For a long time, except in the work of Bartlett, this question was rarely asked inside the psychological laboratory. Psychoanalysis, with its emphasis on repressive defenses against conflict and anxiety, was interested in the problem; but except for the psychoanalytic ego-psychologists, such as David Rapaport and George Klein, psychoanalytic theory had little interaction with experimental psychology.
Things began to change in the late 1940s, when Jerome Bruner and Leo Postman, then working together at Harvard University (Postman later moved to UC Berkeley, where he spent the rest of his career), announced a 'New Look" in perception -- a term inspired by Christian Dior's "New Look" in women's fashion, introduced after the end of World War II. Bruner, himself influenced by psychoanalysis, was interested in how emotional and motivational processes affected perception and attention -- as in the psychoanalytic concept of repression. The New Look in perception never quite caught on, partly because of its attachment to psychoanalysis and partly because psychology just wasn't ready for it, but it was an important precursor to the cognitive revolution of the late 1950s and 1960s. As the cognitive revolution in psychology was followed by a renewed interest in emotion, sometimes amounting to an affective counter-revolution, psychologists took up anew the basic New Look issue question of how emotion affects what we perceive and remember, and the judgments we make about the present and the past (for general reviews, see Blaney, 1986; Bower, 1981; Eich, 1995a; Ellis & Moore, 1997; Fiedler & Forgas, 1988; Forgas, 1991; Christianson, 1992b; Clark & Fiske, 1982; Ekman & Davidson, 1994; Johnson & Magaro, 1987; Kuiken, 1991; Niedenthal & Kitayama, 1994). Much of this literature has focused on memory.
There has also emerged a literature on motivation and memory, but -- in large part, perhaps, because of the enduring attractiveness of the psychoanalytic concept of repression -- the literature on emotion and memory is by far the larger of the two.
But before we consider the relation between emotion and memory, we've got to ask another question: We know what memory is; but what's an emotion?
This turns out not to be an easy question to answer.
As a first pass, we can say that the
domain of affect has two aspects to it:
The fundamental distinction between pleasant and unpleasant feeling states has provided the basis for a number of two-dimensional theories of the structure of emotion. These go by the general label of the affect circumplex, where a circumplex is a circular arrangement of concepts in which the angular distance between them represents the degree of their co-occurrence or similarity.
But, arguably, there is more to our emotional lives than just feeling good or feeling bad. In an extremely popular theory, Ekman and Friesen (1975; Ekman, 1984) distinguished among about a half-dozen "basic" or "discrete" emotions:
These emotions are "basic" because they are innate responses to certain categories of stimuli, and shared an evolutionary commonality with other mammals -- as first proposed by Darwin in his book, The Expression of Emotion in Animals and Men. Note that there is only one positive emotion (surprise can be either positive or negative). For Ekman, the various emotional states are not related to each other along one or another continuous dimension. Rather, there are qualitative differences between them.
And, arguably, there are other emotional states which are elicited only in particular social contexts, and require some conscious awareness of one's relation to other people. Harter (1999) has called these "self-conscious" emotions:
Let us set this (admittedly hugely important) theoretical issue aside, however. So far as memory is concerned, the basic strategies for studying the effects of emotion on memory are clear:
So how do we do this?
First, a number of investigators have gone to the trouble of collecting normative ratings of words on emotional dimensions of positivity and negativity. See, for example, Anderson's (1969) ratings of the social desirability of trait adjectives such as friendly and hostile. And, for other syntactical categories, Toglia and Battig rated a large number of words on a dimension of pleasantness. For nonverbal stimulus materials, the International Affective Picture System is a large set of pictures rated on such dimensions as anger and disgust.
Other investigators have developed a number of techniques
for experimentally manipulating subjects' emotional
state. Thus, Schachter and Singer (1962) induced
physiological arousal by injecting their subjects with
epinephrine. But beyond ostensibly "undifferentiated"
arousal, there are a number of techniques that can induce
particular emotional states.
So far as memory research is concerned, the Velten Technique and musical mood inductions are by far the most popular.
Note: Much of what follows is based on a review article by Kihlstrom, Eich, Sandbrand, and Tobias (2000).
Two emotional effects on memory have to do with the emotional valence, positive or negative, of the material to be remembered and the cues used for retrieval, regardless of the mood state of the person doing the remembering.
The more
familiar of these is the affective valence effect:
in general, information associated with positive affect is
more easily remembered than that associated with negative
affect. The tendency to remember that which is agreeable,
and to forget that which is disagreeable, is sometimes taken
as evidence for the psychoanalytic concept of repression
(for reviews see Singer, 1990), but this is a mistake: as
Rapaport (1942) noted, Freud's notion of repression covers
material which is associated with primitive sexual and
aggressive motives, and the anxiety aroused by their
conflict with social strictures, not the merely unpleasant.
Nevertheless, the general tendency to favor positive over
negative material is so intuitively appealing that it has
acquired its own special label, the "Pollyanna Principle",
after the best-selling 1913 novel by Eleanor H. Porter
(Boucher & Osgood, 1969; Matlin & Stang, 1978).
