The Use of Real Versus Artificial Stimuli in Research on Visual Esthetic Judgments

THE USE OF REAL VERSUS ARTIFICIAL STIMULI IN RESEARCH ON VISUAL ESTHETIC JUDGMENTS

Joel Huber, Duke University

Morris B. Holbrook, Columbia University

[The authors gratefully acknowledge a research grant from the Olympia Brewing Company and the support of Columbia University's Faculty Research Fund.]

ABSTRACT -

The correspondence of experimental results concerning real and artificial esthetic objects is a question that has prompted insufficient empirical attention. The present study investigates the comparability of findings concerning esthetic judgments of 16 realistic architectural photographs and 16 artificially-constructed visual images. It appears (1) that different techniques are most suitable for modeling the two types of esthetic responses, (2) that preference patterns are fairly stable within individuals over time, but (3) that little convergent validity exists in patterns of preference between the two types of stimulus displays. In sum, the results seriously question the degree to which one may legitimately make inferences from one type of visual esthetic stimulus to the other.

INTRODUCTION

In making visual esthetic judgments, many aspects of an image may come into play. Besides belonging to different physical regions of the stimulus field, these visual esthetic factors can also be defined at different levels of abstraction--from the sense data themselves, such as color or form, to the more rarified symbolic or allegorical meanings. In an attempt to control the resulting variability in possible interpretations of esthetic objects, many researchers (e.g., Benjafield 1976; Berlyne 1974; turned to the

Munsinger and Kessen 1964; Nicki 1976) have study of artificially constructed stimuli. Such studies have related the physical features underlying the construction of a stimulus set to perceived complexity and generally have found a curvilinear relationship between complexity and esthetic preference (Berlyne 1971, 1974). in short, people appear to have an ideal level of complexity and to devalue items that are either too complex or not complex enough. Of course, the ideal level of complexity depends on the experience and mental capabilities of the individual as well as on the context in which the image is judged (Berlyne 1960, 1971).

Practitioners involved with designing or evaluating esthetic objects in applied settings face a very different problem. Given a set of possible objects rich in symbolic and sensory value, they must determine which designs are best-liked and apt to exert the maximum market appeal. in such a context, where external generalization is of paramount importance, the research on artificial stimuli is suspect due to the fact that the complexity and familiarity variables differ so dramatically between the laboratorv and real-world environments of art appreciation.y

The present study explores the relevance of esthetic judgments on artificial visual stimuli to those on real artistic objects. To preview briefly, subjects evaluated both kinds of stimuli at each of two successive time periods. The real objects were color slides of 20th-Century buildings. The artificial images were two-dimensional abstract line drawings constructed to contain some of the attributes of the architectural pictures--shading, complexity, angularity, proportion. Given evaluative judgments of the architectural and artificial stimuli, three principal research questions arose. First, what methodology is most suitable for modeling both perceptions and preferences in these two types of data? Second, what is the effect of the time between replications on esthetic judgments of real and artificial images? Finally, and most important, how well do subjects' judgments on artificial stimuli correspond to those on real objects? That is, from judgments on one set, can one safely make inferences concerning the likely pattern of responses to the other?

EXPERIMENTAL PROCEDURE

Sample

Fifteen subjects were recruited from an advanced undergraduate art seminar and were paid $5.00 each for completing the four tasks described below. All students had taken an art survey course, and some had attended other studio and art history courses. It may therefore be inferred that their level of involvement in the evaluation tasks was relatively high.

Tasks

In the first task, subjects were shown 16 color slides of buildings projected onto a screen over an area of approximately ten square feet. All 16 slides were first exposed briefly to acquaint subjects with the stimulus set. Then each slide was shown separately while subjects filled out the 13-scale rating form that appears in Exhibit 1.

EXHIBIT 1

BIPOLAR ADJECTIVES USED TO RATE STIMULI

The first slide was displayed for a period of 120 seconds and succeeding slides for 90 seconds each. This time period allowed ample time to complete the form but did not encourage subjects to ratiocinate or to ponder deeply about the visual images.

