Detecting Demand Characteristics in Laboratory Experiments in Consumer Research: the Case of Repetition-Affect Research

Alan G. Sawyer, University of Massachusetts
[ to cite ]:
Alan G. Sawyer (1975) ,"Detecting Demand Characteristics in Laboratory Experiments in Consumer Research: the Case of Repetition-Affect Research", in NA - Advances in Consumer Research Volume 02, eds. Mary Jane Schlinger, Ann Abor, MI : Association for Consumer Research, Pages: 713-724.

Advances in Consumer Research Volume 2, 1975      Pages 713-724

DETECTING DEMAND CHARACTERISTICS IN LABORATORY EXPERIMENTS IN CONSUMER RESEARCH: THE CASE OF REPETITION-AFFECT RESEARCH

Alan G. Sawyer, University of Massachusetts

[This research was supported by a grant from the Research Council of the University of Massachusetts.]

[Alan G. Sawyer is Assistant Professor of Marketing at the University of Massachusetts at Amherst.]

Demand characteristics in laboratory experiments are various aspects of the experimental environment which may be used by a subject as cues to the experimenter's hypothesis and as a guide to appropriate behavior. Four research methods--the non-experiment, the post-experimental inquiry, manipulation of hypothesized demand cues, and hetero-method replication-can help to detect demand characteristics. Research investigating the effects of repeated stimulus exposures on measures of acceptance, liking, or preference is presented to demonstrate the problems of demand bias in a particular repetition-affect experimental paradigm and to illustrate each of the four methods of detection.

The problem of demand characteristics in laboratory experiments concerns aspects of the experimental environment which may be used by the subject as cues to the experimenter's hypothesis and as a guide to appropriate behavior in the experiment.

Insofar as the subject cares about the outcome, his perception of his role and of the hypothesis being tested will become a significant determinant of his behavior. The cues which govern his perception - which communicate what the experimenter hopes to find - can therefore be crucial variables.... They include the scuttlebutt about the experiment, its setting, implicit and explicit instructions, the person of the experimenter, subtle cues provided by him, and, of particular importance, the experimental procedure itself. All of these cues are interpreted in the light of the subjects' past learning and experience. Although the explicit instructions are important, it appears that subtler cues from which the subject can draw covert or even unconscious inference may be still more powerful (Orne, 1969, p. 146).

The various types of demand characteristics and how they may affect subjects' behavior have been extensively discussed by social psychologists (e. g., Rosenthal and Rosnow, 1969; Weber and Cook, 1972; Rosnow and Aiken, 1973) and by consumer researchers (e.g., Olson, 1974; Sawyer, 1974). Four methods can help to detect demand characteristics. To both demonstrate the problems of demand characteristics and to illustrate the four methods, research investigating the effects of repeated stimulus exposures on measures of acceptance, liking, or preference for that stimulus will be discussed.

REPETITION AND AFFECT

Complete reviews of the effects of repetition on affective measures are available (e.g., Zajonc, 1968; Sawyer, in press) so only a brief summary of various procedures and results will be presented here. Zajonc (1968) reported that increasing the number of exposures of such stimuli as nonsense syllables, Chinese characters, and Turkish adjectives in a series of experiments resulted in increased liking for those stimuli. Zajonc exposed slides of the stimuli for two seconds each and varied the exposure level from 0, 1, 2, 5, 10, to 25 and found a positive monotone relationship between the logarithm of frequency of exposure and liking as measured by the subjects' estimates of whether the foreign words meant something good or something bad. This relationship has since been replicated several times. One experiment also found that the frequency-affect relationship remained the same when measured one week later. Moreover, with the Chinese characters as stimuli, the effects of increased exposure upon rated goodness of their meaning remained positive at exposure levels as high as eighty-one (Zajonc et al., 1971).

Research into the effects of repeated exposures on affective responses has produced mixed results in more consumer-oriented investigations. Becknell, Wilson and Baird (1963) studied the effect of repeated nonsense syllables on "brand choice." Along with print advertisements and landscapes, a set of nonsense syllables with frequencies of one, four, seven, and ten were shown. After one showing of the set, each respondent was asked to choose a gift of nylon stockings from two brands marked by nonsense syllables seen in the presentation. A second showing was followed by a paper-and-pencil ranking of four "brands." Responses to both measures were positively affected by repetition. Miller, Mazis, and Wright (1971) found positive effects of 0, 2, 10, and 20 exposures of different nonsense syllable brand names on rated liking for different characteristics linked to the brand names after they had been repeated. However, no effect of repetition on liking was found when subjects were also able to actually sample the product with which the repeated brand name was linked.