Matlin and Stang (1978) reviewed a wealth of literature indicating that, on average, subjects favor positive over negative material in attention, perception, language, word associations, learning, memory, thinking, judgment, and social cognition, and they concluded that "This preference for the pleasant seemed to invade every area of experimental psychology that we examined, and the effects were usually remarkably robust" (pp. 2-3). They further proposed, following Erdelyi (1974), that human information processing is necessarily selective, and that the seeking of pleasure, and avoidance of displeasure, are the most important principles on which this selectivity is based.
The affective valence effect is sometimes interpreted in terms of Freudian repression. Just as Freud's patients repressed their traumatic and conflictual thoughts, feelings, and desires, so the argument goes, so people have a tendency to ignore and forget what is disagreeable. However, this argument fails to do justice to Freud. What is repressed, in Freudian psychoanalytic theory, are thoughts and impulses related to primitive sexual and aggressive motives that conflict with social strictures. As Rapaport (1944) argued long ago, what is repressed is not the "merely unpleasant". Like Rapaport, Freudians who understand their Freud require much more by way of evidence for repression.
While focusing
their attention on the Pollyanna Principle, Matlin and Stang
(1978) also acknowledged the operation of another, secondary
principle, the Intensity Principle: that intense or highly
polarized items are processed more efficiently than neutral
or less intense items (Dutta & Kanungo, 1975). We refer
to this principle as the affective intensity effect:
memories associated with some affect, or emotional arousal,
are better remembered than those which are affectively
neutral, regardless of whether the emotional valence is
positive or negative.
In part, the affective intensity effect is a special case of the familiar salience effect on memory, also known as the von Restorff effect, after the researcher who first noted it: more salient events are more memorable than less salient ones (von Restorff, 1933; see also Fabiani & Donchin, 1995; Hunt, 1995; Tversky & Kahneman, 1973). Apparently, possessing some emotional valence is one of the things that can render a memory salient.
Matlin and Stang (1978) attempted to
reconcile the effects of affective valence and affective
intensity by proposing a J-shaped function relating
emotional valence to memorability. Items associated with
both negative and positive affective valence are more
memorable than those with no valence, but items with a
positive valence are more memorable than those with a
negative valence.
Unfortunately, this compromise is threatened by a methodological flaw noted by Banaji & Hardin (1994). Although studies of affective intensity have usually controlled for the conditions under which the memories in question were encoded, studies of affective valence generally have not done so. There is reason to think that positive events may be more arousing than negative ones, on average, and that when intensity is controlled what looks like an affective valence effect may really resolve to an affective intensity effect (Banaji & Hardin, 1994). There may well be an affective valence effect on memory, independent of affective intensity, but further research will be necessary to demonstrate it convincingly.
The affective intensity and affective valence effects are generated by the affective properties of the target items themselves -- that is, whether they carry positive or negative emotional valence. But other effects of emotion on memory have to do with the subject's internal emotional state.
Historically,
perhaps the most familiar consequence of the subject's
affective state is known as the Yerkes-Dodson Law (Yerkes
& Dodson, 1908; for recent reviews, see Anderson, 1990;
Neiss, 1988): arousal is related to task performance by an
inverted-U-shaped function. Stated more clearly, the
Yerkes-Dodson Law states that, for every task, there is some
moderate state of arousal, not too low and not too high,
which leads to optimal performance. Further, this
optimum arousal level differs from task to task, depending
on task demands.
In the memory context, the "tasks" involve either encoding or retrieval operations or both. Although a number of competing explanations have been put forward for the effects of arousal on memory (for reviews, see Christianson, 1992a; Eysenck, 1976; Heuer & Reisberg, 1992; Mandler, 1992; Revelle & Loftus, 1992), in general terms we may hypothesize that low levels of arousal (e.g., fatigue or drowsiness) are associated with low levels of attention, poor encodings of events as they occur, and poor memory for these events later. Increasing arousal also increases the deployment of attention, resulting in better memory, but very high levels of arousal may increase information processing load, and thus effectively reduce the amount of information being processed -- the result, again, is poor encoding, and poor memory, especially of peripheral details.
An interesting
aspect of the affective intensity effect is its interaction
with time. An often-cited experiment by Kleinsmith and
Kaplan (1963, 1964) recorded subjects' levels of
physiological arousal during encoding, and then tested
memory at delays of 2 minutes to 7 days. At short intervals,
items associated with relatively high arousal were poorly
remembered; at long intervals, however, these same items
were well remembered.
The eponymous Kleinsmith and Kaplan effect has proved difficult to replicate, but it shines through in a meta-analysis of replication attempts (Park & Banaji, 1997), and has been confirmed experimentally (Crowder, Banaji, Wenk, Hardin, Phelps, LaBar, & Ziv, 1997). But in order to replicate the K&K effect, Crowder and his colleagues had to do exactly what K&K had done. This illustrates a general rule of scientific replication: if you want to replicate someone else's effect, you've got to replicate their method. And if you can't replicate their effect, it's probably because you haven't replicated their method.
Most previous studies either measured subjects' levels of physiological arousal indirectly, by asking them to complete questionnaire measures of neuroticism (in H. Eysenck's system, associated with high levels of autonomic arousal) or introversion-extraversion (in Eysenck's system, hypothetically associated with cortical arousal. Sometimes they measured subjects level of overall physiological arousal during the experiment by measures such as the galvanic skin response. Sometimes they attempted to manipulate subjects' arousal levels by asking them to study words that had some emotional valence (whether positive or negative) attached to them. Most of these individual attempts at replication failed to produce significant differences -- although, as noted, they did come through in a meta-analysis of published results. But then again, none of these attempted replications actually replicated K&K's methods. They were merely conceptual replications.
