Journal of Consumer Research, March 1998 v24 n4 p374(21)

The influence of unity and prototypicality on aesthetic responses to new product designs. (includes appendix) Robert W. Veryzer Jr.; J. Wesley Hutchinson.

Abstract: Unity and prototypicality are important visual aspects of product design. These design principles were operationalized by modifying line drawings of existing products. The results of four experiments provide evidence that these two factors positively affect aesthetic response. These effects were strongest when visual properties were the sole basis of judgment and when design variations were easily compared. However, they persisted when aesthetic aspects were combined with other product information and when comparing design features was difficult. The effect of unity was found to be superadditive, suggesting that it has a relational, "all-or-none" character. Finally, regression analyses show that direct effects of the design modifications on aesthetic response exist in addition to possible indirect effects that are mediated by perceived typicality. (Reprinted by permission of the publisher.)

Full Text: COPYRIGHT 1998 University of Chicago Press

The design of products inherently involves aesthetics. Moreover, aesthetic aspects of a product are a potential source of pleasure for the consumer (Holbrook and Zirlin 1985). As a result, the influence of aesthetics is increasingly being acknowledged as an important part of new product development (Whitney 1988), marketing strategy (Kotler and Rath 1984), product quality (Garvin 1984; Zeithaml 1988), product differentiation (Dickson and Ginter 1987), and competitive advantage (Holt 1985; Kotler and Rath 1984). Despite the growing awareness of the influence of aesthetic responses on product preferences and a growing number of empirical studies of the arts in general (e.g., Berlyne 1974; Holbrook 1995; Lombardo 1991; Martindale, Moore, and Borkum 1990; Wohlwill 1981), there is surprisingly little experimental research testing specific hypotheses about how aesthetic responses are related to product design (see Berkowitz [1987], Bloch [1995], and Holbrook and Zirlin [1985] for related discussions).

Much of the existing work in this area has been concerned with the definition and scope of consumer aesthetics rather than with the specific determinants of aesthetic response. Although the former are clearly important issues, aesthetic research is in need of a broad empirical base to support theory development and to validate the implications of those theories for other aspects of consumer behavior. Consumer research also needs to address the important applied issue of which concrete, easily implemented rules of thumb can be supported well enough to merit use in the everyday decisions that product designers face. This article contributes to these goals by providing a systematic empirical analysis of two potential determinants of aesthetic response: unity and prototypicality. "Unity" refers to a congruity among the elements of a design such that they look as though they belong together or as though there is some visual connection beyond mere chance that has caused them to come together (Lauer 1979; Veryzer 1993a, 1993b). In the experiments reported here, we focus on visual "matching" (i.e., the visual similarity between different parts of the same design) as a means for achieving unity. Prototypicality is the degree to which an object is representative of a category (Barsalou 1985; Rosch 1978). In our experiments, we make a sharp distinction between the concrete design principle of prototype distortion (e.g., number of modifications made to a prototype design) and perceived typicality (e.g., goodness-of-example ratings), focusing on the former as a potential determinant of the latter as well as aesthetic response.(1)

In four experiments, we show that the effects of unity and prototype distortion are reliable and robust. These effects replicate previous results for prototype distortion and provide the first demonstration of unconfounded unity effects. The results also support the view that visual features exert both first-order atomistic effects that are independent and additive in nature as well as higher-order relational effects that are interactive and superadditive in nature. These effects were found for both aesthetic response and perceived typicality. In particular, prototype distortion effects are approximately additive, and unity effects are superadditive. Finally, we provide evidence that there are direct effects of unity and prototype distortion on aesthetic response in addition to possible indirect effects that are mediated by perceived typicality.

FACTORS THAT INFLUENCE AESTHETIC RESPONSE

[Expanded Picture] As with any complex behavior, aesthetic response is likely to involve a number of factors. These factors differ in their degree of generality. At one end of this continuum are the various abstract principles of perceptual organization that have been discussed mainly in the art and experimental aesthetics literatures (e.g., Berlyne 1971; Holbrook 1995; Lauer 1979; Lombardo 1991; Martindale et al. 1990); at the other end are learned responses that are specific to particular objects or categories (e.g., Gordon and Holyoak 1983; Loken and Ward 1990; Rosch 1978). This continuum is a useful way of viewing the determinants of aesthetic response because it captures the notion that the determinants vary in the degree to which they are influenced by experience, including external interventions (such as marketing activities, fashion changes, and social norms). In the present research, we have selected two factors, unity and prototypicality, that span this continuum and are representative of each extreme.

Unity

Visual design principles describe perceived spatial relations between the parts of an object. Unity encompasses any aspect of a visual display that connects its parts in a meaningful way. Such connections can be achieved in a variety of ways, including causal and symbolic relations, but most involve some form of matching with respect to a salient visual characteristic (Ching 1979; Lauer 1979). Unity achieved in this way is therefore closely related to the general tendency to perceive groupings of elements as an integrated entity. One approach to explaining how this integration occurs is found in Gestalt theories of figural goodness, or pragnanz (Koffka 1935). Thus, the Gestalt laws of proximity (i.e., elements that are closest to each other tend to form groups), similarity (i.e., elements that are similar tend to form groups), and common destiny (i.e., parts of a figure that have a common destiny tend to form units) may be viewed as ways to achieve unity (Katz 1950; Lauer 1979). According to the Gestalt psychologists, beauty is dependent on the degree to which an object displays relations consistent with the Gestalt laws of organization. Koffka (1935) clearly suggested this when he discussed how violations of such laws as "good continuation" and "good shape" are not only felt as violations but also conflict with our feeling of "fit" and "hurt our sense of beauty" (Koffka 1935, p. 175).

It is surprising that there is very little experimental research that relates unity to aesthetic responses. A study by Bell, Holbrook, and Solomon (1991) examined the impact of gestalt-like ensemble effects and the influence of personality factors on product evaluations. This research provided support for the view that unity may systematically influence aesthetic responses. In Bell et al.'s study, subjects were asked to look at color photographs of living-room furniture. Some sets of furniture were consistent with respect to style (i.e., all contemporary or all traditional); others were a random mix of styles. Bell et al. found that aesthetic response was correlated with perceived unity and that perceived unity was related to style (i.e., consistent or mixed). These findings suggest that the principle of unity influences aesthetic response.

The influence of unity has also been examined in the context of studies of social information processing. Lennon (1990) investigated the effects of clothing attractiveness on social perceptions. Subjects viewed slides of models wearing attractive and unattractive outfits. In the attractive-clothing condition, models wore clothing that was well matched and wore accessories to complement their clothing. In the unattractive-clothing condition, models wore garments and accessories that did not match either in color, style, or pattern. The distinction between match and mismatch was based on pretest ratings of attractiveness. Lennon found that models dressed in well-matched clothing were perceived to be more competent, more desirable as a co-worker, and more sociable than models dressed in unattractive clothing. These results provide further support for the view that configurations of aesthetic elements (i.e., visual organization rules such as unity, color harmony, and repetition) can significantly influence perceptions.

The preceding discussion suggests that aesthetic responses are more positive for products exhibiting high unity than they are for products that are low in unity. However, in these studies unity was operationalized through the use of preexisting perceptions and social norms. Thus, little was learned about the principles that underlie aesthetic responses. In particular, those operationalizations were likely to be confounded with other variables, including familiarity, novelty, and social acceptability. In addition, the generalizability of these studies is limited because the research was conducted within two very fashion-oriented domains. The stimulus design used in the research reported here avoids these problems.

Prototypicality

Prototypicality, or typicality, is the degree to which an object is representative of a category. Prototypes are usually defined as the central representation of a category or as possessing the average or modal value of the attributes of that category (e.g., Homa 1984; Langlois and Roggman 1990; Medin and Smith 1984; Reed 1972; Rosch 1978). In the research reported here, it is important to distinguish between two uses of the term prototypicality. First, prototypicality can be used as a concrete design principle; namely, common designs already existing in the marketplace can be systematically altered to make them less typical (e.g., newer, more novel, etc.). Much cognitive research on categorization and concept formation has manipulated stimuli in this way (e.g., Homa 1984; Hutchinson and Alba 1991; Kemler Nelson 1984; Medin and Schaffer 1978; Palmeri and Nosofsky 1995; Ward and Scott 1987). This is often called "prototype distortion," and we will adopt that term here.(2) Usually, distortion is defined as the number of physical changes made to a designated prototype and as such is an objectively defined property of the stimulus. The second use of prototypicality refers to subjective perceptions of typicality or category representativeness. This is often measured by goodness-of-example ratings (e.g., Barsalou 1985; Rosch 1975). Although a direct relationship between the two has been observed or assumed in cognitive research, it is also the case that perceived typicality is often affected by factors in addition to prototype distortion (e.g., frequency of instantiation and preference; see Barsalou 1985; Nedungadi and Hutchinson 1985). In the present research we are mainly interested in the effects of concrete design principles (such as prototype distortion and unity) on aesthetic response. However, we also examine the effects of these design principles on perceived typicality.

