Memory, Mood, and Consumer Judgment

Thomas S. Srull, University of Illinois
ABSTRACT - The present paper serves as a progress report of a current program of research aimed at understanding the effects of subjective mood states on the processing of information presented in advertisements. In particular, the distinction between retrieval and computational processes is used to construct a model of how subjective affective states influence product evaluations. Four experiments are used to test various aspects of the model. The data are consistent with the model in most respects-and they suggest that several important variables have previously been ignored.
[ to cite ]:
Thomas S. Srull (1987) ,"Memory, Mood, and Consumer Judgment", in NA - Advances in Consumer Research Volume 14, eds. Melanie Wallendorf and Paul Anderson, Provo, UT : Association for Consumer Research, Pages: 404-407.

Advances in Consumer Research Volume 14, 1987      Pages 404-407

MEMORY, MOOD, AND CONSUMER JUDGMENT

Thomas S. Srull, University of Illinois

ABSTRACT -

The present paper serves as a progress report of a current program of research aimed at understanding the effects of subjective mood states on the processing of information presented in advertisements. In particular, the distinction between retrieval and computational processes is used to construct a model of how subjective affective states influence product evaluations. Four experiments are used to test various aspects of the model. The data are consistent with the model in most respects-and they suggest that several important variables have previously been ignored.

INTRODUCTION

Researchers are now paying considerable attention to the effects of subjective mood states on various aspects of consumer behavior. The underlying reason for this attention is clear: mood matters. Reliable mood effects have been reported from several different laboratories, and by researchers of widely divergent theoretical persuasions. Moreover, a wide array of experimental paradigms and mood manipulations have been employed. In short, people starting at very different places have all come to the same conclusion. Mood is important and its effects need to be accounted for.

Although mood can affect many different facets of consumer behavior (see e.g., Gardner 1985), most work continues to be in the area of memory and judgment. Bower and his colleagues have reported an impressive series of studies demonstrating that mood can have strong and consistent effects on memory (Bower 1981; Bower, Gilligan, and Monteiro 1981; Bower, Monteiro, and Gilligan 1978). Other studies have found that these effects extend into the domain of judgment and behavioral decision making (e.g., Isen 1984; Isen, Clark, and Schwartz 1976; Isen, Means, Patrick, and Nowicki 1982).

Researchers concerned with advertising have found many implications of this work. It is possible, for example, that media contexts can create a particular mood state which, in turn, affects a consumer's memory, brand evaluation, and so on. Alternatively, the ad itself may create a mood that has similar effects (cf. Aaker, Stayman, and Hagerty 1986). In both cases, the ultimate effect of ad exposure would be a function of both the information conveyed in the ad and the consumer's subjective affective state. In other words, the affective state of the consumer modulates the effect of the ad.

The effects of mood pose a serious challenge to information processing approaches to consumer behavior. Philosophers and psychologists have long made a distinction between cognition and affect, thinking and feeling, and the rational and irrational sides of existence. How seriously should such dichotomies be taken today? What role, if any, should they play in our theoretical models? Do they suggest that the "laws" of behavior will be different, depending upon whether the cognitive system or the affective system dominates?

One might think that mood effects are important because they are outside the boundaries of information processing models. I believe that this is a mistake. If information processing is truly a metatheoretical system, then mood effects must be accounted for in the same way as any other type of effect, namely by specifying the computational mechanisms that are involved. Mood effects pose an important and difficult challenge because information processing theorists have seldom dealt with the affective system. However, there is nothing in principle that prevents an explanation of such effects in information processing terms. At the very least, it would seem premature to conclude otherwise. In other words, there is no reason for why this should not be treated as an open question.

Let us assume that information processing theorists must, at the very least, attempt to account for mood effects in terms of their models. If for no other reason, this should be done to test the limits of the information processing approach. How might this be accomplished? One way to begin is to delineate when mood does and does not have an effect in terms of the cognitive processes that are activated at any given time. Another related approach is to look across groups of subjects when there is reason to believe the groups use different processes to perform the same task.

The general issue of whether mood can be accounted for in information processing terms is an extraordinarily difficult one. In fact, it is probably premature at this time to even ask such a question because so little is known about mood effects themselves. My objective in the remainder of this paper is much more modest. Specifically, I will try to introduce a general approach that can be used to study mood within an information processing framework.