Crowder and his colleagues, by contrast, replicated K&K precisely.
- K&K didn't use words, they used nonsense syllables, and Crowder et al. used the very same nonsense syllables as they had done.
- K&K didn't just measure subjects' overall levels of physiological arousal. Instead, they assessed each individual subject's physiological response to each individual nonsense syllable. Which is just what Crowder et al. did.
- K&K then examined the physiological measurements on an individual basis, and constructed for each individual subject lists of nonsense syllables to which that subject had given a weak or strong physiological response. These individualized lists of CVCs were then presented to the subjects at the time of test.
- Just to be clear: K&K's original experiment, like Crowder et al.'s replication, was totally individualized. No shortcuts. Both sets of investigators went to the trouble to measure each individual's idiosyncratic response to each individual item.
And when they did that, they got what K&K got. Just to hammer the point home, the fate of the memories could not be predicted from the affective valence of the items in the abstract, or from the subjects' overall state of physiological arousal. The effect only emerged when the researchers considered the subject's actual affective response to each item. In other words, the subject's affective state is at least as important as the affective valence of what he or she is trying to remember. It may well be that encoding or retrieving any affectively valenced material induces a corresponding emotional state in the subject, but this should not be assumed, and investigators are well-advised to make some assessment of the subject's actual emotional state.
The Kleinsmith and Kaplan (1963, 1964) effect is not an artifact of statistical regression, but rather is a genuine crossover interaction: Memories associated with high levels of arousal are remembered poorly over the short term, but well over the long term. The effect is of great interest in the context of the controversy over "recovered" memories of childhood sexual abuse, although it should be noted that the retention intervals involved in these experiments, minutes and days, pale before the years and decades claimed in cases of recovered memory. In the final analysis, the important issue in the recovered memory debate remains that of providing independent corroboration for any incidents of abuse that a subject or patient might remember.
The effect of
physiological arousal on memory is nicely illustrated by
famous experiments by Cahill, McGaugh, and their colleagues
on both human and animal models of learning and
memory. In one experiment, Cahill et al. (1994) had
subjects read a three-part story, the middle portion of
which was intended to be emotionally arousing (in a control
condition, the middle portion was affectively
neutral). Prior to reading the story, one group of
subjects was injected with a beta-blocker, a drug
that dampens physiological arousal; another group was
injected with a placebo. Subjects who read the
arousing story under conditions of placebo showed
particularly good memory for details of the story --
especially details drawn from the middle, arousing,
part. Subjects who read the arousing story under the
influence of the beta-blocker showed no such
advantage.
Experiments like Cahill's underscore a general principle that emotional arousal tends to enhance memory. This principle would seem to contradict Freud's notion of repression, though it may be that very high levels of arousal impair memory, along the lines of the Yerkes-Dodson Law. In fact, there is a relatively large literature indicating that high levels of arousal do, in fact, impair memory for peripheral details (e.g., Reisberg, 2006; Mather, 2007). However, central details are spared this effect, so that whatever impairment in memory results from high levels of arousal, it is nothing like a general amnesia.
It should be noted that the foregoing phenomena have to do with the effects of undifferentiated arousal on memory. What about particular emotions, such as happiness or sadness, anger or disgust. As it happens, as noted by Levine & Pizarro (2004, 2006), there has been relatively little such research until relatively recently. And much of that has been based on the simple distinction between happiness and sadness.
Shifting from
undifferentiated arousal to valenced mood, the effects on
memory of low arousal levels are paralleled in the resource
allocation effect of depression on ongoing
information-processing tasks, including the encoding and
retrieval of memories (Ellis & Ashbrook, 1988, 1989; see
also Hertel, 1994). In particular, depressed mood appears to
increase the person's information-processing load (Ellis
& Ashbrook, 1988, 1989) -- or, alternatively, his or her
level of initiative (Hertel, 1994). In the former case, the
added internal cognitive activity drains attentional
resources that would otherwise be devoted to other tasks; in
the latter case, available attentional resources simply are
not devoted to the task in the first place. Either situation
results in poor encoding of material presented during a
depressed mood, and performance deficits on subsequent
retention tests -- especially if these tasks themselves are
cognitively demanding.
The resource allocation effects of negative moods are fairly well documented at this point, but we have little information about comparable effects of positive mood states. Comparing the effects of positive and negative moods is of considerable theoretical importance in determining underlying mechanisms (Kihlstrom, 1989). For example, if resource allocation effects are mediated by the distracting effect of the person's mood-relevant thoughts, as Ellis and Ashbrook (1988, 1989) suggest, we should expect happy and sad moods to have the same effects on information processing. On the other hand, it might be that depressed moods have a more specific effect on attentional allocation policy -- for example, by reducing people's interest in their surroundings and consequently their motivation to pay attention to them, as suggested by Hertel (1994). In this case, happiness and sadness should have opposite effects: the sad person's disinterest should impair cognitive processing, while the happy person's peppiness should facilitate it.