There is evidence that people respond most favorably to objects that are highly prototypical and less favorably to objects that are less prototypical (Barsalou 1985; Carpenter and Nakamoto 1989; Gordon and Holyoak 1983; Langlois and Roggman 1990; Loken and Ward 1990; Martindale and Moore 1988; Martindale, Moore, and West 1988; Nedungadi and Hutchinson 1985). A number of explanations have been suggested for the relationship between prototypicality and preference. One explanation proposes that highly prototypical items are perceived as more familiar and therefore are better liked (Gordon and Holyoak 1983; Kunst-Wilson and Zajonc 1980). Another explanation maintains that highly prototypical category members are preferred because they have more valued attributes. This explanation does not hold that prototypicality per se leads to product preference, but rather that as product categories evolve one or a few products tend to become market-share leaders because they have attributes widely desired by consumers who buy the product. Competitive brands are designed to appeal to the same segments of consumers, so they are similar in many ways to market leaders (Loken and Ward 1990). It has also been suggested that the link between prototypicality and preference may in part be due to the information value of proto-types (Rosch 1978, p. 37). That is, the ability of a prototypical object to represent the category as a whole may have some value in and of itself.

Although prototypicality seems to provide a satisfactory explanation of aesthetic response in some cases (see, e.g., Gordon and Holyoak 1983; Kunst-Wilson and Zajonc 1980; Loken and Ward 1990), such an explanation does not seem adequate in others (see, e.g., Meyers-Levy and Tybout 1989). In fact, in some cases it is the opposite of prototypicality (e.g., novelty, distinctiveness) that seems to be associated with positive aesthetic response (see, e.g., Woll and Graesser 1982). It may be that people prefer more novel products because they are seeking variety (Holbrook and Hirschman 1982; Hutchinson 1986; McAlister and Pessemier 1982) or perhaps because of a product's salience relative to other products (Loken and Ward 1990; Woll and Graesser 1982). There is also a purely economic explanation for these product preferences in some situations. The very best products tend to be expensive; therefore, they are rare and purchased primarily by wealthy individuals. Thus, the most preferred (and least affordable) products may be very atypical.

One potential explanation of these apparently contradictory findings is that there is an inverted-U-shaped relationship between familiarity and affect. That is, moderate familiarity is preferred to both extreme familiarity and extreme novelty. It is reasonable to assume that prototypical objects will range from extreme to moderate familiarity and atypical objects will range from moderate familiarity to extreme novelty. Thus, either a positive or negative relationship between prototypicality and preference might be observed depending on the levels of familiarity associated with the particular set of items used in any specific experiment.

Support for this explanation can be found in classic research on preferences for novelty and complexity (Berlyne 1970; Eisenman 1968). More recently, Mandler (1982) has theorized that the level of congruity between a product and a more general product category schema may influence the nature of information processing and thus product evaluations. Products that are moderately incongruent with their associated category schemas are said to stimulate processing that leads to a more favorable evaluation relative to products that are either congruent or extremely incongruent with the category schema. (Of course, the beneficial effects of incongruity on memory are well known; see Stangor and McMillan [1992] for a review.) Support for Mandler's hypothesis was demonstrated in a series of studies conducted by Meyers-Levy and Tybout (1989). However, work by Martindale and his colleagues in a number of domains has failed to find the inverted-U-shaped relationship hypothesized by Berlyne (see, e.g., Martindale and Moore 1988; Martindale et al. 1988; see also Drolet, Keller, and Zajonc 1996). In fact, to the extent that nonlinearity is present in their results, it has been U-shaped, indicating some preference for extreme novelty in addition to a basic typicality effect.

In summary, a simple, positive effect of prototypicality is frequently found, However, there is also evidence that novelty and schema incongruity sometimes have effects.

Atomistic versus Relational Properties of Product Designs

Our definitions of unity and prototypicality illustrate a theoretically and pragmatically important distinction: atomistic versus relational properties of product designs. Most products have easily identified parts. These parts may be perceived by consumers as distinct components of the product and, therefore, exert independent effects on consumer reactions. We refer to such properties of the product as "atomistic," and prototype distortion is defined in this simple, additive way (i.e., number of modified parts). Alternatively, consumers may perceive various relationships among the parts that disrupt this independence and create reactions that are interactive. We refer to such properties as "relational." Unity is a relational property because it requires that parts match in some way. Moreover, designs can vary in the coherence of their relational properties, and unity requires a high level of relational coherence. Thus, relational properties should have an "all-or-nothing" nature because, strictly speaking, the relationship does not hold unless each relevant part is appropriately designed. This suggests that the effect of unity should be superadditive. In particular, as parts are modified to match each other in some way the largest positive effects should be observed when all of the parts match (e.g., the effect of making a partially unified design completely unified should be larger than the effect of making a design with no matching parts partially unified). Put differently, it is natural to expect that a modification that disrupts a perfectly unified design will exert a greater negative effect than the same modification applied to a design whose unity has already been disrupted.

[Expanded Picture] This distinction has important pragmatic implications. For example, one popular approach to new product development matches a large number of specific customer needs with an equally large number of engineering characteristics of the product (see, e.g., Griffin and Hauser 1993; Hauser and Clausing 1988). The product is viewed as a system of components that can be combined in various ways to best provide value to the customer. Thus, the nature of the combination rule that determines perceived value is very important. The rules used in practice are often additive and implicitly assume that the most important properties of the product are atomistic rather than relational. This results primarily from pragmatic constraints because interactive effects are a major complication. If there are 100 components corresponding to consumer needs (which is typical), then there are 4,950 two-way interactions, 161,700 three-way interactions, and so on. An exhaustive analysis of interactions is impossible. Hauser and Clausing's "House of Quality" approach focuses on the use of technical expertise to identify key interactions between engineering characteristics (i.e., the "roof" of the house), but it does not address the issue of perceptual interactions. In fact, this is a classic problem in conjoint analysis, and most applications ignore the interactive effects of product attributes on consumer preferences (see, e.g., Green and Srinivasan 1978, 1990).

Clearly, it would be desirable to have a general theory of perceptual interactions for aesthetic response. Although such a theory does not yet exist, several research areas in cognitive psychology have addressed this issue at least in part for nonaesthetic responses. Early work on the processing of attribute information led to the discovery of clear differences between so-called separable, integral, and configural dimensions (see, e.g., Garner 1976). The manipulations of configurality in these experiments and our manipulations of unity are similarly based on perceptual matching relations. Recent work by Goldstone, Medin, and Gentner (1991; see also Medin, Goldstone, and Gentner 1993) on perceived similarity is more directly relevant. The central premise in their approach is that relational characteristics often dominate atomistic characteristics, and the relative importance of each is context dependent.(3) Similarly, Murphy and Medin(1985) have argued that "mere similarity" (that is based on atomistic characteristics alone) cannot account for the accumulated body of experimental results on categorization and concept formation. Rather, they argue that higher-order relations, and especially causal relations, are needed for a full explanation.

One major goal of this article is to examine the differences between atomistic and relational properties in product designs. This provides a first step in developing a theory of perceptual interactions for aesthetic response.

Summary and Overview

Research from a wide variety of problem areas suggests that unity and/or prototypicality might affect aesthetic responses. None of this research, however, (1) manipulates these factors orthogonally; (2) examines the robustness of the effects with respect to different product types, visual properties, and other contextual variables; or (3) specifically examines the roles of atomistic and relational aspects of visual design in determining aesthetic response. In the research reported here, we constructed stimuli such that differences in unity were not confounded with differences in prototype distortion. Because prototypicality and unity are generally correlated in naturally occurring products, these manipulations were crucial for isolating specific determinants of aesthetic response.

In experiment 1, we establish that unity and prototype distortion exert strong effects on both aesthetic response and perceived prototypicality. The design allows us to test specific hypotheses about the linearity of these effects that are related to the theoretical issues discussed earlier. We also examine via two-stage least squares the extent to which the effects of unity and distortion on aesthetic response are mediated by perceived prototypicality. In the remainder of the article, we test the generality of unity and distortion effects. In experiments 2 and 3, we use a broader set of stimuli and modify stimulus presentation to simulate consumer environments that enhance or inhibit the visual comparison of competing product designs. In experiment 4, we test generality by combining verbal information with the purely visual stimuli of the first three experiments.