A PRELIMINARY MODEL

The purpose of the first experiment was to examine whether an information processing model can make a specific a priori prediction about when mood will affect performance on a particular task. In this case, the task is one of forming a product evaluation and the model is one that was initially outlined by Lichtenstein and Srull (1985).

The model of Lichtenstein and Srull (1985) distinguishes between on-line and memory-based processing (for related discussions, see Hastie and Park 1986 and Lynch and Srull 1982). Consider first on-line processing. This would occur whenever a person acquires product-related information with the implicit or explicit objective of making an evaluation of that product. When this occurs, a global evaluation of the product will be made at the time of information acquisition and stored in memory separately and independently from the specific information presented in the ad.

Memory-based processing is quite different. In many cases, a person will acquire product-related information with no specific objective in mind, or only a very general objective such as to comprehend the information being presented. Under these conditions, a global evaluation of the product will typically not be made at the time of information acquisition. If later asked to make a specific evaluation of the product, the person will be forced to retrieve the previously acquired information from memory, or some subset of it, and use this information as a basis for his/her evaluation of the product. In other words, a judgment will need to be computed on the spot.

Such a model has been found to be very heuristic. For example, one prediction is that the correspondence between the evaluative implications of whatever facts can be recalled and the global evaluative judgment that is made should be much higher in memory-based than on-line processing conditions. In an initial test of this hypothesis, Lichtenstein and Srull (1985) found in 12 out of 12 independent comparisons that the correlation between recall and judgment was higher in memory-based than on-line conditions. Moreover, the correlations in the memory-based conditions were universally large and statistically different from zero. The mean correlation across 12 conditions was .64 in the memory-based condition and .22 in the on-line condition.

More recently Hastie and Park (1986) have replicated these effects by using a slightly different paradigm. The online condition was very similar to that described earlier. However, for a memory-based condition, they had subjects anticipate making one judgment, but later asked them for a different, unrelated judgment. Across four separate experiments, the average correlation between recall and judgment was .51 in the memory-based condition and .16 in the on-line condition. Srull (1985) also reported several conceptual replications using slightly different orienting tasks. In short, these findings are very strong and replicate across a variety of laboratories, stimulus sets, content domains, and delay intervals.

EXPERIMENT 1

The theoretical model described above has been quite successful in accounting for past research, and it has been directly supported in several recent experiments (see e.g. Srull 1985). For both of these reasons, it is useful to ask whether the model can also be used to conceptualize the role that subjective affective states might have on product evaluation. In fact, extending the model into this domain would appear to be a very straightforward process. If the processing objectives of the subject are a determinant of exactly when a product evaluation occurs, they should also affect the influence of subjective affective states on the judgments that are made.

The model presented suggests that affective states at the time of encoding will influence the judgments of subjects using on-line processing because the judgment is being formed at the same time the affective state is being experienced. On the other hand, affective states during encoding should not affect the evaluation of subjects using a memory-based processing strategy. This is because the evaluation is not computed until the time of judgment.

Subjects in the first experiment were undergraduate students who came into the laboratory and were put into either a positive, neutral, or negative mood using the involved recollection procedure previously described by Srull (1983). In brief, subjects enter a quiet, dimly lighted room and are encouraged to completely relax. They are then asked to privately recall everything possible from a previous strongly affectively toned event in their personal life (except in the neutral condition). Every few minutes, subjects are given a "probe" that encourages them to concentrate on every detail concerning what they were thinking and how they felt during the actual experience.

The procedure is similar to those that attempt to induce mood states with hypnotic suggestion (see e.g., Bower 1981), except that subjects are not self-selected and there is no attempt to put them into a true hypnotic trance. Nevertheless, the mood states that result can be quite intense. The procedure has proven to be quite effective and some of the effects it produces are inconsistent with the demand characteristics of the situation (Srull 1983). Thus, there is converging evidence that an actual mood state is phenomenally being experienced.

Following the mood induction procedure, subjects were shown a single (modified) print ad. It contained ten separate attribute values for the Mazda RX7. The information was adapted from ads actually used by the company. It was changed simply by revising the sentences and putting them into paragraph form.

Subjects read the information at their own pace under one of two conditions. Subjects in the on-line condition were told to read the ad with the purpose of forming an evaluation of the product so that they would later be able to judge how desirable it would be relative to other competing brands. Subjects in the memory-based condition were told to simply comprehend what was being said.