Some of the
most dramatic effects of mood state are seen in mood
dependent memory, which occurs when retrieval of
material is enhanced by reinstating the mood that the
individual was in when the material was initially encoded
(Bower, 1981). The notion of mood dependent memory, is based
on an analogy with the state dependent memory produced by
pharmacological substances which act directly on the central
nervous system (Overton, 1984; for reviews see Eich, 1980,
1989); conceptually similar environment dependent memory
effects have been observed in experiments where the encoding
and retrieval phases take place in different physical
environments (e.g., Godden & Baddeley, 1975; Smith,
Glenberg, & Bjork, 1978), at different times of day
(e.g., Holloway, 1978), or while listening to different
kinds of music (e.g., Balch & Lewis, 1996). In all these
cases, the memorability of an event is a function of the
congruence between the context in which the memory is
encoded and that in which retrieval is attempted. These
context effects, in turn, exemplify the more general encoding
specificity principle in memory, which states that
memory is best when the processing operations performed at
the time of retrieval match those that were performed at the
time of encoding (Tulving & Thomson, 1973; see also
Kihlstrom & Barnhardt, 1993).
Some early support for mood-dependent memory came from studies of patients with bipolar affective disorder (Weingartner, Miller, & Murphy, 1977) and normal subjects threatened with electric shock (Macht, Spear, & Levis, 1977). Following those early studies, however, mood dependence assumed a "now you see it, now you don't" quality. Mood dependent memory has rarely been observed in straightforward designs in which subjects study a single word list in one mood, and are tested for memory in the same or a different mood (e.g., Bower, Monteiro, & Gilligan, 1978, Experiments 1-2). Bower et al. (1978, Experiment 3) were able to produce the effect using an interference design employing two word-lists, but subsequent conceptual replications have yielded mixed results; interestingly, environment dependent effects on memory also appeared to be unreliable (Fernandez & Glenberg, 1985). In the face of these failures to replicate, Bower and Mayer (1989) concluded that mood dependence was evanescent rather than robust, and that most of the earlier positive results, including Bower's own, may well have been spurious outcomes -- a point of view apparently shared by some other reviewers of this literature (Blaney, 1986; Leventhal & Tomarken, 1986).
In the most recent turn of events, Eich and his colleagues have developed a paradigm in which mood dependence can be reliably produced in the laboratory (Eich, 1995a). For example, Eich and Metcalfe (1989) employed a musical mood-induction procedure, and asked subjects to study a list of words; later, they went through the musical mood-induction procedure again, and attempted to remember the items they had studied previously. Testing revealed strong evidence of mood dependence: items studied while subjects were sad were remembered better when the subjects were also sad during the memory test, and items studied while they were happy were remembered better when they were happy. The same experiment also illustrated the resource allocation effects of sad mood on encoding and retrieval: compared to the condition where subjects were happy during both encoding and retrieval, memory was relatively poor when subjects were sad while encoding, regardless of whether the retrieval state was happy or sad; and when subjects were sad while retrieving, regardless of whether the encoding state was happy or sad. Other experiments in Eich's laboratory have yielded similar effects for the retrieval of autobiographical memories as opposed to word lists (Eich, Macaulay, & Ryan, 1994). Most recently, Eich, Macaulay, and Lam (1997) have reported good evidence for mood dependent memory in bipolar affective disorder (manic-depressive illness).
Mood dependence can affect memory processes even if subjects are not engaged in tasks that would ordinarily be considered to involve memory. At issue here is the distinction between explicit memory, or conscious recollection, and implicit memory, or the influence of memory on task performance, independent of conscious recollection (Schacter, 1987). Almost all studies of emotion and memory involve explicit memory tasks such as free recall, cued recall, and recognition. Implicit memory tasks have been studied in the literature on resource allocation effects, where it has been found that depression impairs performance on explicit memory tasks, which typically make considerable demands on attentional resources, but not on implicit memory tasks, which are less dependent on attention (e.g., Ellis & Ashbrook, 1988, 1989). Macaulay, Ryan, and Eich (1993) found evidence of mood dependence in performance on a category-generation test of implicit memory.
From one point of view, it might be thought that mood dependence would have little or no effect on implicit memory. After all, as a general rule, implicit memory transfers when explicit memory does not (Roediger & McDermott, 1993). As an example, performance on at least some implicit memory tasks is spared in cases of multiple personality disorder (also known as dissociative identity disorder), even when the inter-personality amnesia profoundly impairs performance on explicit memory tasks (e.g., Eich, Macaulay, Loewenstein, & Dihle, 1997; Nissen, Ross, Willingham, MacKenzie, & Schacter, 1988; for a review, see Kihlstrom & Schacter, 1995). On the other hand, Tobias (1992; Tobias, Kihlstrom, & Schacter, 1992) suggested that mood dependence might be greater in implicit, compared to explicit, memory. It is known that drug-state dependency effects on explicit memory are cue dependent: the effect is found when the memory task employs relatively impoverished retrieval cues, such as free recall, but not with relatively rich cues, such as cued recall or recognition (Eich, 1980, 1989). According to Tobias (1992), implicit memory tasks involve the most impoverished retrieval cues of all -- after all, they don't even specify that the subject should retrieve a memory. Her first study, which compared stem-cued recall to stem completion, found no evidence of mood dependence on either explicit or implicit memory. However, her second study, which compared free recall to a novel test of free association, found a small but significant mood dependent effect on implicit, but not explicit, memory. Although further studies along these lines need to be performed, the important point is that mood dependent memory may affect performance even on tasks that do not seem to involve conscious recollection of the past.