STIMULUS DESIGN

The general method of constructing stimuli that operationalize unity and prototype distortion was common to all four experiments. Therefore, we provide a brief overview at the outset.

Manipulating Unity and Prototype Distortion

Many aspects of an object's appearance (e.g., color, perspective, shading) have the potential to affect aesthetic response. To isolate specific factors such as unity and prototypicality, all other visual properties must be eliminated or controlled. Therefore, in the experiments reported here line drawings of products were used as stimuli.

The stimulus sets were constructed by first creating a realistic line drawing of a commonplace product on the basis of photographs from a magazine or merchandise catalog. We refer to this product as the prototype design, and in experiment 1 we verify that these designs were indeed perceived as prototypical. Two prominent parts of the prototype were then selected for modification. Design variations were produced by altering either one, the other, or both of the two selected parts with respect to shape, trim, or texture. For example, Figure 1a shows a stimulus set consisting of four members and three levels of prototype distortion (i.e., zero, one, or two modifications, or steps away from the prototype). Unity is also affected by these modifications because each part was modified in the same way. In particular, the two one-step distortions are less unified than the prototype because the two principal parts no longer match with respect to shape (i.e., straight vs. curved shape). However, unity is restored when both parts are modified in the same way (i.e., the two-step distortion). In this case, the parts match even though the resulting design is rather unusual. In experiment 1, a second type of modification is introduced into each stimulus set, producing five levels of distortion and three levels of unity (e.g., trim in addition to shape; . This design provides two two-step distortions, one that is unified with respect to both types of modification and one that is mismatched with respect to both. Experiments 2-4 use four-item stimulus sets similar to the one in Figure 1a.

Product Categories and Types of Modifications

[Expanded Picture] In order to increase the generalizability of the results, nine product categories and three types of modification were selected for use in constructing stimuli. The nine product categories were: bathroom scales, clocks, dressers, flashlights, hair dryers, lamps, refrigerators, telephones, and television remote controls. The three types of modification were: shape, texture, and trim. In some cases, multiple modifications were used within the same stimulus set (i.e., experiment 1). Across all four experiments, 144 different products designs were used.(4) Many of the modifications resulted in extremely unusual, "new-to-the-word" designs like those in Figure 1. However, some modifications yielded designs that were plausible but less typical (as confirmed by manipulation-check measures reported subsequently). Thus, our results are not dependent on extreme novelty.

EXPERIMENT 1

In the introduction, we characterized prototype distortion as an atomistic property of product designs and unity as a relational property. Consistent with this, we expect that when multiple visual modifications are combined, the effects of distortion will be additive and the effects of unity will be superadditive. That is, each modification moves the new design incrementally away from the original prototype. However, unity may exhibit more of an "all-or-none" effect in which the first modification that disrupts a unified design exerts a much larger negative effect on aesthetic response than subsequent modifications (even when those modifications disrupt the unity of the design in a new way). Thus, the effect of unity is superadditive in the sense that the joint effect of unity with respect to two visual aspects of a product design are greater than the sum of the two effects measured separately (specific tests of superadditivity are discussed in more detail subsequently). These predictions are summarized in the following formal hypotheses.

H1a: In general, as prototype distortion increases, aesthetic responses become less favorable.

H1b: More specifically, because prototype distortion is an atomistic property the effect of each modification on aesthetic response should be approximately additive (i.e., linear).

H2a: In general, as unity increases, aesthetic responses become more favorable.

H2b: More specifically, because unity is a relational property the effect of each unified visual aspect on aesthetic response should be superadditive.

The second theoretical issue that is addressed in experiment 1 is the extent to which prototype distortion is the sole determinant of perceived typicality. Clearly, prototype distortion should decrease perceived typicality. Testing this hypothesis provides a manipulation check on our assumptions that the prototype products we chose were reasonably typical and that the modifications we introduced generally made the designs less typical. In addition, because most products found in the marketplace are visually unified (or at least more unified than our nonunified designs) it would be reasonable to expect a positive effect of unity on perceived typicality. These predictions are summarized in the following formal hypotheses.

H3a: In general, as prototype distortion increases, perceived typicality decreases.

H3b: More specifically, because prototype distortion is an atomistic property, the effect of each modification on perceived typicality should be approximately additive (i.e., linear).

H4a: In general, as unity increases, perceived typicality increases.

H4b: More specifically, because unity is a relational property the effect of each unified visual aspect on perceived typicality should be superadditive.

Our earlier discussion of the effects of typicality on aesthetic response taken together with Hypothesis 4 suggest that the effect of unity on aesthetic response might be mediated by perceived typicality. That is, prototype distortion might reduce perceived typicality, which would in turn make the aesthetic response less positive. Alternatively, unity is primarily an aesthetic construct, so it should affect aesthetic response directly, and this effect should be larger than its effect on perceived typicality. Conversely, prototype distortion should exert a larger effect on perceived typicality than on aesthetic response because, on average, such modifications will make the design less representative of the category as whole. These differences in the relative effect sizes of unity and prototype distortion argue against the hypothesis that the effect of unity on aesthetic response is completely mediated by its effect on perceived typicality. Of course, both direct and indirect effects might occur (i.e., partial mediation). In fact, this outcome would be most consistent with the literature discussed earlier. Thus, we expect that independent (or direct) effects of unity will persist in regression models of aesthetic response that include perceived typicality (and therefore an indirect effect of unity) as a predictor variable. These predictions are summarized in the following formal hypotheses.

H5a: The effect of prototype distortion is greater for perceived typicality than for aesthetic response.

H5b: The effect of unity is greater for aesthetic response than for perceived typicality.

H6: The effect of unity on aesthetic response is partially mediated by its effect on perceived typicality; however, an independent direct effect also exists.

Earlier we discussed the seemingly contradictory findings regarding the effects of prototypicality on aesthetic response. In experiment 1, we examine this issue using regression models. A significantly negative quadratic component in such models would support the schema-incongruity hypothesis (Mandler 1982; Meyers-Levy and Tybout 1989). Given the reasonably large amount of variation in perceived typicality for our stimuli, the absence of such a quadratic component, together with a significantly positive linear component, would replicate the simple prototypicality effects found in other studies (e.g., Martindale and Moore 1988; Martindale et al. 1988).

H7: When schema incongruity is defined by perceived typicality rather than prototype distortion, moderate levels of incongruity are most preferred relative to low or high levels.

Method

[Expanded Picture] Stimulus Design. In order to keep the task manageable for subjects, six of the nine product classes were used in this experiment.(5) Also, any given subject judged only six designs for each product class. Two types of modification were chosen for each product such that each type of modification occurred equally often across products. The products and types of modification were as follows: clock (texture and shape), dresser (trim and shape), hair dryer (texture and shape), lamp (trim and texture), refrigerator (trim and shape), and TV remote control (trim and texture). For each complete stimulus set of [2.sup.4] = 16 designs, four six-item subsets were constructed such that each subset contained five levels of distortion and three levels of unity and permitted orthogonal linear contrasts for a distortion effect (denoted [P.sub.1]), a unity effect (denoted [U.sub.1]), and an additivity test for each (denoted [A.sub.P] and [A.sub.U], which are positive when superadditivity is indicated). The four subsets and the weights used in the four contrasts are given in Table 1, and an illustrative stimulus set is shown in Figure 1b.(6) Note that the contrast for prototype distortion has been reverse scored so that positive values of the contrast support Hypotheses 1a and 3a. This also maintains sign consistency with subsequent indicators of the prototypicality effect.

Four decks of 36 paper cards were constructed by randomly assigning each of the subsets for a given product to one of the stimulus sets. Each card contained a design, a label indicating the product class, and an identification code. This stimulus design provided within-subject measures of product type, prototype distortion, unity, and additivity but used a small number of stimului per subject so that the judgment task was not too burdensome. The relationship of this stimulus design to Hypotheses 1-4 can be seen in Figure 2. The three levels of unity are represented by different markers. The prototype distortion effect is evident within each level of unity (i.e., the slopes of the connected markers in ). The unity effect is represented by the degree of separation between the three markers. Superadditivity is indicated when the highest level of unity (i.e., circles) is separated from the other two levels (i.e., squares and diamonds), and those levels are relatively close to each other . [A.sub.U] provides a test of this that is independent of the linear effects of distortion and unity (see Table 1). Also, even though a unity effect induces a quadratic effect of prototype distortion, this design permits a test of the additivity of prototype distortion when unity is held constant. A superadditive effect of prototype distortion would be indicated by convexity in the curve connecting the three stimuli that have the highest level of unity (i.e., Ap [greater than] 0; see Table 1 and . To aid comparison, the six stimulus types are numbered in the same way in all tables and figures.