Upon completing the first task, subjects were given a packet of 16 abstract drawings, each to be rated using the same 13 scales. These ratings were returned at the next class meeting two days later. Thereupon, and in subsequent classes, a series of lectures on architectural styles was presented. These involved about 50 slides, 6 of which were taken from the 16 architectural pictures in the original task. Finally, in a class meeting conducted ten days after the initial ratings, both tasks were replicated.

In sum, the experiment obtained evaluative judgments of real and artificial visual images, both before and after a learning experience that related directly to the esthetic content of one of the two stimulus sets.

ANALYSIS OF THE ARCHITECTURAL PICTURES

The architectural pictures were all color slides of the exteriors of 20th-Century buildings. The 16 photos used are shown in Exhibit 2. These had been culled from a larger set of 22 pretested items. The pictures dropped were those on which perceptions, as measured in the pretest, appeared to be inconsistent. This inconsistency often stemmed from ambiguity in the photo, for example, where a building might be simple but its surroundings complex so that subjects did not agree as to its level of complexity. Items that appeared to be strong outliers were also eliminated from the set to insure that no stimulus would be too dissimilar from others on any one dimension. This precaution tended to force trade-offs among attributes and to prevent a few stimuli from dominating esthetic judgments.

While stimuli that prompted extremely heterogeneous perceptions were dropped, those that varied with respect to affect or preference were retained. Affect was defined a priori as the sum of an item's ratings on Beautiful + -Enjoyable + Pleasing (with scales scared in the appropriate direction). We expected that subjects would differ in their preferences thus measured. Indeed, one purpose of modeling evaluative judgments was to account for such affective heterogeneity. By contrast, after the weeding-out process described above, we anticipated that individuals would agree on their perceptions but that individual preference vectors (or ideal points) would indicate a pattern of heterogeneous preferences (e.g., see Holbrook and Huber 1979; Huber and Holbrook 1976). The following section describes such an aggregate perceptual space while the next discusses the projection of individual vectors to represent patterns of preference.

Representation of Perceptions

Discriminant analysis, with photos as the dependent category variables and the 13 ratings as the independent variables, generated the two-dimensional perceptual space shown in Exhibit 2 (F. Johnson 1971). The dimensions of such a space are linear combinations of the original 13 rating scales. These are optimal in the sense of being ~he best dimensions to use to predict the identity of each stimulus. This approach possesses an advantage over a principal components analysis in that its dimensional structure tends to be oriented toward attributes about which subjects are most homogeneous in their ratings, thus emphasizing those attributes about which there is the most consensus (Huber and Holbrook 1979a).

EXHIBIT 2

ARCHITECTURAL STIMULI IN DISCRIMINANT SPACE

The space clustered stimuli in a pattern that appears quite plausible. The complex and more unbalanced buildings in the upper right quadrant contrast sharply with the geometric monolithic forms in the lower left. Subjectively, this suggests an interpretation of the two axes as representing balance and complexity respectively.

The meaning of the dimensions can be determined more rigorously, however, by projecting the original rating scales into this space, as shown in Exhibit 3. These projections were derived for each scale by regressing the mean ratings of items against their coordinates on the two dimensions of the discriminant space (Carroll 1972). Items that project farthest out on the resulting vector have the highest predicted scores on the corresponding attribute.

As a graphic indicator of how well the space accounts for each attribute, the length of each vector is proportional to the percentage of variance accounted for by the multiple regression (shown parenthetically). Only those vectors with fits of R2:,.30 are included in the diagram.

The first axis is strongly correlated with disorderliness and lack of balance while the second is strongly associated with activity, vibrancy, and dynamism. Interestingly, complexity appears to be the resultant of these two sets of perceptual vectors. Hence, the mathematically derived vector orientations support one's intuitions concerning the meaning of the perceptual space while providing more detail on the relations among balance, activity, and complexity.