Mood dependent memory is not merely a curiosity of mental life, but may have broad practical and theoretical implications. For example, mood dependent memory may be implicated in the perseveration of clinical episodes of depression in affective disorder patients. That is to say, negative mood may make negative memories more available, thus reinforcing the negative mood, which makes negative memories even more available, and so on in a vicious cycle. More importantly, in the present context, it means that it may be hard for people who are currently depressed to remember times when they were happy, and vice-versa, making cognitive-behavioral therapy more difficult, and distorting their memories of childhood and other possibly critical events.
Concern about
memory distortion is strengthened by another set of
mood-memory effects, which go by the label of mood
congruent memory: mood state facilitates the
processing of material with a similar emotional valence, and
impairs the processing of material with the opposite valence
(for reviews, see Blaney, 1986; Bower & Forgas, 1997;
Johnson & Magaro, 1987; Leventhal & Tomarken, 1986).
Conceptually, mood congruent memory effects can be separated
into those that operate at the encoding stage (mood
congruent encoding) and those that operate at the
retrieval stage (mood congruent retrieval); but the
outcome of these processes is the same: better memory for
information whose affective valence matches the valence of
the person's mood at the time information processing occurs.
Mood congruent encoding appears to be mediated by attentional processes: happy subjects pay more attention to positive items, while sad subjects pay more attention to negative ones. Among patients, mood dependent encoding is more likely to occur in cases of depression than of anxiety disorder, perhaps because anxious individuals are hyper-vigilant, and deploy attention away from threat cues (Mathews & MacLeod, 1994; Mineka, 1992). In the present context, though, mood congruent retrieval is the more important effect, because it raises the possibility that the subject's mood state at the time of inquiry may bias his or her self-reports of past experiences -- happy subjects more likely to retrieve happy memories, and sad subjects more likely to retrieve sad ones. In fact, prima facie evidence for mood congruent retrieval comes from studies of autobiographical memory where both clinical and non-clinical subjects have been found to access personal memories that are affectively congruent with mood state at time of retrieval more readily than incongruent ones (e.g., Lloyd & Lishman, 1975; Teasdale & Fogarty, 1979; Teasdale & Russell, 1983).
Demonstrations of mood congruent memory have sometimes been interpreted as instances of mood dependence, on the assumption that unpleasant events induce unpleasant mood states in those who experience them, and pleasant events induce pleasant mood states. This is an interesting theoretical controversy -- unfortunately, one with little empirical evidence on either side. As a practical matter, though, whether taken together or separately, mood dependence and mood congruence raise the question of whether subjects' self-reports of past experience might be distorted by the mood state they are in at the time they make the report. Thus, for example, depressed or anxious individuals might exaggerate the frequency or severity of trauma, loss, abuse, or other negative events from childhood, because such memories are more accessible to them at the time of retrieval. This is an extremely important empirical question, but it is also a fiendishly difficult one to answer. One reason for the difficulty, of course, is that the best way to evaluate memory bias is to compare the subject's memories to an objective record of what happened in his or her life, and this sort of information is generally not available except in longitudinal databases. As an alternative, one could test memory when people were in a negative mood state, and then again in neutral and positive mood states, but this is also difficult to manage.
To complicate
things further, some investigators have found evidence for mood-incongruent
memory as well (e.g., Perrott & Sabini,
1990). That is to say, subjects may remember negative
events better when they are happy, and positive events
better when they are sad. Note, however, that the
initial demonstration of mood-incongruence has not received
very much attention: it deserves replication and extension
before too much is made of it.
Precisely how to reconcile mood-incongruence (if it occurs) with mood-congruence (which definitely occurs) is not clear, though it may be that the resolution may be similar to the J-shaped function proposed by Matlin and Stang (1978). That is to say, mood-congruence may be stronger than mood-incongruence, though both effects may be bigger than the effect of mood on affectively neutral material.
If so, mood-congruent and mood-incongruent memory (if it occurs) may be analogized to the schematic processing principle.
Recall that, for Bartlett and others, a schema is a kind of cognitive background that forms a framework for remembering. According to Hastie (1980):
By analogy, we may say that the person's emotional state
provides an emotional background that also forms a
framework for remembering (in fact, Bartlett asserted that
reconstruction begins with the subject's general attitude
toward the event being remembered). By analogy, then:
But, of course, this is an explanation in search of a phenomenon. Before we explain why mood-incongruent memory occurs, we should make sure that it really does occur.
In principle, both mood-congruent and mood-dependent retrieval pose a big problem for investigators who rely on self-reports. But is it a problem in actual practice? A recent paper by Brewin and his colleagues (Brewin, Andrews, & Gotlib, 1993) concludes that it is not. Brewin et al. cover three common criticisms of retrospective reports of early experiences: (a) that memories of childhood are imperfect and unreliable; (b) that syndromes such as anxiety and depression generally impair memory function; and (c) that depression biases retrieval away from positive material and toward negative material. They find the evidence for the first two criticisms wanting. They conclude that, in general, our recollection of the past is "reasonably free of error" (quoting Baddeley, 1990, p. 310), and that anxiety and depression confer no special impairment on memory functioning. Most importantly, Brewin et al. find the evidence that negative mood biases memory toward negative events inconsistent and unconvincing; and they conclude that although mood may distort memory for relatively recent events, it has little impact on memory for childhood. This leads them to conclude that retrospective reports of childhood experiences can be taken at face value, although they do propose some strategies for enhancing accurate recall and minimizing bias and error.
