INTRODUCTION

Humans as well as honey bees need to remember important events to survive. Given limitations of cognitive capacity, however, they also need to forget those moments that are least important. For it is essential in a competitive environment to allocate scarce attentional and memory resources to critical events rather than to the mundane. To be successful when threatened by hungry carnivores our early mammalian ancestors could not be lost in thought considering the pleasures of the last meal. Similarly, for insects with brains the size of that of a honey bee, not too much memory can be allocated to remembering beautiful sunsets (McGaugh, 1990). The way that motivational and affective reactions to environmental stimuli and outcomes modulate memory seems to be one solution to the problem of remembering critical life events and forgetting the trivial. In the following pages we consider evidence from the personality and motivational literature that suggests that arousal, an intensity component of motivation, has a critical role in modulating the processes involved in the storage and retrieval of information. We propose that high levels of arousal facilitate the detection and long term retention of information but at a seeming cost of inhibiting immediate access to that information. We suggest that because many manipulations of affect and mood are also manipulations of arousal that supposed mood effects upon memory may be accounted for partly by the effect of arousal upon memory. We conclude that some of the controversy surrounding the effects of mood upon memory can be resolved by a proper appreciation of the effects of arousal upon memory.

Although our emphasis upon the arousal or intensity aspect of motivation is common in the individual differences and motivational literature, we suspect that our arousal oriented approach to the problem of how affect or emotion relates to memory differs from those with a more cognitive interpretation of affect (e.g. Ortony, Clore & Collins, 1990). Our orientation is quite different from those who view affect as just another memory code, the effects of which can be modeled in the same way as the type face or the color of the stimuli. We prefer to emphasize the important role that motivational and affective intensity play upon the very way in which information is detected, processed, and stored for later retrieval.

We have previously suggested that the motivational construct of arousal has a critical impact upon memory and that any study of the relationships between affect, mood and memory needs to take individual differences and situational sources of arousal into account (Revelle and Loftus, 1990). In the following pages we first review arousal as a useful psychological construct, discuss some of the controversy in its measurement and manipulation, consider how the construct of arousal has proven to be helpful in discriminating among competing models of personality, and then propose four different ways in which arousal affects cognitive performance. We then summarize some of the findings relating affect to memory and suggest that many of the manipulations used to change affect are in fact manipulations of arousal as well. We propose that some of the confusion relating affective state to memory performance is due to a lack of concern for the direct effects of arousal upon memory. Based upon our review we propose a model of how arousal and affect modulate memory. Finally we suggest what steps need to be taken in the study of mood and memory if the implications of the effect of arousal on memory are to be taken seriously.

In all of our discussion we emphasize that individual differences in arousal need to be taken into account if particular effects are expected to replicate across various situational manipulations. This partly reflects our own biases, but also reflects a good deal of prior research (reviewed in Anderson, 1989; Revelle, 1987, 1989) that has shown that what seems at first to be a morass of conflicting findings relating arousal manipulations to performance is actually much more orderly when the effects of several personality variables are taken into account.

AROUSAL AS A PSYCHOLOGICAL CONSTRUCT

In everyday terms, to be aroused means to be wide awake, alert, vigorous, excited, and full of pep (Thayer, 1989). To be unaroused means to be sleepy, sluggish, tired, and relaxed. More formally, arousal or activation "describes a condition conceived to vary in a continuum from a low point in sleep to a high point in extreme effort or intense excitement" (Duffy, 1962, p 5). Moderate levels of arousal can be equated with feelings of peppiness and vigor, although extremely high levels are felt as tense and unpleasant (Thayer, 1989). Arousal as a general construct may also be conceived of as the negative probability of falling asleep (Corcoran, 1965, 1972, 1981). It has been referred to as the non-specific component of motivation that reflects the intensity rather than the direction of motivation (Humphreys and Revelle, 1984) as well as the intensity rather than the evaluational quality of affect (Whissel, Fournier, Pelland, Weir & Makarek, 1986).