[Expanded Picture] [Expanded Picture] Procedure. Fifty undergraduates in an introductory marketing course at the University of Pennsylvania received course credit for participating in the experiment. Subjects were randomly assigned to one of the decks of stimuli. Twenty-three subjects provided goodness~ of-example ratings, and 27 subjects rated visual attractiveness. Thus, the experiment used a 4 x 2 between-subjects design. Subjects were given a deck of 36 stimuli and a response sheet. The response sheet had written instructions that were read aloud to the subjects and reviewed. The response criterion for the rating task was either visual attractiveness or goodness of example. For visual attractiveness, the subjects were told, "Your task is to rank these designs in terms of how visually attractive you find them to be. Your rankings should reflect your personal tastes; there are no right or wrong answers." For goodness-of-example ratings, the subjects were told, "Your task is to rank these designs in terms of how representative, or typical, the design is of the product category as a whole. For example, some birds, such as robins and sparrows, are very good examples of the category 'birds' while other birds, such as penguins and ostriches, are very poor examples of the category. Your rankings should reflect your personal knowledge of the category; there are no right or wrong answers. Notice also that there is no necessary relationship between this judgment of goodness of example and your personal preferences. For example, your favorite bird may be either a penguin or a robin." More detail was provided for goodness-of-example ratings because, unlike attractiveness, it is not a commonplace concept. Also, these instructions closely followed those used in cognitive research on the graded structure of categories (e.g., Barsalou 1985; Rosch 1975). Questions about the criterion were then solicited and discussed. Subjects were run in small groups that were blocked by task.

After the response criterion was explained, subjects were asked to (1) go through the deck once to familiarize themselves with the product designs, (2) separate the cards into several groups according to the criterion, (3) rank order the cards according to the criterion, and finally, (4) enter the identification codes of the design on each card into the appropriately ranked box on the response sheet and assign a numerical rating to the design using a 100-point scale. The highest-ranked design was assigned a rating of 100, and other designs were given smaller numbers, reflecting judgment magnitudes. Subjects were instructed to use the same number twice only if they could not decide between the two designs.

Results

To provide tests of our hypotheses, [P.sub.1], [U.sub.1], [A.sub.P], and [A.sub.U] were dependent measures in separate mixed-design ANOVAs in which the between-subjects factors were deck and task, and product was a repeated measure. This provides a simpler, more direct approach to testing our hypotheses. Means for each measure are provided in Table 2. Means for each stimulus type are plotted in Figure 2D. Main effects of deck and interactions of deck with other factors are predicted whenever the individual effects of the four modifications vary considerably. These effects are not of direct interest and are counterbalanced across decks in our design. Therefore, to simplify the exposition we do not report them here.

Prototypicality. The mean value of [P.sub.1] was significantly greater than zero (F(1, 42) = 95.4, p [less than] .0001). There was also a main effect of task (F(1, 42) = 17.9, p [less than] .0001); however, follow-up analyses confirmed that [P.sub.1] was significantly greater than zero for each task (F(1, 23) = 13.2, p [less than] .002 and F(1, 23) = 137.4, p [less than] .0001 for visual attractiveness and goodness-of-example ratings, respectively). Prototype distortion clearly had a strong positive effect on both aesthetic response (supporting Hypothesis 1a) and perceived typicality (supporting Hypothesis 3a). The main effect of task supports Hypothesis 5a and means that the differences in slope evident in Figure 2D are statistically significant. As there was also a main effect of product (F(5, 210) = 19.9, p [less than] .0001) and an interaction between product and task (F (5, 210) = 6.19, p [less than] .0001), [P.sub.1] was tested separately for each product. For goodness-of-example ratings, [P.sub.1] was greater than Zero for all products and five of six values were statistically significant (see Table 2). For visual attractiveness ratings, [P.sub.1] was greater than zero for five of six products and three of six values were statistically significant (again, see Table 2).

Additivity of the Prototypicality Effect. Although the lines connecting the high-unity stimuli in Figure 2D (i.e., the circles) appear straight, supporting Hypotheses lb and 3b, analyses of [A.sub.P] revealed a small but significant level of subadditivity (F(1, 42) = 5.66, p [less than] .05). When [A.sub.P] was analyzed separately for visual attractiveness and goodness-of-example ratings, however, neither reached traditional levels of statistical significance. As there was a main effect of product (F(5, 210) = 4.55, p [less than] .001), [A.sub.P] was tested separately for each product. For goodness-of-example ratings, [A.sub.P] was less than zero for four of six products and one of six values was statistically significant (see Table 2). For visual attractiveness ratings, [A.sub.P] was less than zero for four of six products and two of six values were statistically significant (again, see Table 2). Overall, the effect of distortion is approximately additive, but there is a weak trend toward subadditivity. This subadditivity was not predicted, but one interpretation is that designs assimilate to the prototype. Thus, the same distortion exerts a smaller effect when applied to a prototype than when applied to a nonprototype.

[Expanded Picture] Unity. As predicted by Hypotheses 2a and 4a, the mean value of [U.sub.1] was significantly greater than zero (F(1, 42) = 131.9, p [less than] .0001). The main effect of task was not significant; therefore, Hypothesis 5b was not supported. However, there was a main effect of product (F(5, 210) = 10.2, p [less than] .0001) and an interaction between product and task (F(5, 210) = 2.36, p [less than] .05), so follow-up analyses were undertaken. These analyses revealed that [U.sub.1] was significantly greater than zero for each task (F(1, 23) = 52.1, p [less than] .0001 and F(1, 23) = 86.3, p [less than] .0001 for visual attractiveness and goodness-of-example ratings, respectively). Also, [U.sub.1] was tested separately for each product. For goodness-of-example ratings, [U.sub.1] was significantly greater than zero for all products; for visual attractiveness ratings, [U.sub.1] was greater than zero for all products and five of six values were statistically significant (see Table 2). Thus, the effects of product and the interaction between product and task were due to differences in the size of the unity effect and not to qualitative reversals. Overall, the effect of unity was quite robust.

[Expanded Picture] Superadditivity of the Unity Effect. The observed pattern of results exhibited in Figure 2D is clearly most similar to the predictions in Figure 2B. That is, the gap between the unified stimuli (i.e., those numbered 1, 3, and 6 in and partially unified stimuli (i.e., those numbered 2 and 5) is greater than the gap between the partially unified stimuli and the nonunified stimulus (i.e., number 4). Therefore, the superadditivity of unity (i.e., Hypotheses 2b and 4b) was supported. Overall, [A.sub.U] was significantly greater than zero (F(1, 42) = 12.6, p = .001). There was no main effect of task or product, nor did they interact. As can be seen in Table 2, [A.sub.U] was more consistently positive than [A.sub.P] (i.e., five of six products for each dependent measure). Also, the degree of superadditivity is greater for visual attractiveness than for goodness of example (i.e., 10.3 vs. 5.8, and [P.sub.1], [U.sub.1], and [A.sub.P] exhibit the opposite trend). This superadditivity contrasts with prototype distortion, which is approximately additive (i.e., slightly subadditive) and adds to growing evidence that atomistic and relational properties behave differently in creating an overall impression.

Testing Mediation and Schema Incongruity. Given this strong evidence that unity affects perceived typicality, it is possible that the effect of unity on aesthetic response is mediated by perceived typicality. We argued earlier (i.e., Hypothesis 6) that unity is primarily an aesthetic concept; therefore, a direct effect of unity on aesthetic response should be found even if an indirect effect (mediated by perceived typicality) also exists. Similarly, even though an effect of moderate schema incongruity is not evident when prototype distortion represents schema incongruity, an effect might emerge when perceived typicality represents schema congruity (i.e., Hypothesis 7). To reveal these relationships, the overall means presented in Figure 2 have been replotted in Figure 3 to show mean visual attractiveness ratings as a function of mean goodness-of-example ratings. This plot clearly shows that an effect of unity persists, but there is little evidence of an effect of moderate schema incongruity (i.e., there is no inverted-U shape to the plot, overall or considering only the unified stimuli).

In order to statistically test these conclusions, a series of regression analyses following the approach suggested by Baron and Kenney (1986) were conducted. Because visual attractiveness and goodness of example are both measured with error and because our prior discussion suggests that reciprocal causation is possible, a two-stage least-squares approach was used. When either measure served as an independent variable, predicted values from a first-stage regression were substituted for observed values (Baron and Kenny 1986; Greene 1993; James and Singh 1978).(7) This method removes measurement error and allows separate models for each causal direction.