Consumer studies with actual product advertisements have generally found either neutral or negative affective responses to repetition. Light (1967) measured ad evaluation and ad satiation (the trend to neutrality on five semantic differential scales). Color slides of print advertisements were exposed in frequencies ranging from one to sixteen in two experimental conditions: high exposure density (fifty total exposures) and low density (seventy-five total exposures). Light found no evidence of a trend over repeated exposures to ad satiation and no effect of repetition on ad evaluation. Moreover, in the high density condition, there was a negative relationship between ad frequency and ad evaluation. The lack of any positive relationship of reactions to the ad and repetition in comparison to Zajonc's positive results is noteworthy.

Ray and Sawyer (1971) developed a laboratory procedure which measured several responses including brand evaluation (rating on a good-bad scale) and brand purchase intention (which brand in a product class would the subject be most likely to purchase). The true purpose of the experiment was disguised, and volunteer female shopper subjects were led to believe they were looking at a new shopping-by-television demonstration. Color slides of print ads were repeated from one to six times for three brands within several product classes. Although brand purchase intention was positively affected by repetition, there was no significant effect of repetition on brand evaluation. Moreover, when the relative rating of the brand in question was used to develop a score of preference to other brands in the product class, the effects of repetition coincided with a significant decrease in preference. With a different version of Sawyer's disguised laboratorY procedure and with television as well as print advertisements as stimuli, Heeler (1972) also failed to find any positive increases in brand evaluation due to repeated ad exposures.

DEMAND CHARACTERISTICS IN REPETITION RESEARCH

From the above review, it seems that positive effects of repetition on affect are more likely to result from experimental procedures in which the repeated exposures are presented with little or no accompanying "cover" or rationale for the subject and where no information other than the experimentally manipulated number of stimulus exposures is available to the subject. Such artificial settings may be antecedents of demand bias. Inconsistent past results might be explained by differences in the environment of the experiments and in the presence of demand characteristics which may cue a subject to respond positively to repeated stimuli.

One experimental paradigm that includes both of these demand-prone characteristics is the one used by Zajonc and his colleagues. An examination of the typical Zajonc experimental procedure from a subject's viewpoint reveals that his role in the experiment might be perceived as quite strange and artificial. The task of rating the foreign adjectives is a near impossible task. However, a subject realizes that the experimenter is performing the experiment for some purpose and expects the subject to rate the stimuli in some meaningful way. Thus, many subjects are likely to search for any available cues to differentiate the stimuli. Since the stimuli differ somewhat in appearance, some patterns of "goodness" may be discerned in these differences. However, the major difference in the stimuli is that, for some unexplained reason, the words were repeated several times but at different exposure frequencies. The fact that no warning or rationale for that repetition was offered by the experimenter adds to the artificiality of the experimental setting and may further focus the subject's attention to that variable. A subject who does not reject the task as too difficult or inane is likely to seize upon the widely manipulated number of exposures to differentiate his ratings on the stimuli. To the extent that the subject perceives that the experimenter expects some use of the exposure variable in the rating task and to the extent he believes the relationship should be positive, the subject may increase his ratings for stimuli that were exposed more often.

Such hypothesized demand cues might help to explain some of the inconsistent results between Zajonc's research and consumer research based on his work. Repeated consumer stimuli such as advertisements, actual products or brand names are presented in the context of the consumers' experiences and current perceptions and not in a vacuum where the number of exposures is the only available information. Also consumer research finding no positive effects of repetition on liking has often included deception or other explicit attempts to present the repeated stimuli in some natural or justified manner. Demand characteristics might account for the fact that brand evaluation or attitudes are often not affected by repeated ad exposures but measures of purchase intention are positively affected. Subtle demand bias might affect the simpler purchase intention response of simply checking a brand but might be not strong enough to significantly alter a subject's value structure of his perceptions of a brand or its characteristics.

There are four research modes that can examine the validity of such conjecture about demand characteristics. These include: l) the non-experiment, 2) the post-experimental inquiry, 3) manipulation of suspected demand cues, and 4) hetero-method replication. Investigations of each of these types have recently focused upon the Zajonc paradigm. Details of new research and summaries of published research are presented both to expand on the potential demand bias in repetition-affect research and to illustrate the use of these methods.