It is difficult to find an area of psychology where the construct of arousal has not played an important role. For instance, in real time theories of conditioning, arousal is necessary for some neural associations to form (Grossberg, 1987). In psychophysiology, arousal is the cause of desynchronization of the electroencephalogram (EEG), is a non-specific response to orienting stimuli (O'Gorman, 1977), and is a response to increases in task complexity (Berlyne, 1960). In some personality theories arousal level is thought to be maintained at an intermediate value as part of a homeostatic mechanism believed to cause differences in stimulation seeking and hyperactivity (H.J. Eysenck, 1967, 1981; Zuckerman, 1979). In theories of human performance, arousal has been discussed as a determinant of sustained performance in vigilance situations (c.f., Davies and Parasuraman, 1982; Mackie, 1977), and has been used as a way of relating individual differences in personality to the effects of stimulant drugs and time of day on cognitive performance (Revelle, Humphreys, Simon and Gilliland, 1980).

Theories of generalized arousal are not without their critics (e.g., Hockey, 1979, 1984; Neiss, 1988; Venables, 1984) or more accurately, the supporters of theories of generalized arousal are sometimes hard to find (but see Anderson, 1989; Gale, 1986, 1987; Humphreys & Revelle, 1984; Revelle, 1987, 1989). While many theorists recognize the appeal of a unified construct of intensity, the experimental evidence for the dissassociation of various arousal manipulations seems to some to be quite compelling[1]. Indeed, some disillusioned arousal theorists now refer to the study of the intensity aspects of performance with the less theory-laden term of "energetics" (Hockey, Gaillard & Coles, 1986).

Although the evidence is quite clear that there are many different biological responses to increased demands for energy expenditure, it is less clear that functionally these responses are not serving a similar purpose. Following Corcoran's very compelling definition of arousal as the inverse probability of falling asleep (1965, 1981), or the related hypothesis that arousal level is a positive function of energetic requirements, we propose that the low levels of energy expenditure associated with sleeping are a means of conserving physiological resources for more demanding activities. Conversely, high levels of arousal are associated with higher rates of metabolic response to task demands. That is, variations in arousal may be seen as serving the function of varying the resources available for information processing: "cardiovascular changes during information processing reflect the regulation of the organism's energetic economy in support of the specific processes required by the task at hand" (pg. 641, Jennings, Nebes & Brock, 1988).

Compelling demonstrations of how separate arousal systems can serve the same function are the studies by McGaugh and his colleagues who have examined how various stimulants, depressants, and neurotransmitters have similar effects upon memory modulation (McGaugh, 1990). The peripheral stimulant epinepherine, the central stimulant, norepinepherine, and the opiate antagonist, naxolone, all can be shown to modulate memory by their effect on the amygdala. Similarly, arousal changes associated with such widely disparate sources as stable personality traits (Introversion/Extraversion), time of day, stimulant drugs, white noise, evaluation by others and exercise have resulted in highly similar effects on memory (Revelle & Loftus, 1990).

Even if all variations in arousal might be serving the same purpose, it is not necessary that they are caused by similar mechanisms. That is, for primates accustomed to gathering fruits and seeds in the daytime while looking out for predators, conserving energy by sleeping at night is a very useful adaptation to lowered visual accuity and associated higher risk of predation at night. It is functionally more efficient to reduce one's rate of metabolic activity at times when demands are low than it is to keep a constant rate. Consequently, a diurnal variation in arousal is a useful means of coping with the diurnal variation in task demands from the environment. An organism with a very strong diurnal metabolic system, however, will be particularly sensitive to predation at low points in its metabolic cycle. (Monkey's that fail to wake up when attacked by pythons late at night will have a lower reproductive fitness than those that do wake up and are able to flee.) Thus, there is a need to have a separate system that can override or compensate for the normally low nocturnal levels of arousal - a system more capable of responding in an acute sense to immediate demands within the environment. It is possible to argue for a cascade of such control systems, each of which may have evolved separately to compensate for under or over control of the previous level of energy allocation. The function of all of these systems, however, may be seen as controlling the current level of energy expenditure. It is the functioning of this complex of systems that we choose to describe in terms of arousal.[2]