After subjects had read the ad, they were dismissed and asked to return in 48 hours. At that time, all subjects were put into a neutral mood and asked to evaluate the product without being re-exposed to the original ad.

Specifically, subjects were asked, "Assuming you wanted to purchase a product similar to the Mazda RX7, how desirable do you think this particular brand would be?" Subjects made their ratings on a scale ranging from O ("very undesirable") to 20 ("very desirable").

The results provided considerable support for the model. In the on-line processing condition, the mean evaluative rating was 13.6 in the positive mood condition, 12.4 in the neutral mood condition, and 10.2 in the negative mood condition. Thus, the mood state of the subject had a strong and consistent impact. Ratings were displaced from their neutral baseline in both the positive and negative mood conditions, indicating that the mood effects were symmetrical.

Ratings in the memory-based condition were quite different. The mean evaluative rating was 12.6 in the positive mood condition, 11.9 in the neutral mood condition, and 12.2 in the negative mood condition. Note that subjects in a positive mood gave less positive ratings than those in the on-line condition, and subjects in a negative mood gave less negative ratings than those in the on-line condition.

In summary, mood had a much greater influence in the online condition than in the memory-based condition. It is important to remember, however, that only mood at the time of encoding was manipulated. Theoretically, subjects in the on-line condition were making their judgments of the product at the precise time the mood state was being experienced. According to the model, however, this was not true of those in the memory-based condition. Theoretically, these subjects would be forced to compute the judgment at the time it was requested, a time at which they were all in a neutral mood.

EXPERIMENT 2

The model suggests that the mood state of the subject will only have an effect on product evaluation if the mood is being experienced while the evaluation is taking place. The results of the first experiment are consistent with this in that mood at the time of encoding only had an effect in the on-line processing condition. Theoretically, those engaged in on-line processing were computing the judgment at the time of input (encoding).

Subjects who engage in memory-based processing are assumed to use a much different process. Because no evaluation is computed on-line, any such judgment must be made at some later time. In order to form a judgment, subjects must search memory for relevant information and integrate it using some type of combination rule. In other words, the product evaluation is computed at the time of judgment rather than at the time of input. One implication of this is that affective states at the time of judgment should affect the evaluations of those subjects using memory-based processing. This is so because the computation will be made while the affective state is being experienced. However, when subjects form their evaluations on-line, affective states at the time of judgment should have no effect.

Experiment 2 tested these predictions by using a procedure comparable to that used in the first experiment. All subjects were first placed in a neutral mood. Then they were-shown the ad for the Mazda RX7 and told to read the information at their own pace. As before, subjects in the on-line condition were told to read the ad with the purpose of forming an evaluation of the product so that they would later be able to judge how desirable it would be relative to other competing brands. Subjects in the memory-based condition were told to simply comprehend what was being said.

After subjects had read the ad, they were dismissed and asked to return in 48 hours. At that time, they were put into either a positive, neutral, or negative mood and then asked to evaluate the product without being re-exposed to the original ad.

Although not quite as strong as in the first experiment, the data are quite consistent with the model here as well. For those subjects who engaged in on-line processing, the mean product evaluation was 11.8 in the positive mood condition, 11.9 in the neutral mood condition, and 12.2 in the negative mood condition. Note that the means associated with each of the mood conditions are very close to the neutral baseline. As expected, mood at the time of judgment has no effect when on-line processing has al

ready occurred.

Those subjects who engaged in memory-based processing showed a quite different pattern of results. As predicted, mood had a strong and consistent effect in this case. The mean evaluative rating was 12.9 in the positive mood condition, 12.2 in the neutral mood condition, and 11.1 in the negative mood condition. Theoretically, this is due to the fact that only memory-based subjects were making their evaluations of the product at the time the mood state was being experienced.

EXPERIMENT 3

In the first two experiments the processing objectives were experimentally manipulated and comparisons were made across experimental conditions. The philosophy behind the third experiment is somewhat different. Subjects are first classified as either "novice" or "expert" and the comparisons are then made across the groups of subjects. This is relevant to testing the model if one assumes that experts and novices naturally use different processing strategies. In other words, rather than experimentally manipulate processing objectives, they are left free to vary naturally.