The space shown in Exhibit 2 was derived from both waves of data pooled together. Separate discriminant spaces had been run for each wave separately, but the results were quite compatible, both in terms of the positioning of the items in the spaces and the orientations of the attribute vectors. Thus, it appears that the intervening lectures exerted little effect on perceptions of the architectural photographs. This finding differs from the demonstrable effect of the lectures on architectural preferences, to which the discussion now

Preferences for Architectural Pictures

When the aforementioned index of affect was first averaged across subjects and then represented by a vector projected into the space shown in Exhibits 2 and 3, its fit was quite weak (R2=.2). This result is typical of attempts to project aggregate preferences into a discriminant space. The reason is that discriminant spaces are oriented to those attributes on which different judges agree in their ratings so that affect is included only if subjects share the same likes and dislikes. Where individuals disagree on preferences, each must be represented by a separate affective vector. Such individual affect vectors, for both the first and second time periods, appear in Exhibit 4. Again, vector lengths are proportional to explained variance. For simplicity, only those subjects with R2>.50 for both waves of data are included. However, the vector orientations for these eight subjects are typical of those not shown.

EXHIBIT 3

PROJECTION OF BIPOLAR RATINGS INTO DISCRIMINANT SPACE

Several conclusions emerge from Exhibit 4. First, preferences ranged widely--particularly with respect to the vertical activity dimension. Second, there is a rather remarkable stability in orientation between the first and second waves for any given individual, as indicated by the generally small shifts between paired single- and double-headed arrows. Third, the index of fit tended to increase in the second wave, as shown by the greater lengths of the double-headed arrows. Specifically. across all 15 subjects the mean-squared correlations was .58 for the second wave as opposed to .48 for the first, a difference significant at the .05 level. This shift may be due to what Wyer (1974) called a "socratic effect," a tendency to become more consistent with practice in forming evaluative judgments. Fourth, subjects' evaluations converged somewhat over time. Although considerable affective variation across respondents remained in the second wave, there was a general movement toward preferring more orderliness and balance. This trend may reflect an effect of the lecture in promoting an appreciation of the esthetic virtues of symmetry, balance, and order in architectural design.

EXHIBIT 4

PROJECTION OF INDIVIDUAL AFFECT INDICES INTO DISCRIMINANT SPACE

In further reference to the lecture, the six slides used in the class discussion (flagged by asterisks in Exhibit 2) increased in affect between the two waves, with a mean change of +1.0 as compared with -0.1 for the remaining 10 slides. This difference was significant at the P<.05 level. It appears, therefore, that the art-appreciation lecture did increase liking for the items mentioned and generally encouraged homogeneity of preferences across subjects and consistency of preferences within individuals.

ANALYSIS OF THE ARTIFICIAL IMAGES

The 16 artificial images shown in Exhibit 5 were drawn by creating all possible combinations of four design features. On the first design dimension (double/golden), quadrilaterals were drawn such that their sides (a and b) were proportioned either as a golden rectangle (a/b b/a+b) or as a double rectangle (a/b = b/2a). This contrast appears in Exhibit 5 in the difference between rows I and 3 (golden) versus 2 and 4 (double). The potential importance of this design feature has been suggested by considerable psychological research indicating that the golden rectangle is inherently pleasing (Benjafield 1978; Berlyne 1971). The present study provided a test of this proposition in a more complex setting.

The second design feature varied the number of components composing an image from few (6) to many (10). The third feature involved shading three of the component areas or ieaving them blank. The fourth introduced irregularity into the abstract images by changing the perpendicular lines around the edges to oblique angles.

These last three design features were viewed as alternative manipulations of complexity and were expected to be related to the relevant perceptual judgments. However, the complexity manipulations were chosen less for their role in hypothesis testing than for their potential capacity to tap the same sorts of evaluative processes engaged by the architectural displays. In other words, the artificial stimuli were viewed as abstractions of the richer architectural photographs. This viewpoint encouraged an examination of the degree of correspondence between the perceptual and affective processes in both sets of data.

