Perhaps due to the pervasiveness of affect in everyday functioning, the study of the relationship between affect and memory is an intriguing one. A quick reading of the chapters available in this book provides only a brief sampling of the wide variety of issues that exist in the more general area of affect and memory. For instance, one might choose to focus on the effects of pathological affective states (i.e. depressive or manic disorders) or alternatively on the effects of more naturally-occuring affect such as the shifts in emotional state that occur as one goes about the activities of a normal day.
Given the vast number of phenomena encompassed by the study of affect and memory, perhaps it is not surprising that some areas of research have been plagued by inconsistencies in attempts to replicate promising early findings. Seminal attempts to study systematically the changes in information encoding and retrieval that occur under affectively valenced conditions garnered evidence of two effects - affect state dependent and affect congruent effects - that offered exciting possibilities for theories of both naturally-occuring and pathological shifts in affective states (Bartlett & Santrock, 1977; Bower, Montiero & Gilligan, 1978, Study 3; Bower, 1981). Much of this work was based on Bower's influential Network Theory of Affect (Bower, 1981; Gilligan & Bower, 1984). The original theory detailed a cue-based model of information processing in which affective information can serve to both cue retrieval of previously stored information or filter the nature of incoming information through spreading activation within the memory system.
Affect state dependent recall refers to a tendency exhibited by subjects in either a positive or negative affective state to remember information that had been learned under the same affective state (congruent conditions) better than that learned under a different affective state (incongruent conditions). Affect congruent (or mood congruency) effects are reflected in a tendency to encode or retrieve information of the same hedonic quality as the current affective state (i.e. happy or sad). Singer and Salovey (1988) further differentiate between encoding congruent and recall congruent effects within affect congruency to reflect the stage at which the mood condition is of interest. A less-well researched but related prediction made by the Network Theory of Affect is that of mood intensity effects. This effect predicts that learning is positively correlated with the intensity of a mood (Gilligan & Bower, 1984; Singer & Salovey, 1988).
While affect congruency effects have stood up fairly well to the passage of time and further study, affect state dependent recall effects have not (see Blaney, 1986 and Singer & Salovey, 1988 for more thorough reviews). Research of this latter effect has led to a confusing "now you see it, now you don't" pattern of results. Several further strong demonstrations of affect state dependent recall (Bartlett, Burleson & Santrock, 1982, Study 2; Mecklenbrauker & Hager, 1984) stand in stark contrast to frustrating failures to replicate even the original study itself (Bower & Mayer, 1985; Wetzler, 1985). Although there are doubtless several factors that might be at play in this confusing pattern of results, we have suggested (Revelle & Loftus, 1990) that one of the most straightforward explanations exists in the uncontrolled and unmeasured effects of the arousal present in the experimental conditions.
This finding, however, does little to explain the inconsistencies in this area of research. It is interesting to note that arousal may also result in state-dependent recall but hardly surprising. Rather, the focus here is on the process-based differences that may occur as a result of the uncontrolled and unmeasured presence of arousal within the experimental situation. The likelihood that arousal-based information processing effects are present is based on confounds seen most clearly in several aspects of these experimental paradigms: the affect inductions, the intervals between learning and recall, and the analyses performed.
More specifically, the affect or mood inductions employed in mood and memory studies vary widely. Different studies regularly employ different mood induction techniques (i.e. hypnosis, reading valenced statements, listening to music, false feedback) to achieve what are considered comparable levels of "positive" and "negative" or "happy" and "sad" mood states. Although this strategy enhances the generalizability of the results, it invites confounding of differential levels of arousal within similar classes of affect. For instance, it seems likely that listening to uplifting, bubbly music might induce a higher arousal state than hypnosis (which involves deep relaxation) - although both have been used to derive "happy" mood conditions. There has also been some suggestion that a commonly-used mood induction, the Velten procedure (Velten,1968), differs significantly in the percentages of arousing and nonarousing words that are used in each of the mood conditions (depression, elation and neutrality) (Whissell & Levesque, 1988).