On the other hand, Brewin et al. (1993) may have been premature in their conclusions about the accuracy of childhood memory and its invulnerability to emotional distortion. Remembering is inherently a reconstructive process (Bartlett, 1932; Ross, 1989), with lots of opportunity for bias and error (Roediger, 1996), and one's emotional state must cast a light -- or a shadow -- over that process. Moreover, anxiety and depression very likely put a considerable drain on cognitive resources -- even if, like sleep-deprived subjects, the anxious and the depressed can pull themselves together for a short while and perform adequately on standard laboratory tests.
Most importantly, the evidence on which Brewin et al. (1993) base their sanguine conclusions about mood congruent memory seems inadequate. Very few experimental studies involve clinically significant alterations in mood, or for that matter clinically significant memories; very few have independent confirmation of the quality that is really necessary to evaluate memories for distortion and bias. Moreover, surprisingly little of the evidence pertaining to childhood memory comes from the recall and interpretations of discrete events. In fact, a great deal of it involves test-retest correlations on scales rating such things as parental discord or care; another large segment involves factual knowledge, such as where the subject lived or who took care of the subject during the birth of a sibling. This is not within the definition of episodic memory.
In this context, a strong cautionary note is offered by a recent study of memory in soldiers who participated in the Persian Gulf War (Southwick, Morgan, Nicolaou, & Charney, 1997). One month after their return from the theatre of operations, the soldiers completed a questionnaire concerned with combat-related traumatic events (e.g., an extreme threat to personal safety, observing bizarre disfigurement from wounds, seeing others killed or wounded). Two years later they completed the questionnaire again, and this time they reported significantly more traumatic experiences than they had initially. Moreover, scores on a scale of combat-related post-traumatic stress disorder (PTSD) were significantly correlated with these changes in the follow-up reports. One interpretation of these findings is that individuals who show symptoms of PTSD exaggerate their histories of traumatic events, possibly as part of an attempt to explain their current problems. The same danger exists in other studies of memory in individuals with anxiety, depression, and PTSD.
One area in which we agree with Brewin et al. (1993) is in their call for more research in this area. Beginning with classical psychoanalysis, a variety of theories implicate childhood experiences as the cause of adult psychopathology -- a viewpoint that has been revived in recent speculations about the role of incest and child sexual abuse in eating disorders and other mental illnesses. It is a fact that many mental patients frequently report a high level of childhood trauma, abuse, and neglect, but what we need to know is whether these reports are reliable, and whether there is any causal relation between the trauma and the illness. We do not know this for sure. Everything we know about mood and memory justifies our suspicion that memory-based self-reports may not be reliable, but this knowledge is also limited. Therefore, clinical practitioners and researchers should treat retrospective reports of childhood and adult experiences as just that -- reports to be verified, not historical truth. If we want to know the truth about such things as the association between childhood experience and adult depression, the approach is not through memory, with all its vagaries, but through history: not through retrospective self reports, but through prospective studies of objective data.
Certainly the most dramatic claims about emotion and memory harken back to the Freudian concept of repression: that psychological trauma can induce a defensive process of repression that renders the victim amnesic for the trauma itself. Some versions of this trauma-memory argument invoke a concept of dissociation rather than repression, but the essential argument is the same:
A good example of research ostensibly supporting the
trauma-memory argument is a study by Williams (1994), which
followed up 129 women who had been treated at a particular
hospital, as children, for sexual abuse. She
interviewed hem 17 years later, under the guise of a
consumer survey of "the lives and health of women who had
been treated at" that hospital. The interviewees were
not reminded of their victimization history. But when
they were asked about their childhood sexual history, 38% of
the respondents did not report the index experience of
childhood sexual abuse. Williams suggested that this
reporting failure reflected a trauma-induced amnesia for the
victimization experience. But, of course, there are a
lot of problems with the study. Chiefly, it confuses
memory failure with a failure to report a particular
memory. Imagine, for a moment, participating in what
is essentially a consumer satisfaction survey: you are not
necessarily going to tell the interviewer that you were the
victim of sexual abuse as a child.
The trauma-memory argument has served as a plot device in
countless novels and films, but it turns out that
there is very little evidence for it.
And that pretty much accounts for all the evidence.
It is important to note that most of the evidence supporting the trauma-memory argument is retrospective in nature, meaning that both the trauma and the amnesia for it are reported after the fact. Consider, for example, the following question, close variants on which feature in many retrospective studies of amnesia for childhood sexual abuse:
"During the period of time between when the first forced sexual experience happened and your eighteenth birthday was there ever a time when you could not remember the forced sexual experience?"
The problem is that, in these cases, neither the claim of childhood sexual abuse nor the claim of intervening amnesia is amenable to independent corroboration. In other words, the ostensible trauma victim reports having been abused, and also reports having forgotten the abuse. This isn't traumatic amnesia. It is solipsism.
