicit memory. Results of the study indicated that there was an MCM effect for explicit memory, and that the depressed participants also showed more priming of negative words. To ensure that participants were not using explicit strategies to help them complete the word

16 402 Barry, Naus, and Rehm stems, Ruiz-Caballero and Gonzalez performed a second experiment. To eliminate the possible influence of explicit strategies being used during an implicit memory task, Ruiz-Caballero and Gonzalez provided half of their participants with intentional encoding conditions, and half of their participants with incidental-physical encoding conditions. For the former condition, participants were told that their memory for the words would be tested. For the latter condition, participants were instructed to count the number of vowels in each word as quickly as they could. Results indicated that once again, depressed participants showed a larger priming effect on the word completion task for words of negative valence. Importantly, this effect was not different for intentional versus incidental learning conditions. Contrary to the earlier research reviewed, the authors concluded that if memory bias is assessed from the ease with which a representation of an emotional word can be accessed once it has been primed (implicit memory), the present research also provides evidence that depressed and nondepressed Ss do indeed show a bias favouring emotional information (p. 566). Ruiz-Caballero and Gonzalez interpreted their results in terms of Bower s (1981) network model of mood and memory. In a companion to their 1994 study, Bradley et al. (1995) examined implicit and explicit memory for emotional information, using clinical populations of chronically depressed and clinically anxious participants. This study is also summarized in Table III. Their purpose was twofold: to include anxious participants in addition to depressed participants, and to find differences between anxious and depressed patients on implicit tasks. Using the same type of materials as in their former study (Bradley et al. 1994), and contrary to their hypothesis, they found that depressed participants showed cognitive MCM bias on three types of memory tasks: implicit tasks comprised of both subliminal priming and supraliminal priming, and an explicit memory task (free recall). Bradley et al. (1995) suggested that both automatic and strategic processes are consistently biased in depressed participants toward depression-relevant information. To round out their series of experiments, Bradley et al. (1996) conducted two additional experiments, in which they combined features of each of their studies discussed above. Because of their design differences, these experiments are listed separately in Table III. In Experiment 1, Bradley et al. (1996) hypothesized that dysphoric individuals would show less priming for neutral words than for depression words, in comparison with nondysphoric individuals, especially in the subliminal condition. Their results supported this hypothesis, showing subliminal priming of depression-related words by non-clinical dysphoric participants. In Experiment 2, they hypothesized that clinically depressed individuals would show the same pattern of results as the dysphoric individuals from Experiment 1. In addition, they predicted that the enhanced priming of depression-related words by depressed individuals would also extend to supraliminal priming. Bradley et al. (1996) found enhanced subliminal priming and supraliminal priming of depression-related words in clinically depressed participants. Together, these results replicated their previous findings, and they concluded, [e]vidence of depression-congruent effects in automatic memory processes seems more consistent with the schema and network theories considered earlier... (p. 877). Bradley et al. (1996) pointed out, however, that the important perceptual/conceptual distinction of Tulving and Schacter (1990) needed to be considered in future studies of depression and implicit memory. In other

17 Depression, Implicit Memory, and Mood Congruency 403 words, because certain implicit tasks rely on perceptual priming, whereas others rely on conceptual priming, this distinction needs to be experimentally manipulated with regard to depressed participants. Further, Bradley et al. noted that purely perceptual priming implicit memory tasks may mask possible mood-congruent effects if the perceptual priming effects are masking emotion-congruent priming effects (p. 877). Apparently unaware of the research just reviewed, Watkins et al. (1996) used a conceptual implicit memory test to measure mood-congruent bias in depression. As noted in Table III, they selected clinically depressed individuals and contrasted them with non-depressed controls, hypothesizing that MCM bias would be demonstrated using a word association implicit memory task. Their prediction follows logically from the Transfer Appropriate Processing framework (which is similar to the Bower model and is discussed in the next section), which they pitted against the elaboration hypothesis of Williams et al. (1988). Without citing any of the studies discussed above, Watkins et al. (1996) claimed to be the first authors to find an implicit mood congruent memory (MCM) bias in clinical depression. They first presented participants with positive, neutral, and negative words and instructed them to imagine themselves in a scene that involved the word for 15-s. Watkins et al. (1996) then provided participants with a conceptual implicit memory test. Specifically, they provided participants with a series of association cues to which they were to respond with as many associates to each as they could. This continued for 30-s per cue. In fact, these authors did demonstrate MCM bias by their clinically depressed participants. That is, depressed participants provided more studied negative words in response to association cues than did controls and controls provided more studied positive words in response to association cues than did depressed participants. These authors concluded that their results were consistent with the spreading activation approach of Bower (1981) and the schema theory of Beck (p. 39). Finally appearing in Table III, non-clinically depressed and non-depressed participants were studied by Ruiz-Caballero and Gonzalez (1997). They assessed the effects of semantic versus nonsemantic processing (a conceptual vs. perceptual levelsof-processing distinction) on implicit memory in their participant population, hypothesizing that this manipulation would have no effect on implicit task performance, but would have an effect on explicit task performance. Under the semantic (conceptual) processing condition, and to encourage elaborative processing, participants were informed that their memory would be tested and then were presented with emotionally valenced words that they rated on a Likert scale for pleasantness. Nonsemantic (perceptual) processing instructions did not inform participants that their memory would be tested, and then presented them with emotionally valenced words for which they were to perform a letter counting task. All participants were then given a word stem completion task followed by a free recall test. Results indicated that there was a larger priming effect for mood-congruent words than for words that were not moodcongruent. Ruiz-Caballero and Gonzalez (1997) also found an MCM effect in free recall. The authors suggested that mood-congruent information is more accessible than mood-incongruent information, consistent with Bower s (1981) network theory of emotion. Each of the studies just discussed and presented in Table III used a perceptual implicit memory task (word stem completion and lexical decision), with the exception