Representation of Perceptions

The artificial images were placed into a perceptual space constructed by multiple discriminant analysis in a manner analogous to that used with the architectural slides. This discriminant space contained three dimensions that corresponded closely to the three complexity-manipulating features: few/many, clear/shaded, and rectangular/angled. Indeed, as measured by the aforementioned vector-fitting procedure, the three discriminant functions accounted for over 90% of the variance in each of the three dichotomous stimulus features. By contrast, the double/golden feature showed no consistent orientation in the perceptual discriminant space, perhaps because the differences in side-length proportions of the quadrilaterals were too subtle or interactive to detect reliably.

The strong association between the physical and discriminant dimensions implies a high degree of isomorphism based on perceptual veridicality. Indeed, this degree of correspondence is quite remarkable given that the perceptual ratings were collected without telling subjects anything about the physical rules underlying the construction of the images. This result demonstrates the power of discriminant analysis to uncover latent dimensions in a data set. In the present situation, however, this power carries the almost ironic corollary that the perceptual space yields little additional information beyond that already available from the physical features themselves. Since design decisions are made more easily on the basis of such physical dimensions rather than from positions in a perceptual space, it follows that for the stimuli used here a consideration of physical features should be sufficient to predict affective responses as a basis for pragmatic applications to esthetic design. Accordingly, the rest of this analysis considers the relationship between the manipulated features and preferences rather than developing the kind of two-stage model that uses perceptions as intervening variables between physical attributes and affect (Holbrook 1980; Huber 1975). The very high correspondence between the physical and perceptual dimensions indicates that very little predictive (as opposed to explanatory) information is lost by such a simplification.

EXHIBIT 5

16 ARTIFICIAL STIMULI USED IN STUDY

Preferences for Artificial Images

The numbers beneath the images in Exhibit 5 indicate the rank orders of the average affect scores for the first/second waves. Generally, the simpler images were better liked though exceptions abound. To attain a more detailed analysis of the effects of stimulus features on preferences, dummy-variable regressions were used to perform metric conjoint analyses on each subject (Green and Srinivasan 1978). One series of regression runs obtained simple additive part-worth values for the physical features coded as four zero-one independent variables. Another set of regressions included six multiplicative terms to estimate feature Y feature interactions. Such cue configurality might occur if the affective impact of one feature depended upon the level of another. Finally, regressions were run at the aggregate level (i.e., over all subjects) on both waves of data.

Typical of studies on cue configurality (for a review, see Holbrook and Moore 1980), the analysis of feature interactions produced rather mixed results. Among 180 possible interactions at the individual level, 22 reached statistical significance at the .05 level. Across waves, however, none of the feature interactions significant for a subject in the first wave reappeared in the second. This finding raises serious questions concerning the reliability of the interaction terms. Furthermore, in the aggregate analysis, none of the interactions reached significance in either wave. Thus, the feature X feature interactions proved inconsistent both within-subjects-across-waves and across-subjects-within-waves . Given this instability of the interaction terms, we present results only for the main effects of features.

Exhibit 6 shows part-worth coefficients for each subject on each physical dimension in each wave. Summary statistics give the mean coefficient, its standard deviation across subjects, and the average percentage of variance in affect accounted for by each feature. Although rather wide variation occurs across subjects, certain generalizations apply. Notice, first, that an initial mean dislike for shading diminishes in the second wave (-.21 vs. -.06). Similarly, the average coefficient for angle begins at a negative level but grows less unfavorable over time (-.26 vs. -.12). Since the shading and angularity features account for the majority of the individual variance (44%), these results suggest an initial aversion to complexity moderated by increased familiarity over time. By contrast, the average coefficients of few/many are too small to reach significance in either wave of data.

EXHIBIT 6

COEFFICIENTS OF THE PHYSICAL DIMENSIONS PREDICTING AFFECT FOR ARTIFICIAL STIMULI

The coefficients for abstract stimuli in Exhibit 6 are analogous to the positions of preference vectors for architectural photos in Exhibit 4. In both cases, individual preferences are summarized by the marginal values of various dimensions. In the architectural case, these dimensions are represented by the axes of a perceptual space derived from a set of subjective ratings. For the artificial images, on the other hand, the dimensions of interest are defined by the physical features themselves. While the results for architectural and artificial stimuli are not strictly comparable because of the differences in these underlying dimensions, it appears in general that changes in preference for two types of stimuli were in opposite directions. For architectural designs, subjects came to desire more balance while, for abstract images, they acquired a higher degree of favorability toward features associated with complexity. This divergence of results is examined in the next section, which considers the relation of preferences to indices defined the same way across data sets.