What is needed, then, are prospective studies of trauma victims -- studies in which trauma victims are objectively identified, and then interviewed to determine the extent of their memory for the trauma. But even a prospective study does not guarantee success. After all, the Williams (1986) study was a prospective study, with independently verified cases of childhood sexual abuse; and it still was highly problematic.
In an important review, Pope et al. (2000) surveyed all published prospective studies of victims of various kinds of trauma such as the "Coconut Grove" nightclub fire in Chicago, the victims of the Chowchilla kidnapping, and survivors of the Holocaust (as well as other concentration-camp inmates and prisoners of war), children (aged 2-7) who had been swept up in the Buffalo Creek flood, and the Loma Prieta earthquake, the amnesia associated with "battle fatigue" or "war neurosis", and crime victims (as well perpetrators who claimed to be amnesic for their crimes) and found not one single instance of a trauma victim who could not remember the event, and whose amnesia, if indeed that was what it was, could not be accounted for by "organic" factors or normal, non-pathological processes. Not a single case.
Still, some were not convinced. Brown et al. (1999) offered eight studies in which, they claimed, some trauma victims were amnesic for their trauma. However, Piper et al. (2000) systematically dismantled each claim.
In order to understand what's going on in this debate, it's
important to understand the criteria that are necessary to
confirm the diagnosis of traumatic amnesia.
As Pope et al. note, almost all of the studies claiming to provide evidence for traumatic amnesia fail to meet even elementary standards of methodological rigor -- mostly because they are retrospective studies that do not provide independent corroboration of the claimed trauma; in addition, most of them rely on the subject's retrospective self-report to document the amnesia as well as the trauma. What is needed are prospective studies in which the trauma is independently (and objectively) corroborated, and in which we can distinguish between amnesia and nondisclosure, and between psychogenic amnesia (i.e., "repression") and amnesia from other sources (e.g., infantile and childhood amnesia). There exist fewer than a dozen methodologically adequate prospective studies, and they indicate that amnesia is a vanishingly rare response to trauma. The vast majority of trauma victims remember their traumas perfectly well.
In the final analysis, traumatic amnesia, mediated by a defensive process of repression or dissociation, appears to be part of clinical folklore, but not part of clinical science. It is, if you will, a sort of urban myth.
Another familiar phenomenon of emotional memory is the flashbulb memory, in which subjects remember the circumstances under which they first learned about a surprising, consequential, affect-laden event. For members of the "baby-boom" generation, a classic example is the assassination of President John F. Kennedy. For younger individuals, as well as boomers, other familiar examples are the space shuttle Challenger disaster of 1986 and the terror attacks of September 11, 2001.
In a classic study, Brown and Kulik
(1977) studied a number of flashbulb memories (including the
Kennedy assassination), and identified a number of features
that flashbulb memories have in common:
Flashbulb memories for the JFK
assassination occurred in the vast majority of subjects;
flashbulb memories for the assassination of Malcolm X and
Martin Luther King, Jr., were much more common in blacks
than in whites.
Of course, remembering where you were when you learned of something is not all that special: environmental context is, after all, a feature of every episodic memory, which is defined as memory for events that are unique in space and time. And representation of the subject's own emotional response could be taken as an aspect of self-reference, which is also a necessary feature of every episodic memory. Still, hardly any of us remember where we were when we learned that Columbus discovered America in 1492, or that Abraham Lincoln was assassinated at Ford's Theater, or how we felt about either event. Still, it's the whole package of features that makes flashbulb memories seem special.
Mullane Swar
and Kihlstrom (2002) examined each of these features in a
study of memory for the Challenger disaster, which occurred
in 1986. In 1989 a group of college undergraduates
were asked to write down their recollection of the
Challenger disaster. They then completed a
questionnaire concerning their memories.
Their
free-recall
protocols were then coded for Brown and Kulik's 6 "canonical
categories", plus 3 additional categories suggested by other
research. Most subjects remembered where they were
when they learned the news, who the informant was, and who
else was present. Most subjects' memories had at least
five of the nine features that were coded.
The
features
occurred even more frequently when the subjects were asked
directly about them on the questionnaire. In this
case, the vast majority of subjects' memories had all nine
features.
The average
subject's free-recall protocol included four of the six
"canonical categories" identified by Brown and Kulik (1977),
while the probes on the questionnaire tended to elicit all
six.
Flashbulb memories get their name because it seems that the mind has taken a snapshot of the situation -- by virtue of what Livingston (1967) referred to as a "Now Print!" mechanism -- one which has evolutionary significance because it insures a prompt, enduring record of critical events.
Brown and Kulik expanded on Livingston's idea, suggesting that events associated with a high level of surprise, and a high level of consequentiality (at least, for the subject himself), and a correspondingly high level of emotional arousal, induce overt and covert rehearsal processes, resulting in a richly detailed, highly elaborated autobiographical narrative.
Flashbulb memories also seem to reflect the operation of the affective intensity effect, and underscore the finding by Cahill et al. (1999) that high levels of emotional arousal are associated with good, as opposed to poor, memory. Not for no reason are vivid, intrusive memories a cardinal symptom of post-traumatic stress disorder.