The Non-Experiment

A non-experiment involves a reenactment of all experimental procedures except the actual treatment which is instead only simulated (Orne, 1969). New subjects from the same population sampled in the experiment in question are asked to role-play the subjects in the actual experiment. They then undergo the same experimental procedure as the actual subjects. They are shown the same room and the employed equipment, read the same instructions, and, after the experimental treatment is described, asked to produce data just as if they had actually undergone the experimental treatment. A comparison of results from the non-experiment and the actual experiment can offer insights into the role of hypothesized demand characteristics. If results are similar, the cause-and-effect relationship cited in the experiment is open to question, and further investigation is warranted. Although similar results do not prove the presence of bias due to demand characteristics, such results do demonstrate the plausibility of the alternative demand characteristics explanation that cues in the experimental procedure--instead of or in addition to the cited independent treatment variable--helped to produce the observed experimental result.

This writer designed a "non-experiment" that simulated the Zajonc procedure but did not actually repetitively expose any stimuli to the subjects. The experimental hypothesis was that there would be a significant effect of "exposure" even though no actual exposure occurred. Seventy-two subjects were told that they were going to participate in an experiment where they would be read only a description of an actual experiment and then would be asked to behave as if they were subjects in an experiment like the one described. They were then read the following description:

A year ago, a psychological experiment about the learning of a foreign language was performed at the University of Michigan. Here is what happened: Subjects were seated at a table in front of a slide screen. After being told that they were participating in an experiment about the learning of a foreign language, they were asked to study a following series of slides. The room was darkened, and they were shown a series of 86 slides--each of which was exposed, one after another, for 2 seconds, or a total of almost 3 minutes. The slides were pictures of characters which looked like Chinese letters or words. During the slide presentation, it became apparent that many of the characters were being shown more than once. In fact, by the time the slides were over, some of the slides had been repeated as many as ten or twenty times.

We would like you to now sit and try to imagine sitting and watching these slides for almost 3 minutes--one slide every 2 seconds and several being repeated quite frequently.

After sitting for three minutes, the subjects were read Zajonc's explanation that the characters were Chinese symbols and that, as an estimate of their meaning, their rating of the "goodness" of the symbols was sufficient. In addition, they were given the following rating instructions:

Below are the characters which were shown in the series of slides. Since you did not actually see the slides, we have indicated, next to each word, the number of times that word was shown in the previous experiment. Note that the number of repeated exposures varied from O to 25. Please circle the point on the evaluation scale that represents your best guess of the degree of favorableness of the meaning of that foreign adjective word. (Note that "O" and "6" represent the two extremes of the scale and that "3" is the neutral scale point.)

The subjects then rated the twelve Chinese characters from the Zajonc experiment. After rating the characters, they were asked four questions about criteria used in the rating task and about the experimental hypothesis.

As in the Zajonc experiment being duplicated, the twelve characters used as stimuli were counterbalanced against six indicated frequencies (0, 1, 2, 5, 10, and 25), such that each character occurred equally often in each of the six frequencies, but for different subjects. Thus, for each subject, two characters were indicated as having been exposed for each frequency level, and each character appeared in each exposure condition to some subjects. In addition, to assess the importance of measurement "set," half of the subjects used scales marked "bad-good" and half used scales marked "good-bad." Six subjects per scale-character-frequency cell were used.

The experimental hypothesis was that subjects would increase the rated goodness of the characters as "frequency" increased from O to 25, and the results supported this hypothesis (see Figure l). The rated goodness increased from 2.56 on a seven-point scale to 3.50 (F=6.21, 5 and 720 df, p < .001). As in Zajonc's experiment, the ratings of eleven of the twelve characters were higher at 5, 10, and 25 exposures than at 0, l, and 2 exposures. The fact that subjects who did not actually view the repeated stimuli but who did experience the experimental setting and accompanying cues produced results quite similar to those of subjects who did actually view the repeated exposures lends some support to a hypothesis that demand cues in the experimental setting and procedure might have helped to produce Zajonc's observed results.