DIFFERENCES IN RESPONSES ACROSS STIMULUS TYPES

For both the architectural and artificial stimuli, ratings were collected on the same 13 scales shown in Exhibit 1. These scales had been chosen to reflect four underlying factors repeatedly revealed in other studies (Berlyne 1971, 1974; Osgood, Suci, and Tannenbaum 1957) and supported by pretests of the present questionnaire. Indices were formed by simply summing items associated with each factor. Specifically: (1) Affect = Beauty + Pleasing + Enjoyable; (2) Activity = Active + Dynamic + Vibrant; (3) Potency = Loud + Hard + Strong; and (4) Complexity = Complex + Disorderly + Unbalanced. The relationships among these indices were directly comparable across data sets since the measures were defined equivalently in both cases. We first consider the consistency of the components of affect, then examine the stability of responses to complexity across waves and data sets, and finally conclude with the stability of overall attitude structure over time and across stimulus types.

Consistency of the Affect Index

One measure of the consistency of evaluative judgments is their correspondence across the variables that compose an affective index. In this case, the average part-whole correlation of affect with its components was used. This assesses the relative degree to which subjects who rated a stimulus high on one component of affect (e.g., "beautiful") also tended to rate it high on others (e.g. , 11 pleasing"). This measure was lower for the artificial images (r = .83) than for the architectural photos (r = .94), a difference significant at the .05 level.

The lower consistency for the artificial images may be explained in part by the fact that less time was spent making these judgments and by a lower interest expressed by the subjects in this task.

Stability of Responses to Complexity

The second issue examined with the indices concerns the affective impact of perceived complexity over time and across data types. This evaluative orientation was measured by the average correlation of the complexity index with the index for affect. The results appear in Exhibit 7 and indicate that, for both sets of data, affect is negatively related to complexity, but that the degree of this relationship increases for the architectural pictures and decreases for the artificial images. In other words, preferences change in opposite directions for the two types of stimuli.

The general negative response to complexity can be accounted for by assuming that complex stimuli of both types lie beyond the optimal point on the Wundt Curve shown in Exhibit 8. According to Berlyne (1971), if a set of stimuli is so complex as to generate arousal beyond some optimal point, then the affective response to further increases in complexity will be negative. Both the architectural photos and the artificial images appear to be more complex than either the simple architectural drawings used by Breuer and Lindauer (1976) or the monochromatic polygons studied by Munsinger and Kessen (1964). Given this complexity, the results found here do not conflict with previous findings of positive associations between perceived complexity and affect, but may simply reflect the implications of different starting points on the Wundt Curve. Specifically, in Exhibit 8, if the artificial images began at point B on the curve and the more complex architectural photos began at point C and if the effect of increased familiarity between waves were reflected in a small movement of both sets to the left, then the changes in the marginal value of complexity shown in Exhibit 7 would be explained. That is, the architectural images would be moved to a point where the marginal value of complexity has a stronger negative slope while the artificial drawings would be moved to a point where that slope is flatter. Needless to say, such a post hoc explanation requires further testing before it can be accepted with any degree of confidence.

EXHIBIT 7

EVALUATION OF COMPLEXITY ACROSS DATA SETS AND TIME

EXHIBIT 8

HYPOTHESIZED RELATIONSHIP BETWEEN COMPLEXITY AND AFFECT GIVEN BY WUNDT CURVE

Stability of Attitude Structure

A final issue concerns the degree of consistency in the relationships of the three perceptual indices (activity, potency, and complexity) to affect. For each Subject a correlation was computed between the index of affect and each of the perceptual indices. These correlations measure the "affective orientation" of each individual. Two relevant consistency measures were computed to measure the stability of these affective orientations. The first estimated temporal stability across waves in a way that is analogous to test-retest reliability. The second computed stability across tasks to estimate stability of affective orientation between stimulus types in a manner analogous to convergent validity.