The results of the second-stage regressions are given in Table 3. Consistent with Baron and Kenney's approach, models 1 and 2 are preliminary analyses of visual attractiveness that suggest linear effects of predicted goodness of example, prototype distortion, and unity and a quadratic (superadditive) effect of unity.(8) Model 3 provides the main test of mediation. Consistent with Hypothesis 6, the results suggest a direct (superadditive) effect of unity persists even when perceived typicality (i.e., goodness of example) is included as a potential mediator. The reduction in the size of the coefficient suggests that both direct and indirect effects are present. An effect of prototype distortion also persists, and the sign reversal reflects the fact that this effect is weaker for visual attractiveness than for goodness of example. To explore the possibility of reciprocal causation we conducted similar analyses for goodness-of-example ratings. Model 6 provides the main test of mediation and suggests that direct effects of prototype distortion and unity persist even when preference (i.e., visual attractiveness) is included as a potential mediator. Although we did not specifically predict this outcome, it is noteworthy that the coefficient for the linear effect of prototype distortion on goodness-of-example ratings is quite stable compared with the coefficients for predicted visual attractiveness and the linear effect of unity. Finally, we note that there is no evidence of a quadratic effect of predicted goodness of example on visual attractiveness, so Hypothesis 7 is not supported.

[Expanded Picture] Our primary interest in the present analyses is to establish that the effect of unity on aesthetic response is not completely mediated by its effect on perceived typicality, and the results support this conclusion. In this regard, the changes we observed in coefficient size should be interpreted cautiously. The first-stage model was much more complex (i.e., including interactions and main effects) than the second-stage model so that the test of the persistence of direct effects would be conservative. A more complex model of the direct effects of unity and distortion (e.g., one that included interactions with product) might exhibit less change in coefficient size. We leave for future research the task of establishing the detailed causal relations that exist between prototype distortion, unity, aesthetic response, and perceived typicality.

EXPERIMENT 2

Experiment 2 tests the generality of unity and prototype distortion effects by expanding the number of product classes to nine and using all three types of modification (i.e., shape, texture, and trim) in each product class. Also, stimuli from each product class were presented simultaneously on a single page. This makes visual comparisons easier and more salient than they were in experiment 1. This simulates shopping environments in which entire product lines for several manufacturers are displayed together in the same area of the store (or on the same page of a catalog). Experiment 3 explores the opposite extreme by presenting only one design from each product class to each subject.

Method

[Expanded Picture] Experimental Design. The overall design of the experiment was a 2 (modification of part 1) x 2 (modification of part 2) x 9 (product) x 3 (type of modification) x 3 (version) x 2 (order) mixed-factorial Latin-square design. Version and order were between-subject factors. Twenty-seven sets of stimuli (9 products x 3 types of modification) were constructed and organized into three questionnaire versions. Each stimulus set contained four designs representing modifications of two parts with respect to a single visual property (i.e., shape, texture, or trim; for an example). Each questionnaire version contained nine stimulus sets. Each product category occurred once, and each type of modification occurred three times in each questionnaire version. Thus, modification of part 1, modification of part 2, product, and type of modification were all within-subjects factors. The (randomized) order in which the 36 stimuli were presented was reversed for half of the subjects. The four stimuli in each set were presented together on a single page in randomized locations.

Experimental Procedure. One-hundred and ninety-seven volunteer subjects enrolled in an introductory marketing course at the University of Florida participated in this experiment and received extra course credit as a result. The subjects were run in groups of 10-20 participants.

The stimulus materials were contained in a booklet. The introductory page informed the subjects that they would be shown drawings of products and asked to evaluate the appeal of the product designs based on their appearance. The subjects were told that the purpose of the study was to obtain consumers' reactions to potential new products that were in the early stage of development. Subjects were then told that each of the designs of a given product performed equally well and that they were to rate all four designs that were shown on each page. Subjects proceeded through the task at their own pace. They rated each product design on a nine-point semantic differential scale that measured visual attractiveness (i.e., anchored by "beautiful" and "ugly").(9)

Results

The data were analyzed using a mixed ANOVA design. Version and order were treated as between-subjects factors and modification of part 1, modification of part 2, product, and type of modification were treated as within-subjects factors.

Dependent Measures. Because the design for each stimulus set is simpler than in experiment 1 (i.e., modification of part 1 x modification of part 2), distortion and unity effects can be tested using the main effects of modification (for distortion) and their interaction (for unity). As in experiment 1, however, some analyses are more straightforward, using contrasts for distortion and unity. These contrasts are [U.sub.2] = ([y.sub.++] + [y.sub.--] - [y.sub.+-] - [y.sub.-+])/2 for unity and [P.sub.2] = [Y.sub.++] - [Y.sub.--] for distortion, where [y.sub.++], [y.sub.+-], [y.sub.-+], and [y.sub.--] are visual attractiveness ratings for each member of the stimulus set (e.g., designs A, B, C, and D, respectively, in ).

Unlike the measures used in experiment 1, [U.sub.2] is simply the interaction of the two modifications, and it is possible that it could be positive because of a floor effect for prototype distortion (i.e., a single modification is almost as damaging as two). This alternative explanation is unlikely, given the nearly linear distortion effect observed in experiment 1 (also, a floor effect predicts that [A.sub.P] should be positive, but the observed value was negative). Nevertheless, as a check against such a spuriously positive outcome, a second contrast representing a crossover-interaction effect of unity was computed at the individual level, [U[prime].sub.2] = [y.sub.--] min{[y.sub.+-], [y.sub.-+]}. If there is no true unity effect and [U.sub.2] is positive because of a floor effect (or more generally because of nonlinearity in the effect of distortion), then [U[prime].sub.2] should be negative or zero. Another benefit of analyzing [U[prime].sub.2] is that it eliminates certain potential aggregation biases. In particular, mean ratings could exhibit a crossover interaction consistent with a unity effect even though only floor effects were present at the individual level (see Hutchinson, Kamakura, and Lynch [1997] for a more detailed discussion).

[Expanded Picture] Prototypicality. The presence of a prototypicality effect (i.e., Hypothesis 1a) was supported by significant main effects for modification of part 1 and modification of part 2 (F(1, 184) = 23.97, p [less than] .0001, and F(1, 184) = 11.86, p [less than] .0001, respectively). Also, the mean value of [P.sub.2] was .44, which was significantly greater than zero (t(184) = 6.04, p [less than] .001).

Unity. The unity effect (i.e., Hypothesis 2a) was supported by significant interaction between the modification of part 1 and modification of part 2 (F(1, 184) = 338.2, p [less than] .0001). Also, the mean value of [U.sub.2] was 1.05, which was significantly greater than zero (F(1, 184) = 338.2, p [less than] .0001; and the mean value of [U[prime].sub.2] was 1.48, which was significantly greater than zero (F(1, 184) = 473.5, p [less than] .0001).

Other Effects and Follow-Up Tests. Significant main effects were also observed for product and type of modification (F(8, 1,472) = 40.78, p [less than] .0001, and F(2, 1,472) = 24.91, p [less than] .0001, respectively).(10) Also, although no significant main effects were observed for the remaining design variables (i.e., version and order), several significant higher-order interactions were observed, including product x modification of part 1 (F(8, 1,472) = 24.28, p [less than] .0001), product x modification of part 2 (F(8, 1,472) = 22.49, p [less than] .0001), and product x modification of part 1 x modification of part 2 (F(8, 1,472) = 53.26 p [less than] .0001). Given the possibility that these interactions might compromise the observed effects supporting Hypotheses 1a and 2a, follow-up tests were conducted.

Table 4 presents mean visual attractiveness ratings, [P.sub.2], [U.sub.2], and [U[prime].sub.2] for each stimulus set (i.e., each combination of product and type of modification). The prototypicality effect was tested separately for each stimulus set using [P.sub.2]. In support of the prototypicality effect, this difference was significantly greater than zero for 13 of the 27 stimulus sets and was never significantly less than zero (although the sign was negative for 11 stimulus sets). Thus, the observed interactions mainly reflected variations in the size of the prototypicality effect. However, the relatively frequent occurrence of reversals (even though they were never statistically significant) suggests that novelty effects might sometimes occur.