It is reasonable to assume that experts and novices will process information obtained from an advertisement differently. Novices, by definition, have little prior information and only limited experience with the product class. As a consequence, they will be forced to find some algorithm to combine specific items of information. In contrast, experts will have a much richer knowledge base to begin with. They will know which attributes are most important, they will have prior beliefs about how attributes are related to one another, and they will have already formed many brand evaluations. In many cases, brand evaluations will be based on past judgments rather than new information (see Lynch and Srull 1982). To the extent this is true, mood states at the time of encoding should have little effect.

The procedure of the experiment was similar to that used in the first study. Undergraduate students were first put into either a positive, neutral, or negative mood. They were then shown the same ad pertaining to the Mazda RX7 and given the same on-line processing instructions. Subjects were then dismissed and asked to return in 48 hours. At that time, all subjects were first put into a neutral mood and then asked to evaluate the product. At the end of the second session, subjects rated their self knowledge of automobiles in relation to the rest of the population and a median split was used to identify "novice" and "expert" subjects. This procedure was adapted from that used by Johnson and Russo (1981).

It was predicted that mood at the time of encoding would have a strong effect on the judgments of novice subjects but not on those of expert subjects. The data for novices show clear support for the hypothesis. The mean product evaluation of novice subjects was 14.16 in the positive mood condition, 12.09 in the neutral mood condition, and 10.87 in the negative mood condition. The means in both mood conditions are displaced from their neutral baseline condition. Theoretically, this is due to the fact that subjects were forming their evaluations at the time they were experiencing the mood state.

The data for expert subjects are a little more ambiguous. Contrary to what was predicted, mood at the time of encoding did have a systematic effect. On the other hand, the size of this effect was much smaller than that found with novice subjects. The mean product evaluation for experts was 14.24 in the positive mood condition, 14.17 in the neutral mood condition, and 14.03 in the negative mood condition. Relative to the novice condition, mood had the same type of effect but the magnitude of the differences was much smaller. a us, qualified support for the more general model was obtained.

EXPERIMENT 4

The procedure of Experiment 4 was identical to that used in the last study with two exceptions. First, >11 subjects were put into a neutral mood at the time of encoding. Second, subjects were put into either a positive, neutral, or negative mood at the time of judgment. It was assumed that novice subjects would form their evaluations of the product at the time of input. a us, when asked to make a specific judgment during the second session, they would simply retrieve the prior evaluation from memo wi. To the extent this is true, mood at the time judgment should have no effect.

The prediction for expert subjects is more or less the same. These subjects should also not have to compute their evaluation at the time of judgment. Rather, one would expect that the judgment has already been formed and stored in memory. If this is true, one would expect that mood at the time of judgment would have no effect on the evaluations of expert subjects.

The data offer mixed support for both hypotheses. In both cases, mood at the time of judgment had a systematic effect but, in each case, the magnitude of the effect was very small. The mean product evaluation of novice subjects was 14.12 in the positive mood condition, 13.18 in the neutral mood condition, and 12.67 in the negative mood condition. One can see how small these effects are by comparing them to those obtained in Experiment 3.

The data from expert subjects are similar. the mean product evaluation for expert subjects was 14.63 in the positive mood condition, 14.08 in the neutral mood condition, and 13.76 in the negative mood condition. As in Experiment 3, mood had a systematic effect even though the model predicts otherwise, It is also true, however, that these effects are very small compared to those obtained in the previous experiments.

DISCUSSION

AS noted earlier, mood effects are sometimes thought to be outside the boundary conditions of information processing models. Such a conclusion is premature and certainly inconsistent with the data reported in the present paper. In fact, the present results suggest that at least some mood effects can be accounted for quite well in information processing terms. A reasonably simple model was developed and used to illustrate how specific a priori predictions pertaining to mood can be derived. Moreover, the effects reported in the present paper are quite similar to those obtained in many other paradigms. Thus, the generalizability of this approach would appear to be very promising.

Experiments 1 and 2 provided strong support for the model. The data from the last two experiments, both of which pertained to expert-novice differences, were a little more ambiguous. Strong and consistent mood effects were observed in each case the model predicted they would be. However, mood effects were also observed in several cases where the model suggests mood should not have an impact. It is useful to consider why the model broke down under these conditions.

m e fact that these mood effects were so small offers some clue. One possibility is that the criterion used to classify subjects as "expert" was simply too liberal. Is it reasonable to assume that 50% of the population is expert in any given domain? If not, some "true novices" would have been classified as experts, and these misclassified subjects could have produced the small effects that were observed. Consider Experiment 3 for example. If some of the subjects classified as experts were actually computing their evaluation at the time of encoding, the effects observed would not be surprising.