Temporal consistency requires that respondents who display a high or low orientation to some perceptual dimension in the first wave continue to show that orientation in the second. Accordingly, temporal consistency was gauged by the correlation across subjects between the affective orientations in the two waves. As shown in Exhibit 9, temporal consistency was quite high for both data sets and for all three perceptual indices indicating considerable test-retest reliability in the measures of affective orientation.

EXHIBIT 9

CONSISTENCY OF AFFECTIVE ORIENTATION ACROSS TIME AND TYPES OF STIMULI

The second measure of consistency is the degree to which affective orientation remains constant across stimulus sets. Cross-stimulus consistency was estimated by the correlation of affective orientations across subjects between the architectural and artificial stimuli. This measure would be high, for example, if respondents who liked complexity in architecture also liked it in the artificial images. Such convergent validity is quite important if one wishes to make legitimate inferences from one type of data to the other. Therefore, this issue bears directly on the question of whether people's responses to artificial stimuli tell us anything about their reactions to real objects. As shown in the right-hand portion of Exhibit 9, the cross-stimulus consistencies were positive but, with the exception of the activity index, did not reach statistical significance. The next section discusses the implications of these results.

SUMMARY AND CONCLUSIONS

This study employed two stimulus sets (architectural and artificial), each evaluated at two time periods (before and after intervening lectures). The principal research questions concerned the appropriate way to analyze evaluative judgments of objects in the sets; the effect of time and the intervening lectures on such judgments; and, finally, the degree of correspondence among affective reactions to the two types of stimulus. Conclusions concerning each of these issues may now be summarized in turn.

In terms of the appropriate analytic method, the perceptual space generated by a discriminant analysis of the architectural ratings did quite a good job of providing a meaningful representation of architectural perceptions on dimensions that might not have been clear a priori. By contrast, the corresponding perceptual space based on the ratings of artificial images did little but duplicate the physical features used to construct the stimuli. Accordingly, individual preferences for the architectural photos were represented by vectors in the perceptual space while those for the artificial images were modeled by part-worth coefficients in conjoint analyses conducted via dummy-variable regressions. This latter method had the additional advantage of estimating the feature X feature interaction terms. While these were not stable across either respondents or time periods in the present case, they have been shown to be important in other contexts (Holbrook and Moore 1980). Similar relativity holds for the decision to use the physical rather than the perceived dimensions in modeling evaluative judgments. There may be situations where lack of redundancy between perceived and physical attributes requires a two-stage processing model. Various research has shown that the perceptual dimensions of a product set may (Holbrook 1980) or may not (Huber 1975) enhance one's ability to predict or understand consumer preferences.

The second issue posed a question concerning the effect of the intervening ten days and the architecture lectures on perceptions and preferences. Generally, perceptions remained quite stable, as evidenced by the high correspondence between the perceptual spaces from the two waves of data. The individual-specific structures of affect were also quite stable across waves--whether measured as the directions of the individual preference vectors in the architectural space (Exhibit 4), the part-worth coefficients of the physical features of the artificial images (Exhibit 6), or the affective orientations toward the various perceptual indices (Exhibit 9). In all cases, differences in an individual's preferences over time were much smaller than differences across individuals. The one systematic change in affect that did occur concerned the five architectural slides discussed in the intervening lectures. These increased in value, indicating that esthetic judgments may provide one more area in which to study the persuasive effects of communication.

Finally, questions about the stability of affective responses across types of stimuli were considered. In this light, the most salient issue concerns the degree to which one may safely make inferences from artificial visual stimuli to more realistic esthetic objects. Here the results indicate clearly that such inferences might be threatened by rather weak convergent validity. First, the part-whole correlational index of consistency in evaluative judgments suggests that subjects may be more careful with realistic pictures than with artificial drawings. This result may be due to the combination of a greater interest in the former together with a greater stability of preferences toward more familiar kinds of objects. Second, the change in preferences in the real objects was away from complexity while, for the simpler artificial images, complexity became more acceptable. While such a result might be explicable by reference to the Wundt Curve (Exhibit 8) it offers little solace to those who might study simple artificial stimuli in search of guidance on designing real esthetic objects. Third, the summary measures of cross-stimulus consistency in affective orientation showed that neither complexity nor potency judgments possessed significant degrees of convergent validity. This facet of the findings again signals the danger inherent in generalizing from artificial to real esthetic judgments.