Measurement of Arousal

Given the long history of arousal in psychological theory, it would seem that the measurement of arousal would be straightforward. Unfortunately, this is not the case. Rather, the failure of different measures of arousal to agree with each other is perhaps the greatest source of dissatisfaction with the use of arousal as a unifying theoretical construct. Some of the sources of the controversy surrounding the use of arousal as a psychological construct include debates about broad issues in measurement, the importance of individual differences, the level of inference between the measure and the construct, and the effect of various situational manipulations.

Problems of measurement

At least four different issues need to be considered when discussing the measurement of arousal. Three of these involve the timing and the temporal resolution of the measures: the first considers how arousal measures vary across the psychological spectrum of temporal resolution; the second concerns the time course of various arousal measures; and the third involves the problem of relating measures based upon different sampling frames and latencies. A fourth issue is the problem of between versus within subject measures of arousal.

Arousal and the psychological spectrum. Psychological phenomena range across a temporal spectrum of at least 12 orders of magnitude: from the milleseconds used to index firing rates of neurons, to the seconds of a verbal learning study, to the hours of a vigilance experiment, and finally to the decades that make up a lifetime. Different psychological phemenona typically are measured at different temporal frequencies (or durations) across the spectrum. For example, Event Related Potentials (ERPs) have durations of 100-600 ms, priming effects for reaction time persist over only a few seconds, changes in affect takes tens of seconds to occur, stimulants such as caffeine have effects only after 30-40 minutes (~2 x 10³sec), and cognitive performance changes can range over the day (~10⁵sec), the menstrual cycle (~10⁶sec), a year of schooling (~3 x 10⁷sec), and over the life span (~3x10⁹sec).

Measures of arousal may be classified in terms of this psychological spectrum. At the shortest intervals are measures that include indices of cortical activity such as the EEG, both event related and resting frequency. At somewhat longer intervals are the autonomic measures of Skin Conductance (SC) and Heart Rate (HR). At even longer levels are endrocrine measures such as the level of Mono-Amine Oxidase (MAO) and general metabolic measures such as core Body Temperature (BT). Other measures with fine temporal resolution include psychophysical sensitivities to light and sound. At a less fine resolution are measures of activity level. In addition to the direct physiological measures are measures of self report of arousal. By asking subjects how peppy, active and vigorous they feel, it is possible to show general effects with durations from a few minutes to a few hours. In fact, as Robert Thayer (1986, 1989) has shown, self report measures seem to reflect the general factor of many of the finer grain physiological measures (see also Clements, Hafer & Vermillion, 1976 and Matthews, 1987, 1989).

Time course of arousal measures: growth, level, and decay of arousal. Arousal changes over time. Therefore, we need to be concerned with the rates of growth and decay, as well as the average level of arousal. Variables that affect one rate of change need not affect the others. Consider the various patterns of arousal that can occur as a result of different patterns of stimulation duration (both the on and off times) if change in arousal (dA) is represented as a function of excitatory sensitivity (e) to stimulation (S) and habituation which is a function of arousal level and some decay parameter (c): dA = eS - cA. Average arousal level is a function of e, c, and the length of time the stimulation is on or off. When measuring arousal level it is important to remember that these parameters of arousal are logically, and frequently empirically, independent.[3]

Latency and sampling frame. A classic problem in the measurement of arousal is how to relate measures with different onset latencies after the stimulus. Each index of arousal has different temporal parameters, with different delays and sampling rates, so that different indices reflect arousal averaged over different lengths of time and with different delays (e.g., EEG activation will occur within milliseconds of a stimulus onset, skin conductance changes after several seconds, self reports of activity and energy reflect arousal sampled over a longer period than either EEG or SC, and body temperature seems to indicate average activity over a period of minutes to hours). Depending upon the fineness of the temporal resolution of the stimulus, these different measures will respond differently. A missing stimulus in the P300 oddball paradigm that elicits a change in EEG evoked potentials, and perhaps an Orienting Response as indexed by a change in SC or HR, should not be expected, however, to affect ratings of alertness or core body temperature. These broader measures reflect changes in the system that are so slow as to be undetectable or are discarded by those who record ERPs or GSRs.