The Post-Experimental Inquiry

Upon completion of a subject's participation in an experiment, it should be standard procedure to interview the subject to determine his reaction. Such a practice takes advantage of the fact that human subjects can talk and reflect upon their experience. This inquiry may include a debriefing which explains the purpose of the research, the subject's contribution, and the reason for the procedure including deception if used. Concerning demand bias, however, the prime purpose of the post-experimental interview is to determine any suspicions the subject may have developed about the purpose of the experiment, the subject role he might have adopted, and why he behaved as he did.

In the above non-experiment, answers to post-experimental questions designed to reveal subjects' reactions to the experiment were analyzed. When asked to describe any guidelines or criteria used in rating the characters, fourteen or almost 2096 indicated that they used repetition to indicate goodness. The remainder indicated that they used only the appearance of the characters or used no criteria at all. An examination of the ratings of those who claimed to use repetition as a positive indicator revealed that these subjects accounted for a large portion of the overall increase with frequency. Table 1 shows the rated goodness of the repeated characters for the two groups. Those who claimed frequency as a criterion exhibited pronounced increases whereas the others showed only a slight increase. A post-hoc analysis of variance with frequency and subjects' cited rating criterion as factors indicated a significant effect of both frequency (F = 5.87, 5 and 852 df, p < .001) and its interaction with the cited criterion (F = 9.97, 5 and 852 df, p < .001).

FIGURE 1

COMPARISON OF FREQUENCY EFFECTS IN ZAJONC AND "NON-EXPERIMENT"

TABLE 1

REPETITION-AFFECT RESULTS OF SUBJECTS CLAIMING TO USE FREQUENCY AS AN INDICATOR OF GOODNESS AND OF THOSE WHO DID NOT

In answer to a question which asked the subjects to state what they believed to be the experimenter's hypothesis, eleven or 15% stated that more frequent exposures were expected to increase goodness. Seven others stated an opposite hypothesis; repetition would result in less goodness. The remainder of the subjects either paraphrased the opening instructions or stated some psychological hypothesis about the appearance of the characters. Table 2 shows how these subjects reacted to the indicated frequency of the rated characters. The first group demonstrated a large increase with repetition, and the second showed a decrease. The latter group, which did not mention frequency as part of the experimental hypothesis, showed an increase with greater frequency and nearly duplicated the reactions of all subjects averaged together. Both frequency (F = 5.61, 5 and 846 df, p < .001) and its interaction with subjects' estimates of the experimenter's hypothesis (F = 1.80, 10 and 846 df, p < .06) were significant. Thus, it appeared that subjects' reactions to increased frequency coincided with their estimates of what was expected of them by the experimenter - a result consistent with a demand characteristics explanation of Zajonc's positive relationship between repetition and effect.

TABLE 2

REPETITION-AFFECT RESULTS OF SUBJECTS WITH DIFFERENT ESTIMATES OF EXPERIMENTER'S HYPOTHESIS

Similar results of other post-experimental inquiries have been reported. Stang (1974) found very close relationships between subjects' intuition about the experimenter's hypothesis and their affective ratings of repeated Turkish words. Burgess and Sales (1971) found that subjects who reported feeling relatively positive about the experiment and their participation exhibited a significant positive relationship between the number of exposures and liking, whereas subjects who felt relatively negative about the experimental context exhibited no effect of exposure.

Manipulation of Demand Characteristics

When demand characteristics are suspected, experimental variation of the hypothesized cues can help to explore their effects. If the effects of the hypothesized demand cues are monotonic, then increases (or decreases) in the suspected aspects of the experimental environment should increase (or decrease) their effects.

For example, Suedfeld et al. (1971) manipulated the direction of measurement "set" as either positive or negative ("tell me the extent that this symbol means something good [bad]"). They found that a good set coincided with positive monotonic effects of repetition on attitude whereas a negative set led to an inverted U-curve relationship. Burgess and Sales (1971) hypothesized that classical conditioning between positive evaluation of participation in the experiment and high frequency stimuli might account for Zajonc's positive results. In an experiment that manipulated frequency of nonsense words and the positiveness of paired-associated words, the rated liking of the nonsense words at different frequency levels varied with the context of the associated words. As hypothesized, a negative relationship was found for the negative context conditions, and a positive relationship was found for the POSitive conditions.