Another possibility is that the retrieval process attributed to experts is too simple to model their actual performance. As Logan and Cowan (1984) have pointed out, there may be a "race" between an algorithm that is used to process new information and an attempt to retrieve a prior evaluation. In most cases, the retrieval process will be faster and used to make the judgment. However, the algorithm may win the race in a small percentage of cases. This would also predict a mood effect, but one that is much smaller than that observed for novices.

A final possibility is that the judgment and translation assumptions are too simple. The prior evaluation that is retrieved by an expert is unlikely to be in the same form as that requested by the experimenter. Thus, it will need to be translated and mapped onto whatever particular scale is used. It is possible that mood has some effect on this mapping process.

All of these possibilities are viable and deserve to be explored in future research. Each of them provides a mechanism for providing a fuller account of the data. The model will certainly become more complex as a result. However, this is not too surprising since only very simple processing assumptions were made to begin with.

The more general point is that information processing approaches to studying mood should not be dismissed prematurely. Accounting for affective processes is a definite challenge. But it is a challenge that must be met. Even if such models are ultimately rejected, we are sure to learn a great deal in the process of exploring their implications.

REFERENCES

Aaker, D.A., Stayman, D.M, and Hagerty, M.R. ( 1986), "Warmth in advertising: Measurement, impact, and sequence effects," Journal of Consumer Research, 12, 365-381.

Bower, G.H. (1981), "Mood and Memory," American Psychologist, 36, 129-148.

Bower, G.H., Gilligan, S.G., and Monteiro, K.P. (1981), "Selectivity of Learning Caused by Affective States," Journal of Experimental Psychology" General, 110, 451-473

Bower, G.H., Monteiro, K.P., and Gilligan, S.G. (1978), "Emotional Mood as a Context for Learning and Recall," Journal of Verbal Learning and Verbal Behavior, 17, 573-585

Gardner, M.P. (1985), "Mood States and Consumer Behavior: A Critical Review," Journal of Consumer Research, 12, 281-300.

Hastie, R. and Park, B. (1986), "The Relationship Between Memory and Judgment Depends on Whether the Judgment Task is Memory-Based or On-Line," Psychological Review, 93, 258-268.

Isen, A.M. (1984), "Affect, Cognition, and Social Behavior," in Handbook of Social Cognition (Vol. 3), eds. R.S. Wyer and T.K. Srull, Hill dale, NJ: Erlbaum.

Isen, A.M., Clark, M., and Schwartz, M.F. (1976), "Duration of the Effect of Good Mood on Helping: Footprints on the Sands of Time," Journal of Personality and Social Psychology, 34, 385-393.

Isen, A.M., Means, B., Patrick, R., and Nowicki, G. (1982), "Some Factors Influencing Decision-Making Strategy and Risk Taking," in Affect and Cognition, eds. M.S. Clark and S.T. Fiske. Hillsdale, NJ: Erlbaum.

Johnson, E.J. and Russo, J.E.,(1981), "Product Familiarity and Learning New Information," in Advances in Consumer Research (Vol. 8), ed. K.B. Monroe, Ann Arbor: Association for Consumer Research.

Lichtenstein, M. and Srull, T.K.,(1985), "Conceptual and Methodological Issues in Examining the Relationship Between Consumer Memory and Judgment," in Psychological Processes and Advertising Effects: Theory, Research, and Application, eds. L.F. Alwitt and A. A. Mitchell, Hillsdale, NJ: Erlbaum.

Logan, G.D. and Cowan, W.B. (1984), "On the Ability to Inhibit Thought and Action: A Theory of an Act of Control," Psychological Review, 91, 295-327.

Lynch, J.G. and Srull, T . K . ( 1982 ), "Memory and Attentional Factors in Consumer Choices Concepts and Research Methods," Journal of Consumer Research, 9, 18-37.

Srull, T.K. (1983), "Affect and Memory: The Impact of Affective Reactions in Advertising on the Representation of Product Information in Memory," in Advances in Consumer Research (Vol. 10), eds. R.P. Bagozzi and A. Tybout, Ann Arbor: Association for Consumer Research.

Srull, T.K. (1985), "The Relation Between Memory and Product Evaluation," Paper presented at the Fourth Annual Advertising and Consumer Psychology Conference, Needham Harper Worldwide, Chicago.

----------------------------------------