Within versus between subject measures. Several different questions arise with respect to the problem of measurements taken within subjects versus between subjects. Arousal is normally thought of as something which varies within subjects - a stimulus induces an orienting response, we become excited while giving a talk, exercise increases heart rate. While some measures (such as self report) are very useful indicators of arousal changes within subjects, they are of questionable validity between subjects[4]. For instance, it is difficult to know what different people mean by the term low arousal. What person A reports as being calm, person B might well report as excited and peppy. When one of us feels exhausted and thinks that he is talking very slowly, students report that his rate of talking in a lecture is finally at a normal level. This problem of calibration between subjects is not just for self report; for example, skin conductance differences can reflect skin thickness between subjects as well as differences in excitement within and between subjects.

Finally, it is important to note that when indices covary within individuals, they do not necessarily covary between individuals. Some individuals will respond to stimuli with greater changes in heart rate than in skin conductance, while others will have large increases in skin conductance but only small increases in heart rate. Within subjects, changes in SC and HR can correlate positively even though there is zero or a negative correlation between subjects. The question of how to aggregate data across subjects is particularly important for any consideration of the effect of arousal upon performance.

Individual Differences

That there are individual differences in measures of arousal at all levels of the psychological spectrum is not an issue. If there were no individual differences, then we would need no more than one subject per cell of our experimental designs. But since not all subjects within an experimental condition are the same, the question becomes whether the within experimental condition differences are predictable by extra-experimental means. That is, are there characteristics of subjects that allow us to predict how they will respond (behaviorally or physiologically) in a particular situation. The purpose of searching for individual differences is not to demonstrate that people differ but rather to find ways of predicting behavior across situations. What is important is that differences at long temporal intervals are able to predict differences on tasks with short durations and vice versa. We discuss such evidence later when relating personality to arousal.

Range of environmental conditions.

A consistent issue when searching for an arousal effect of any type is to specify the appropriate range of environmental conditions. There are very few interesting arousal effects or individual differences when subjects are dead or in a deep coma. Although an oversimplification, it is useful to consider person operating curves (POCs) plotting hypothetical levels of arousal (or any other outcome measure) as a function of environmental demand (Figure 1). Arousal is shown as a monotonically increasing function of the environment for three hypothetical subjects. At very high and very low levels of environmental stimulation, there are no meaningful differences. It is only at intermediate values that differences can be detected. In addition, note that one POC reflects a subject who is more sensitive to the changes in environmental demand than the others. That is, for individual C, the POC is steeper at the middle level of stimulation. For this person, low levels of environmental demand lead to very low levels of arousal and high levels of demand lead to very high levels. That is, at low stimulation levels, this person is less aroused than the others, but at high stimulation values, is actually more aroused. Although making the search for reliable arousal effects as a function of individual differences more difficult, it is possible to organize the literature reporting the effects of extraversion on basal frequency of EEG in just such a manner (e.g., Gale, 1981). Reliable differences in EEG frequency between introverts and extraverts only occur at moderate levels of environmental demand.

Insert Figure 1 about here

Direct versus indirect measures of arousal.

Most studies concerned with the measurement of arousal can be divided into those with an emphasis upon peripheral measures of autonomic activity such as skin conductance or heart rate, and the more central measures of cortical activity associated with variations in scalp potential as detected by the EEG. Unfortunately, these measures of the hand, the heart, and the head do not produce evidence for a unified arousal system (Lacey, 1967; Venables, 1984) From a functional point of view, the independence of the separate physiological measures is not evidence against the validity of the arousal construct. Rather it suggests that separate control processes have developed to modulate energy availabilty, and that it is this sum of available energy that people report as arousal.