[Expanded Picture] The unity effect was tested separately for each stimulus set through [U.sub.2] and [U[prime].sub.2]. [U.sub.2] was significantly greater than zero for 21 of the 27 stimulus sets, and [U[prime].sub.2] was significantly greater than zero for 25 stimulus sets (see Table 4). Neither measure was ever significantly less than zero, and a negative value was observed for only one stimulus set. Thus, the unity effect was strongly confirmed, and the observed interactions reflected variations in the size of the effect.

[Expanded Picture] EXPERIMENT 3

In experiment 2, product designs were evaluated in a comparative context. That is, different product variations were viewed and rated simultaneously. As discussed earlier, this context is similar to shopping situations (e.g., retail stores, catalogs) in which competitive product offerings in the same category are displayed together, and, therefore, differences in visual aspects of product design are easily apprehended. However, in some cases products are evaluated individually (e.g., advertising, usage situations). Experiment 3 is a between-subjects replication of experiment 2 in which subjects evaluate only one design from each product class (i.e., a noncomparative context).

Method

Experimental Design. To simplify the task, a subset of the stimuli that were employed in experiment 2 were used in this experiment. One modification type was chosen for each product category. Although all nine product categories were included, the bathroom scales stimulus set was used as an initial warm-up item (and was omitted from subsequent analyses) for two reasons. First, the between-subjects design required a number of stimulus sets that was divisible by four. Second, this stimulus set produced the weakest unity effects in experiment 2. The eight experimental stimulus sets that were employed were clock/shape, dresser/trim, flashlight/trim, hair dryer/texture, lamp/texture, refrigerator/shape, telephone/shape, and TV remote control/trim. Each of these sets exhibited a significant unity effect in experiment 2, although the size of that effect varied (e.g., [U.sub.2] ranged from .44 to 3.06).

[Expanded Picture] The overall design of the experiment was a 2 (modification of part 1)x 2 (modification of part 2) x 8 (product) x 4 (version) x 2 (order) mixed-factorial Latin-square design. Just as in experiment 2, the modifications of parts 1 and 2 resulted in orthogonal manipulations of prototype distortion and unity. Because only one stimulus design per category was presented to each subject, there were four versions of the booklet that contained the stimulus items and their associated response scales. Thus, each version was a one-quarter fraction of the 2 (modification of part 1) x 2 (modification of part 2) x 8 (product) stimulus design. The randomized order in which the stimulus products were presented was reversed for half of the subjects.

Procedure. Two-hundred and forty volunteer subjects enrolled in an introductory marketing course participated in the experiment. The procedure was essentially the same as the procedure for experiment 2.

Results

Data were analyzed by a mixed ANOVA with between-subjects factors of version and order. Product, modification of part 1, and modification of part 2 were within-subjects factors. Only product ratings were analyzed because the between-subjects design precluded the computation of product specific contrasts at the individual level. [P.sub.3] and [U.sub.3] were defined similarly to [P.sub.2] and [U.sub.2] but were calculated by using mean product ratings at the aggregate level and were tested by using appropriate ANOVA contrasts.

Prototypicality. As in experiment 2, the presence of a prototype-distortion effect was supported directionally by the effects of modifications of part 1 and modifications of part 2; however, only modifications of part 2 reached traditional levels of statistical significance (F(1, 230) = 1.44, p = .23, and F(1,230) = 10.38, p [less than] .01, respectively). The value of [P.sub.3] was significantly greater than zero (t(230) = 3.14, p [less than] .0019); however, the absolute size was somewhat smaller (i.e., [P.sub.3] = .33 vs. .44 in experiment 2).

Unity. As in experiment 2, the unity effect was supported by a significant interaction between the modifications of part 1 and part 2 (F(1,230) = 5.88, p [less than] .02). As with prototype distortion, the absolute size of the unity contrast was smaller in experiment 3 than in experiment 2 (i.e., [U.sub.3] = .19 vs. 1.12).

Other Effects and Follow-Up Tests. A significant main effect was observed for product (F(7, 1,610) = 42.51, p [less than] .0001). Also, there was a significant main effect of order (F(1,230) = 6.89, p [less than] .01) and a significant interaction between product and order (F(7, 1,610) = 2.24, p [less than] .05) that were not easily interpreted but are unrelated to our hypotheses. Because of the Latin-square design of the experiment, it was not possible to compute the traditional test for the interactions of product with the distortion and unity effects. Therefore, follow-up tests were conducted.

Table 5 presents mean visual attractiveness ratings, [P.sub.3], and [U.sub.3] for each stimulus set. The distortion effect, [P.sub.3], was tested separately for each stimulus set. It was positive for six of eight stimulus sets, and two reached traditional levels of statistical significance. It was never significantly negative. Thus, even though the distortion effect was not exceptionally strong, serious interactions with product were not present. (Recall that the overall effect was significant.)

The unity effect was tested separately for each stimulus set using [U.sub.3]. [U.sub.3] was greater than zero for six of the eight stimulus sets and reached statistical significance for two. It was never significantly negative. Thus, as with the prototype distortion effect, the unity effect was not exceptionally large, but it was fairly robust and significant reversals did not occur.

EXPERIMENT 4

Experiment 4 extends the earlier results and tests the robustness of prototype distortion and unity effects. In this experiment, subjects were provided with written product descriptions as well as drawings of the products that they were asked to rate. The evaluation task (i.e., liking or quality ratings) did not explicitly require consideration of product appearance. The experiment examined whether or not there were moderating effects of task or additional information on the influence of prototype distortion and unity. We wish to demonstrate that distortion and unity effects are not limited to situations in which only visual information is available or relevant for the task at hand. Thus, we expect these effects to persist in judgments of liking and even when visual appearance plausibly might be deemed irrelevant for the task at hand (i.e., for quality judgments; also see Hutchinson and Alba 1991). These predictions are formalized in the following hypotheses.

H8a: The prototype distortion effect persists in attitude and belief judgments about product designs that combine aesthetic and nonaesthetic information.

H8b: The unity effect persists in attitude and belief judgments about product designs that combine aesthetic and nonaesthetic information.

Method

[Expanded Picture] Experimental Design. The same subset of product drawings that was used in experiment 3 was used in this experiment. Thus, the manipulations of unity and prototype distortion were the same as before. In this study, however, each of three target product designs was shown with a written product description and was rated against a control product. The targets were the two unified designs (i.e., the prototype and the dual-modification design, exemplified by designs A and D in ) and one of the (nonunified) single-modification designs (e.g., design B or C in ). The control product was the other single-modification design and was chosen to be the higher rated of the two in experiment 3. The product descriptions for the control product dominated target product descriptions (i.e., they were more favorable for. one or two attributes and the same for all other attributes). This avoided ceiling effects that might have occurred if targets were superior on both aesthetic and nonaesthetic factors. The product descriptions were developed from catalog descriptions and were pilot tested (without the product drawings) to confirm that the control descriptions were preferred to target descriptions (t(6) = 2.89, p [less than] .05 and t(10) = 8.49, p [less than] .01 for the difference in mean liking ratings and mean quality ratings, respectively). These product descriptions are presented in the Appendix.

In this experiment, subjects compared one target/control pair from each product class (i.e., nine pairs). A Latin-square design was used to construct three versions of the questionnaire such that in each version each of the three levels of product distortion occurred three times, and across versions all combinations of distortion and product class were represented. Thus, product and target were within-subjects factors. The response measure was either a scale measuring general attitude (like/dislike) or a scale measuring quality (high quality/low quality). Subjects received only one of the two scales for all of the products that they rated. The position of the target and control products on each page (left or right side) and the order of the nine pairs were reversed for half of the subjects. Thus, version (three levels) and order/position (two levels) were between-subjects factors.

Procedure. Two-hundred and fifty-seven subjects participated in the experiment. The procedure was very similar to the procedures used in experiments 2 and 3 except that in this study subjects were presented with two product versions (i.e., control and target) on each page with the response scale provided below each description. All conditions of this design were run concurrently.

Results

The dependent measure for each target was a difference score (i.e., target rating minus control rating; e.g., [d.sub.++] = [y.sub.++] - [y.sub.-+], [d.sub.+-] = [y.sub.+-] - [y.sub.-+], and [d.sub.--] = [y.sub.--] - [y.sub.-+], assuming that the -+ design was the control stimulus). This removed individual differences in response bias and/or preference for a given product class (which otherwise would have contributed to the error terms of the analyses). [P.sub.4] was defined as in the earlier experiments except that difference scores rather than ratings were used (i.e., [P.sub.4] = [d.sub.++] - [d.sub.--]). In this design it is not possible to define an interaction contrast for unity analogous to [U.sub.2] and [U.sub.3]; therefore, a contrast similar to [U[prime].sub.2] was defined as the difference in the difference scores between the dual- and single-modification targets (i.e., [U[prime].sub.4] = [d.sub.--] - [d.sub.+-]). Mean difference scores for each target, [P.sub.4], and [U[prime].sub.4] are provided in Table 6. Note that all mean difference scores were values in the neighborhood of - 1. Given that a nine-point scale was used, this indicates that, on average, the verbally presented product features influenced ratings slightly more than did aesthetic aspects of the line drawings. Thus, from the perspective of experimental design, our stimuli were appropriately calibrated and avoided potential floor or ceiling effects that would have occurred had either the verbal or the visual information exerted a dominant influence on the ratings.