The relative effects of other potential demand cues in repetition experiments could be accomplished by manipulating the suspected procedural cues contrary to the direction of the claimed causal variable of repeated stimulus exposure. For example, subject's perceived personal evaluation from responding positively, negatively, or neutrally to repetition could be experimentally varied (e.g., Rosnow et al., 1973). An alternative method would be to first expose subjects to the various levels of exposure and then subsequently mislead them by stating that low frequency stimuli were exposed at high levels and vice versa. A negative repetition-attitude result would indicate a greater influence of demand characteristics than the actual amount of exposure. This latter method would require that the results of the actual repetition conditions remain at the time of the deception and that subjects would forget the real repetition levels and believe the experimenter's false repetition "reminder "

Heteromethod Replication

Multiple operationalism of independent variables and accompanying procedures can help to refute questions about confounding demand cues. If more than one procedure yielded similar results, the result (and accompanying theoretical foundation) would obviously carry more confidence.

Repetition experiments with different procedures or designs have both refuted and supported a demand characteristics explanation. Saegert, Swap, and Zajonc (1973) studied the effects of repeated personal encounters on interpersonal attractions and manipulated the pleasantness of the experimental context by presenting pleasant and noxious odors or liquids to smell or taste. Contrary to the results of Burgess and Sales, positive effects oz frequency were found in both pleasant and unpleasant contexts. In addition, Zajonc and Rajecki (1969) tested the effects on liking of nonsense words when exposure was experimentally varied from one to twenty-five by means of newspaper inserts over a month's time. Randomly selected readers of the newspaper were interviewed and a positive relationship between word exposure and rated goodness of meaning was found. Although the field results are impressive, field replication does not preclude the possibility of demand bias (e.g., Rosen, 1970). A post-experimental interview probing why the more heavily repeated words were rated more positive would have been very helPful.

Moreover, some of Zajonc's own lab experiments with variations in the number of exposures and experimental design have offered additional support for an alternative demand bias explanation. Zajonc, Swap, Harrison, and Roberts (1971) reported that the absolute number of exposures appeared to be much less important than the relative number. The results of an-experiment which tested 0, 1, 3, and 9 exposures did not differ significantly from those of experiments with exposure levels of 0, 1, 3, 9, and 27 and 0, 3, 9, 27, and 81. Such results would be expected if past effects of repetition were actually due to demand bias whereby repetition was perceived as merely a cue to differentiate ratings as expected by the experimenter. Furthermore, no significant effect of exposure of initially unfamiliar characters was found in a fourth experiment in which the manipulated exposure treatment was across-subjects rather than the typical within-subjects design. That is, no effect of frequency was obtained when subjects were not able to compare different levels of exposure--a design that could not have artificially highlighted repetition as a cue to discriminate ratings of the tested stimuli.

CONCLUSION

One objective of this paper was to present research evidence of demand bias in a particular repetition-affect experimental paradigm. It seems safe to conclude that the effects of repeated exposures are highly sensitive to small changes in the experimental procedures and environment. However, the case for demand characteristics as a cause of a positive monotone repetition affect relationship instead of the cited independent variable of the number of exposures has not been conclusively proven. Supporting evidence comes from the facts that a minority of subjects who use a positive connotation of repetition as a response criterion account for a disproportionately high amount of the observed positive relationship, subjects' reactions to repetition coincide highly with their estimates of the experimenter's hypothesis, manipulations decreasing either the positiveness of the experimental context or of items associated with the repeated stimuli can eliminate or even reverse the positive repetition-affect relationship, and positive repetition-affect results depend on the subjects' being able to compare relative exposure levels instead of absolute levels.

However, if subjects do interpret and react to their estimates of the experimenter's hypothesis, none of the demand bias research has provided an adequate explanation of why subjects tend to perceive a hypothesized repetition - good relationship instead of a negative one. Some conjecture might involve the tendency to adopt a "good" subject role (Orne, 1969) or the overall positiveness associated with science and one's participation in an experiment (Burgess and Sales, 1971). The fact that experiments with advertisements as repeated stimuli are more apt than subjects in more sterile experimental contexts to yield neutral or negative effects of repetition might be due to subjects who negatively evaluate the social and aesthetic values of advertising (Bauer and Greyser, 1968) and who are more apt to adopt a "negative" or "apprehensive" subject role (Weber and Cook, 1972) or to feel negative about their participation in the experiment. Currently, however, there is no evidence to support such hypothesized subject negativism.