An alternative to physiological measures is to assess the subjective feel of alertness and vigor by self report. As Thayer (1967, 1978, 1989) and Clements et al. (1976) have shown convincingly, when measures are taken within subjects, self reports of arousal correlate more highly with each of the separate physiological measures than the measures do with each other. This is, of course, just the pattern that would be expected if the self-report measures represented either the common factor of arousal, or the sum of activity in each of several independent systems.

Situational manipulations of arousal

There are many ways to manipulate arousal. Manipulations used to decrease arousal level include sleep deprivation, time on task, and depressant drugs (e.g., alcohol or barbiturates). Manipulations used to increase arousal include increases in stimulation (e.g., increases in noise, light, or somatasensory input), the use of stimulant drugs (e.g., caffeine, amphetamine, or nicotine), and light to moderate exercise. Variations in arousal also occur naturally throughout the day (diurnal rhythms) with most people reaching a peak of alertness in the late morning to mid afternoon and reaching a low point between 3 and 6 am.

Each manipulation of arousal has specific effects as well as a general effect. It is important when examining the relationships between various manipulations and measures to take into account the irrelevancies of each manipulation as well as the common effects (Anderson, 1989). For instance, the behavioral consequences of lowering arousal through sleep deprivation or sustained performance are quite different than those associated with the effects of alcohol. Similarly, noise and caffeine both increase arousal, but their effects on performance differ, at least partly because high levels of noise are distracting and can mask inner speech while high levels of caffeine can cause muscle tremor. Much of the arguement over the utility of the arousal construct centers on the relative importance of the specific versus common effects of arousal manipulations (Anderson, 1989).

The covariation of manipulations of arousal and affect:

That affect and emotion have both a directional and intensity component is well known. A great deal of research and debate has been concerned with the interplay between these two components and, more specifically, the question of which is primary in one's experience of an emotion - the intensity (arousal) or the directional (evaluative) component (Schachter & Singer, 1962). The question of primacy aside, it is clear that while emotions such as fear and happiness are easily categorized into separate evaluational or affective dimensions (positive vs. negative), other emotional terms such as nervous and afraid must rely on an intensity factor (high vs. low in its most simplistic form) for differentiation. Whissell et al. (1986) found that a two-dimensional space (Evaluation x Activation) represented the most parsimonious categorization system for emotion words. A alternative solution to the dimensions of affective space can be found in Watson and colleagues' two orthogonal factors of positive and negative affect (PA and NA; Watson, Clark & Tellegen, 1988; Watson & Tellegen, 1985).

Within the arousal domain, Thayer (1967, 1978, 1986, 1989) has explicated two orthogonal factors that can be related to dimensions of affect: energetic arousal and tense arousal. Energetic arousal and positive affect covary strongly as do tense arousal and negative affect. Thayer (1989) found that energetic arousal varies diurnally along with positive affect. Loftus (1990) found that an exercise induction (as compared to a relaxation induction) tended to increase positive affect while negative affect remained unaffected. Both energetic and tense arousal increased as a result of the exercise induction also. This result was in contrast to an earlier study performed by Thayer that suggested exercise increased energetic arousal and decreased tense arousal (Thayer, 1989).

Although the exact nature of the relationship between positive/negative affect and low/high or energetic/tense arousal has yet to be clearly delineated, it is clear that many manipulations of affect also are manipulations of arousal. The negative affect experienced while watching a horror movie is accompanied by an increase in tense arousal. The positive affect associated with a lively party is also associated with an increase in energetic arousal. It is this confounding of affective manipulations with arousal manipulations that we believe may account for at least part of the inconsistent findings (discussed later) within the study of mood-state dependent effects. In addition, it is likely that the so-called mood intensity effects are more attributable to the effects of the arousal levels present in the experiments than the affect.
More