Liking. A mixed-design ANOVA indicated that there were no main effects of version or order, but the main effect of target was statistically significant (F (2, 242) = 11.3, p [less than] .0001). Separate linear contrasts were used to test [P.sub.4] and [U[prime].sub.4]. Only [U[prime].sub.4] was significantly different from zero (F (1, 121) = 15.5, p [less than] .0001), supporting Hypothesis 8b. The main effect of product was also significant (F (8, 968) = 7.31, p [less than] .0001). The results of tests of [P.sub.4] and [U[prime].sub.4] for each product class are given in Table 6. In summary, [P.sub.4] was mixed in sign, significantly greater than zero twice and significantly less than zero once; [U[prime].sub.4] was positive for all product classes and was significantly greater than zero for two.

Quality. A mixed-design ANOVA indicated that the main effect of target was statistically significant (F (2, 248) = 4.15, p [less than] .02). Separate linear contrasts were used to test [P.sub.4] and [U[prime].sub.4]. Again, only [U[prime].sub.4] was significantly different from zero (F (1,124) = 7.79, p [less than] .01), supporting Hypothesis 8b. The main effects of version and product were also significant (F (2, 124) = 3.87, p [less than] .05 and F (8, 992) = 21.4, p [less than] .0001, respectively). The results of tests of [P.sub.4] and [U[prime].sub.4] for each product class are given in Table 6. In summary, [P.sub.4] was again mixed in sign, significantly greater than zero once and significantly less than zero once; [U[prime].sub.4] was positive for eight of nine product classes and was significantly greater than zero for one.

Summary. Experiment 4 examined the influence of unity and prototype distortion on nonaesthetic responses to products. The results presented here indicated that unity influenced both preference and perceived quality despite the availability of written attribute information, supporting Hypothesis 8b. The effect of prototype distortion on these measures was not reliable. Therefore, Hypothesis 8a was not supported.

GENERAL DISCUSSION

The results of these experiments provide strong evidence for positive effects of prototypicality and unity on aesthetic response. These results replicate prior research on the effect of prototypicality on affective response. More important, they provide the first clear demonstration of a unity effect that is not confounded with other factors and suggest that this effect is fairly general with respect to product class (especially experiment 2) and the presence of verbal information (experiment 4) whenever direct visual comparisons are easy for subjects to make. The effect of unity is greatly reduced when direct visual comparison is prevented (experiment 3).

[Expanded Picture] Experiment 1 showed that prototype distortion exerted a nearly linear negative effect (i.e., a positive mean for [P.sub.1] and a small negative mean for [A.sub.P]) on both aesthetic response and perceived typicality. The linearity of this effect is consistent with our characterization of distortion as an atomistic visual property. Unity was found to exert a superadditive positive effect on both aesthetic response and perceived typicality. The superadditivity of this effect is consistent with our characterization of unity as a relational visual property. Two-stage least-squares analyses suggested that unity and prototype distortion exert direct effects on aesthetic response in addition to indirect effects that are mediated by perceived typicality. None of the reported experiments or analyses provided evidence supporting the hypothesis that people prefer moderate levels of typicality (or schema congruity).

The finding that direct visual comparison is necessary for strong unity effects has pragmatic implications. Many shopping environments display deep assortments of similar products, and most stores organize their merchandise by product category. Thus, products are evaluated alongside their competitors. This is also true when consumers collect brochures in order to directly compare products, and most catalogs display merchandise in a comparative format. Insofar as aesthetic aspects of product designs are important to consumers, shopping formats such as these are useful because they facilitate the appreciation of visual differences.

[Expanded Picture] Finally, these results begin to address a major need in the development of new products. This need is the ability to predict how consumers will react to visual design characteristics. Traditional methods (e.g., conjoint analysis) seem adequate for assessing the marginal effects of these characteristics. However, the sheer number of possible interactions between characteristics makes direct measurement an impossibility. Thus, the establishment of empirically and theoretically sound principles, such as unity and prototypicality, that predict these interactions should provide a useful tools for product designers.

1 Although the term "unity" appears in the design literature, "prototype distortion" originated in cognitive psychology and is not commonly discussed as a design principle, per se. However, in practice designers are influenced by existing products (e.g., they may start with existing designs that they subsequently modify). Our primary goal is to distinguish prototype distortion, as a role of thumb, from perceived typicality.

2 The phrase "distance from prototype" is also frequently used; however, "distortion" conveys the operationalization process more clearly.

3 They use the term "attributional" rather than "atomistic." We have not adopted that terminology here because of its other uses in consumer research and social psychology.

4 A full set of stimuli is available from the first author upon request.

5 Chronologically, experiment 1 was the last to be conducted; however, it is reported first for expositional and conceptual clarity. These six product classes were used in all four experiments.

[Expanded Picture] 6 We thank the area editor for suggesting the form of [A.sub.U] that was used for this experiment (see Table 1). This contrast tests the difference between partially unified stimuli and the appropriately weighted mean for the completely unified and completely nonunified stimuli. This value should be zero if the unity effect is additive.

7 Given our earlier results showing effect size differences across product classes and the natural expectation that some modifications would have larger effects than others, the first-stage ordinary least-squares (OLS) regression included linear and quadratic components for prototype distortion and unity and their interactions with product and deck. Main effects of subject and product were removed from the data by normalization for both stages. Thus, the first-stage model has sufficient degrees of freedom to avoid identification problems in the second stage. It is also important to note that the independent variables, predicted visual attractiveness and predicted goodness of example, are population estimates, not individual-level observations. This is consistent with prototype distortion and unity, whose effects (like all ANOVA fixed effects) are assumed to be homogeneous across subjects.

8 Linear and quadratic components of prototype distortion and unity were included in the same model because, unlike the ANOVA contrasts, [A.sub.P] and [A.sub.U], the two quadratic components are somewhat correlated with each other and with unity even after zero-centering of the original variables because of the oversampling of unified stimuli in our design. The quadratic is used rather than the additivity test contrast because we are potentially interested in the shapes of the functions, not just testing for a violation of additivity.

9 As part of an exploratory study on the influence of aesthetic factors on other product perceptions, subjects were also asked to rate each product design on scales that measured nonaesthetic evaluations: familiarity (familiar/unfamiliar), attitude (like/dislike), quality (high quality/low quality), and price (high price/low price). These data are not reported in detail in this article. Because multiple scales were used, visual attractiveness was measured with a simple bipolar scale instead of the more extensive method of experiment 1.

10 In this design, the main effect of modification type is examined with a Latin-square subset of the complete five-factor design. This is consistent with the treatment suggested by Winer, Brown, and Michels (1991, pp. 706-711) for Latin squares and related designs.

REFERENCES

Baron, Reuben M. and David A. Kenny (1986), "The Moderator-Mediator Variable Distinction in Social Psychological Research: Conceptual, Strategic, and Statistical Considerations," Journal of Personality and Social Psychology, 51 (6), 1173-1182.

Barsalou, Lawrence W. (1985), "Ideals, Central Tendency, and Frequency of Instantiation as Determinants of Graded Structure," Journal of Experimental Psychology: Learning, Memory, and Cognition, 11 (October), 629-654.

Bell, Stephen S., Morris B. Holbrook, and Michael R. Solomon (1991), "Combining Esthetic and Social Value to Explain Preferences for Product Styles with the Incorporation of Personality and Ensemble Effects," Journal of Social Behavior and Personality, 6 (6), 243-273.

Berkowitz, Marvin (1987), "The Influence of Shape on Product Preferences," in Advances in Consumer Research, Vol. 14, ed. Melanie Wallendorf and Paul Anderson, Provo UT: Association for Consumer Research, 559.

Berlyne, Daniel E. (1970), "Novelty, Complexity, and Hedonic Value," Perception and Psychophysics, 8 (November), 279-286.

----- (1971), Aesthetics and Psychobiology, New York: Appleton-Century Crofts.

----- (1974), Studies in the New Experimental Aesthetics, New York: Wiley.