But flashbulb memories also seem to contradict
the Reconstruction Principle. In this case, it seems
as if a veridical representation of an event has been
indelibly stored in memory. Remembering seems to be
more like taking a book off the shelf and reading it, rather
than writing it anew from fragmentary notes. But
appearances may be deceiving.
In fact, beginning with the Challenger disaster, and continuing with such events as the Loma Prieta earthquake of 1989, the resignation of Margaret Thatcher as prime minister of England, the assassination of Swedish prime minister Olaf Palme, and the terrorist attacks of 9/11. researchers have scrambled to collect subjects memories as close as possible to the time of the event itself (often spending a long night at the photocopy machine, preparing questionnaires to distributed to their classes the next day); then these same subjects are asked to recall the event again sometime later, permitting a comparison of immediate and delayed recollection.
In
the first
study of this sort, Neisser and Harsch collected flashbulb
memories of the Challenger disaster in 1986, and then
retested the same subjects two years later, in 1988.
The subjects were highly confident of the accuracy of their
recollections. But N&H found that, compared to the
immediate recollections, the delayed recollections were full
of errors. And so have other investigators, who have
done similar studies. So, flashbulb memories may be
very vivid, and detailed, and remembered with a high degree
of confidence -- but that doesn't make them particularly
accurate representations of past events.
Along the same lines, Larsen (1992) performed
a "self-study" -- that is, a study of his own memories -- to
compare the accuracy of recollections of personal and public
events (there is a long history of such studies, beginning
with one performed by Harriet Linton).
We can conclude from studies like these that,
appearances to the contrary notwithstanding, flashbulb
memories are not exceptions to the rule. Flashbulb
memories are very vivid, and remembered with a high degree
of confidence, but like "ordinary" memories, they are not
particularly complete or infallible; they are not immune to
forgetting; and they are not produced by special
mechanisms. Flashbulb memories, like ordinary
memories, are reconstructions, not reproductions.
A case in point: my own flashbulb memory of the assassination of President John F. Kennedy (viritually every American in my age cohort has one). I was in biology class, in my upstate New York high school, taught by Mr. C., 7th period: an announcement came over the public address system that school was being dismissed early, and we were to go home or head for our school buses. The wall clock over the classroom door registered almost 2 PM. I can see Mr. C., in his white shirt and striped tie, with his lock of white hair (it stood out against the rest of his hair, which was jet black); Nancy J. sitting in front of me, with a white blouse and gray skirt; John Kl. to my right, Alan M. behind me, in chinos and a blue bleeding-madras short-sleeved shirt. It's all very vivid, as if I took a picture of the moment and preserved it forever.
But then again.... There are no contemporaneous records to check my memory against, and I've never checked my memory with the others who were in it, but frankly a lot of it smacks of reconstruction -- that is, a memory constructed based on plausible inferences rather than any kind of mental snapshot. Kennedy was shot in Dallas about 12:30 PM CST, and was declared dead about a half-hour later, which matches my memory of 2:00 PM EST. I don't actually remember whether we were told the President had been shot; I think we were just told that school was being dismissed early (nobody told kids anything then). Mr. C., seated us in alphabetical order, so I know that Nancy was in front of me, John to my right, and Alan in the row behind. Like most male teachers of the era, Mr. C. wore a jacket and tie to school, the shirts usually white and the ties usually striped -- though it's possible he was teaching in shirtsleeves that day. Nancy and I were dating at the time, John and Alan were good friends. Nancy, like most high-school girls at the time, typically wore a blouse (usually white) and skirt (usually dark) -- slacks, much less shorts, were not permitted. Many high-school boys, influenced by pop stars like the Kingston Trio, wore chinos and madras shirts. I owned several, and I bet that Alan did too -- though I doubt he was wearing short sleeves in late November in upstate New York. In any event, unlike Mr. C., Nancy, John, and the clock, Alan wasn't actually in my field of view.
Still, whether they are accurate or not, our memories, including our flashbulb memories, are -- well, they're our memories -- our mental representations of the past. And because they typically represent surprising, consequential, emotional events, they're relevant to the question of emotion and memory.
Regardless of whether they're accurate, flashbulb memories are our recollections of important events. And "importance" is something that differs from group to group (remember, in the Brown & Kulik study, that relatively few white subjects had flashbulb memories for the assassination of Medgar Evers). And it's also something that differs from individual to individual. I have a flashbulb memory for the time I first met my spouse, but I don't have flashbulb memories for everyone I ever met for the first time. An uninvestigated aspect of flashbulb memories are those idiosyncratic events that are important to us as individuals, even if they're not important to the wider public or history. For that reason, our flashbulb memories -- not necessarily of 9/11 and the Challenger Disaster, but the more "mundane" memories for things like our first kiss -- are important expressions of our personality. And, it turns out, they are also important for social interaction. So, we'll take them up again in the lectures on Personality and Memory and Social Memory.
Emotion and motivation are obviously related:
we tend to approach what gives us pleasure, and we tend to
avoid what causes us pain. But they're not the same
thing.
But before we can explore the (rather limited) literature on motivation and memory, we've got to talk a little bit about motivation in general.
As implied by the remarks immediately above, we can define the domain of motivation as a conative state of approach or avoidance.
This page last modified 11/20/2008.