This problem is further complicated by the tendency for many researchers in this area to fail to perform sufficient manipulation checks for the inductions used - and when manipulation checks are performed, arousal is rarely measured. Eich and Metcalfe (1989) represent one exception in that arousal as well as affective levels were measured after a musical mood induction. It is interesting to note that the subjects in the happy mood condition rated themselves as being not only happier but also significantly more aroused than the sad mood condition subjects across seven points in time.
A second factor related to the suspected presence of arousal-type effects within these experimental paradigms is that of the recall intervals employed between initial learning and recall trials. Some studies do not even report the intervals used; others have used intervals than range anywhere from immediate recall to recall some 20 minutes (Schare, Lisman & Spear, 1984), 40 minutes (Bower & Mayer, 1985), 5 hours (Bower, Gilligan & Montiero, 1981), 24 hours (Wetzler, 1985) and 4 days later (Weingartner, Miller & Murphy, 1977). Moreover, most studies employ a multiple-list learning and recall paradigm which involves a variety of intervals for each study. Once again, although such a variety of intervals might enhance the eventual generalizability of results, this mixture of shorter and longer term retrieval effects has likely served to significantly increase the inconsistency inherent in the mood-state dependent research. In addition, perhaps the most significant point to be made here is that in light of the interaction of arousal and recall interval discussed earlier, it seems highly probable that the recall performed in such studies is influenced by both cue-based affect and arousal processes as well as the processing changes due to the interaction of arousal and retentional interval.
The exploration of this literature for these predictions might seems to be the most reasonable tack. However, the paucity of arousal and recall interval information available within these studies makes reanalysis impossible (Loftus, 1990). Moreover, much of the research is analyzed in terms of "congruent" (mood state same at learning and recall) and "disparate" (mood state different at learning and recall) mood groups. Thus, positive and negative mood states are often collapsed in favor of this contrast. If the positive and negative mood states were associated with different levels of arousal due to the type of induction used and/or natural covariation, this practice effectively wipes out evidence for arousal effects.
A more clearcut reinterpretation of affect-dependent effects on memory lies in the research related to the mood intensity hypothesis. The mood intensity hypothesis predicts that affectively intense memories should be better recalled and over longer periods of time. This prediction and the classical evidence in support of it found this tendency to be independent of the affective nature of the material (Dutta & Kanungo, 1967, 1975; Singer & Salovey, 1988). More recent research, however, has suggested that this is particularly true of positive affective states whereas negative affective states interfere with the encoding of material (Ellis, Thomas, McFarland & Lane, 1985; Ellis, Thomas & Rodriquez, 1984). An arousal interpretation of these results would suggest that as the positive emotional states induced in these studies are most likely highly correlated with significant levels of arousal, the tendency for these materials to be better retained over time is most likely an arousal-based enhancement of longer term retention.
It is our suspicion that both arousal and affect have substantive and some
qualitatively different effects on information processing. As such, the
importance of considering both dimensions of emotion in experiments that seek
to change either arousal or affective state should not be minimized. As the
earlier summary of the arousal literature suggested, arousal tends to enhance
information transfer, decrease short term retrieval capacity, and increase long
term retention; most likely, it also has some sort of inverted-U relationship
with retrieval processes in general. Affect, on the other hand, appears to
have a significant tendency to bias the encoding and retrieval of
hedonically-laden materials and, to some degree, affect the retrieval of
neutral information also. Given the natural interrelatedness of affect and
arousal, it would seem that the most reasonable tack would lie in experimental
paradigms that overtly cross both dimensions or, at least, study one dimension
while trying to hold the other constant. In both paradigms, accurate
measurement of the affective and arousal states that are induced and
appropriate analyses based on each are essential. Given the "capricious"
nature of mood-state dependent effects, it may very well turn out that the
existence of such effects is, to some degree, dependent on an interaction with
arousal-based effects on information processing.
More