In interpreting the above results, a cautionary note is required. In both tasks, subjects compared large numbers of relatively novel stimuli. The effect of this unfamiliarity may have been to raise the general level of arousal to a point at which complexity became aversive. Thus, complex items that were disliked in the present setting might have been liked better in a less overloaded context. Thus, just as responses to simple stimuli might not represent reactions to more complex objects, so evaluations in a more complex setting might not correspond to those in a simpler situation. For practitioners there is no current answer to this problem, save the rather simple-minded advice that the evaluation task in the laboratory should parallel that in the real world as much as possible. For the researcher, however, this conjecture may suggest a potentially rich field of esthetic research.

REFERENCES

Fred Attneave, "Physical Determinants of the Judged Complexity of Shapes," Journal of Experimental Psychology, 53 (April 1957), 221-7.

John Benjafield, "The Golden Rectangle: Some New Data," American Journal of Psychology, 89 (1976), 737-43.

D.E. Berlyne, Conflict, Arousal, and Curiosity, New York: McGraw-Hill, 1960.

D.E. Berlyne, Aesthetics and Psychobiology, New York: Appleton-Century-Crofts, 1971.

D.E. Berlyne, Studies in the New Experimental Aesthetics, New York: Wiley & Sons, 1974.

Leslie F. Breuer and Martin S. Lindaur, "Judgments of Historical Architectural Styles," Perceptual and Motor Skills, 42 (1976), 1181-2.

J. Douglas Carroll, "Individual Differences and Multidimensional Scaling," in Multidimensional Scaling: Theory and Applications in the Behavioral Sciences, ed. Roger N. Shepard, A. Kimball Romney, and Sara B. Nerlove, New York: Seminar Press, 1972.

Paul E. Green and V. Srinivasan, "Conjoint Analysis in Consumer Behavior: Issues and outlook," Journal of Consumer Research, 5 (September 1978), 103-123.

Morris B. Holbrook, "Integrating Compositional and Decompositional Analyses to Represent the Intervening Role of Perceptions in Evaluative Judgments," Working Paper, Columbia University, 1980.

Morris B. Holbrook and Joel Huber, "Separating Perceptual Dimensions from Affective Overtones: An Application to Consumer Aesthetics," journal of Consumer Research, 5 March 1979), 272-83.

Morris B. Holbrook and William L. Moore, "Feature interactions and Consumer Judgments of Esthetic Products in Verbal Versus Pictorial Presentations," Working Paper, Columbia University, 1980.

Joel Huber, "Predicting Preferences on Experimental Ratings of Attributes: A Comparison of Models," Journal of Marketing Research, 12 (August 1975), 290-7.

Joel Huber and Morris B. Holbrook, "Using Attribute Ratings for Product Positioning: Some Distinctions among Compositional Approaches," Journal of Marketing Research, 16 (November 1979a), 507-16.

Joel Huber, "The Determinants of Esthetic Value and Growth," Advances in Consumer Research, v. 7, Association for Consumer Research, 1979b.

R.M. Johnson, "Market Segmentation: A Strategic loci," Journal of Ilarketing Research, 8 (February 1971), 13-18.

Harry Munsinger and William Kessen, "Uncertainty Structures and Preference," Psychological Monographs, 78, (1964).

R.M. Nicki, "Approach and Following Behavior of 24-hour Old Chicks as a Function of Stimulus Complexity," Animal Behavior, 23 (1976), 116-123.

Charles E. Osgood, George J. Suci, and Percy H. Tannen baum, - The Measurement of Meaning, Urbana: University of Illinois Press, 1957.

Robert S. Wyer, Cognitive Organization and Change: A Information Processing Approach, Potomic:, Maryland: Wiley, 1974.