The second goal of this paper was to illustrate the use of four research methods to evaluate the possible presence of demand characteristics in a given understanding of the basic process under study as well as improved experimental methodology. Other than repetition, many areas in consumer research would benefit from similar demand characteristic investigations. These areas might include quality connotations of price, conformity in consumer choice behavior, bargaining behavior between buyers and sellers, and post-purchase dissonance (Sawyer, 1974). More careful experimental designs and procedures could help to reduce demand bias. More consumer research should concentrate on hetero-method replications and manipulations of suspected confounding demand characteristics, and the use of non-experiments as pretests for planned experiments and the inclusion of post-experimental inquiries which probe for demand bias should be standard experimental procedures.

REFERENCES

Bauer, R.A. and Greyser, S.A. Advertising in America: the consumer view, Boston: Division of Research, Graduate School of Business Administration, Harvard University. 1968.

Becknell, J.C., Wilson, W.R., and Baird, J.C. The effect of frequency of presentation on the choice of nonsense syllables. Journal of Psychology, 1963, 56, 165-170.

Burgess, T.D.G., and Sales, S.M. Attitudinal effects of "mere exposure": A reevaluation. Journal of Experimental Social Psychology, 1971, 7, 461-472.

Heeler, R.M. A laboratory investigation of inter-related effects of mixed media, multiple copy, and multiple insertions in advertising campaigns. Unpublished doctoral dissertation, Stanford University, 1972.

Light, M.L. An experimental study of the effects of repeated persuasive communications upon awareness and attitudes. Unpublished doctoral dissertation The Ohio State University, 1967.

Miller, S.J., Mazis, M.B., and Wright, P.L. The influence of brand ambiguity on brand attitude development. Journal of Marketing Research, 1971, 8, 445-449.

Olson, J.C. Cue properties of price. Working Paper No. 20, Pennsylvania State University, 1974.

Orne, M.T. Demand characteristics and the concept of quasi-controls. In R. Rosenthal and R. L. Rosnow (Eds.), Artifact in behavioral research. New York: Academic Press, 1969, 143-179.

Ray, M.L. and Sawyer, A.G. A laboratory technique for estimating the repetition function for advertising media models. Journal of Marketing Research, 1971, 8, 20-29.

Rosen, N.A. Demand characteristics in a field experiment. Journal of Applied Psychology, 1970, 54, 163-168.

Rosenthal, R. and Rosnow, R.L. (Eds.), Artifact in behavioral research. New York: Academic Press, 1969.

Rosnow, R.L. and Aiken, L.S. Mediation of artifacts in behavioral research. Journal of Experimental Social Psychology, 1973, 9, 181-201.

Rosnow, R.L., Goodstadt, B.E., Suls, J.M., and Gitter, A.G. More on the social psychology of the experiment: When compliance turns to self-defense. Journal of Personality and Social Psychology, 1973, 27, 337-343.

Saegert, S., Swap, W., and Zajonc, R.B. Exposure context and interpersonal attraction. Journal of Personality and Social Psychology, 1973, 25, 234-242.

Sawyer, A.G. Demand characteristics in laboratory experiments in consumer research. Working Paper No. 74-9, University of Massachusetts, 1974.

Sawyer, A.G. The effects of repetition: Conclusions and suggestions about experimental laboratory research. In G.D. Hughes, M.L. Ray, and G. Haines (Eds.), Consumer information processing. Chapel Hill: University of North Carolina Press, in press.

Stang, D.J. The effects of "mere exposure" on learning and affect. Journal of Personality and Social Psychology, 1974.

Suedfeld, P., Epstein, Y.M., Buchanan, E., and Landon, P.B. Effects of set on the "effects of mere exposure." Journal of Personality and Social Psychology, 1971, 17, 121-123.

Weber, S.J. and Cook, T.P. Subject effects in laboratory research: An examination of subject roles, demand characteristics, and valid inference. Psychological Bulletin, 1972, 77, 273-295.

Zajonc, R.B. Attitudinal effects of mere exposure. Journal of Personality and Social Psychology, Monograph Supplement, 1968, 9, 2, Part 2, 1-27.

Zajonc, R.B., and Rajecki, D.W. Exposure and affect: A field experiment. Psychonomic Science, 1969, 17, 216-217.

Zajonc, R.B., Swap, W.C., Harrison, A.A., and Roberts, P. Limiting conditions of the exposure effect: Satiation and relativity. Journal of Personality and Social Psychology, 1971, 18, 384-391.

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