Bloch, Peter H. (1995), "Seeking the Ideal Form: Product Design and Consumer Response," Journal of Marketing, 59 (July), 16-29.

Carpenter, Gregory S. and Kent Nakamoto (1989), "Consumer Preference Formation and Pioneering Advantage," Journal of Marketing Research, 26 (August), 285-298.

Ching, Francis D. K. (1979), Architecture: Form, Space, and Order, New York: Van Nostrand Reinhold.

Dickson, Peter R. and James L. Ginter (1987), "Market Segmentation, Product Differentiation, and Marketing Strategy," Journal of Marketing, 51 (April), 1-10.

Drolet, Aimee, Punam Anand Keller, and Robert Zajonc (1996), "Category Congruence and Preference Formation," working paper, Marketing Department, Stanford University, Stanford, CA 94305.

Eisenman, Russell (1968), "Novelty Ratings of Simple and Complex Shapes," Journal of General Psychology, 78 (April), 275-278.

Garner, Wendell R. (1976), "Interaction of Stimulus Dimensions in Concept and Choice Processes," Cognitive Psychology, 8 (January), 98-123.

Garvin, David A. (1984), "What Does 'Product Quality' Really Mean?" Sloan Management Review, 26 (Fall), 25-43.

Goldstone, Robert L., Douglas L. Medin, and Dedre Gentner (1991), "Relational Similarity and the Nonindependence of Features in Similarity Judgements," Cognitive Psychology, 23 (April), 222-262.

Gordon, Peter C. and Keith J. Holyoak (1983). "Implicit Learning and Generalization of the 'Mere Exposure' Effect," Journal of Personality and Social Psychology, 45 (September), 492-500.

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

----- and V. Srinivasan (1990), "Conjoint Analysis in Marketing: New Developments with Implications for Research and Practice," Journal of Marketing, 5 (October), 3-19.

Greene, William H. (1993), Econometric Analysis, New York: Macmillan.

Griffin, Abbie and John R. Hauser (1993), "The Voice of the Customer," Marketing Science, 12 (Winter), 1-27.

Hauser, John R. and Don P. Clausing (1988), "The House of Quality," Harvard Business Review, 66 (May-June), 63-73.

Holbrook, Morris B. (1995), "An Empirical Approach to Representing Patterns of Consumer Tastes, Nostalgia, and Hierarchy in the Market for Cultural Products," Empirical Studies of the Arts, 13 (1), 55-71.

----- and Elizabeth Hirschman (1982), "The Experiential Aspects of Consumption: Consumer Fantasies, Feelings, and Fun," Journal of Consumer Research, 9 (September), 132-140.

----- and Robert B. Zirlin (1985), "Artistic Creation, Artworks, and Aesthetic Appreciation: Some Philosophical Contributions to Nonprofit Marketing," Advances in Nonprofit Marketing, 1, 1-54.

Holt, Steven (1985), "Design, the Ninth Principle of Excellence: The Product Half of the Business Equation," Innovation, 4 (Fall), 2-4.

Homa, Donald (1984), "On the Nature of Categories," in The Psychology of Learning and Motivation, Vol. 18, ed. Gordon H. Bower, New York: Academic Press, 49-94.

Hutchinson, J. Wesley (1986), "Discrete Attribute Models of Brand Switching," Marketing Science, 5 (Fall), 350-371.

----- and Joseph W. Alba (1991), "Ignoring Irrelevant Information: Situational Determinants of Consumer Learning," Journal of Consumer Research, 18 (December), 325-346.

-----, Wagner Kamakura, and John G. Lynch (1997), "Unobserved Heterogeneity as an Alternative Explanation for 'Reversal' Effects in Behavioral Research," working paper, Marketing Department, University of Pennsylvania, Philadelphia, PA 19104.

James, Lawrence R. and B. Krishna Singh (1978), "An Introduction to the Logic, Assumptions, and Basic Analytic Procedures of Two-Stage Least squares," Psychological Bulletin, 85 (September), 1104-1122.

Katz, David (1950), Gestalt Psychology, New York: Ronald Press.

Kemler Nelson, Deborah G. (1984), "The Effect of Intention on What Concepts Are Acquired," Journal of Verbal Learning and Verbal and Verbal Behavior, 23,734-759.

Koffka, Kurt (1935), Principles of Gestalt Psychology, New York: Harcourt, Brace.

Kotler, Philip and G. Alexander Rath (1984), "Design: A Powerful but Neglected Strategic Tool," Journal of Business Strategy, 5 (Fall), 16-21.

Kunst-Wilson, William R. and Robert B. Zajonc (1980), "Affective Discrimination of Stimuli That Cannot Be Recognized," Science, 207 (February), 557-558.

Langlois, Judith H. and Lori A. Roggman (1990), "Attractive Faces Are Only Average," Psychological Science, 1 (March), 115-121.

Lauer, David A. (1979), Design Basics, New York: Holt, Rinehart, & Winston.

Lennon, Sharron J. (1990), "Effects of Clothing Attractiveness on Perceptions," Home Economics Research Journal, 18 (3), 303-310.

Loken, Barbara and James Ward (1990), "Alternative Approaches to Understanding the Determinants of Typicality," Journal of Consumer Research, 17 (September), 111-126.

Lombardo, Sergio (1991), "Event and Decay of the Aesthetic Experience," Empirical Studies of the Arts, 9 (2), 123-141.

Mandler, George (1982), "The Structure of Value: Accounting for Taste," in Affect and Cognition: The Seventeenth Annual Carnegie Symposium, ed. Margaret S. Clark and Susan T. Fiske, Hillsdale, NJ: Erlbaum, 3-36.

Martindale, Colin and Kathleen Moore (1988), "Priming, Prototypicality, and Preference," Journal of Experimental Psychology: Human Perception and Performance, 14 (November), 661-670.

-----, Kathleen Moore, and Alan West (1988), "Relationship of Preference Judgments to Typicality, Novelty, and Mere Exposure," Empirical Studies of the Arts, 6 (1), 79-96.

-----, Kathleen Moore, and Jonathan Borkum (1990), "Aesthetic Preference: Anomalous Findings for Berlyne's Psychobiological Theory," American Journal of Psychology, 103 (1), 53-80.

McAlister, Leigh and Edgar Pessemier (1982), "Variety Seeking Behavior: An Interdisciplinary Review," Journal of Consumer Research, 9 (December), 311-322.

Medin, Douglas L., Robert L. Goldstone, and Dedre Gentner (1993), "Respects for Similarity," Psychological Review, 100 (April), 254-278.

----- and Marguerite M. Schaffer (1978), "Context Theory of Classification Learning," Psychological Review, 85 (May), 207-238.

----- and Edward E. Smith (1984), "Concepts and Concept Formation," Annual Review of Psychology, 35, 113-138.

Meyers-Levy, Joan and Alice M. Tybout (1989), "Schema Congruity as a Basis for Product Evaluation," Journal for Consumer Research, 16 (June), 39-54.

Murphy, Gregory L. and Douglas L. Medin (1985), "The Role of Theories in Conceptual Structure," Psychological Review, 92 (July), 289-316.

Nedungadi, Prakash and I. Wesley Hutchinson (1985), "The Prototypicality of Brands: Relationships with Brand Awareness, Preference and Usage," in Advances in Consumer Research, Vol. 12, ed. Elizabeth C. Hirschman and Morris Holbrook, Provo UT: Association for Consumer Research, 498-503.

Palmeri, Thomas J. and Robert M. Nosofsky (1995), "Recognition Memory for Exceptions to the Category Rule," Journal of Experimental Psychology, 21 (May), 548-568.

Reed, Stephen K. (1972), "Pattern Recognition and Categorization," Cognitive Psychology, 3 (July), 382-407.

Rosch, Eleanor (1975), "Cognitive Representations of Semantic Categories," Journal of Experimental Psychology: General, 104 (September), 192-233.

----- (1978), "Principles of Categorization," in Cognition and Categorization, ed. E. Rosch and B. B. Lloyd, Hillsdale, NJ: Erlbaum, 27-47.

Stangor, Charles and David McMillan (1992), "Memory for Expectancy-Congruent and Expectancy-Incongruent Information: A Review of the Social and Social Developmental Literatures," Psychological Bulletin, 111 (January), 42-61.

Veryzer, Robert (1993a), "Aesthetic Response and the Influence of Design Principles on Product Preferences," in Advances in Consumer Research, Vol. 20, ed. Leigh McAlister and Michael L. Rothschild, Provo, UT: Association of Consumer Research, 224-228.

[Expanded Picture] ----- (1993b), "The Influence of Unity and Prototypicality on Aesthetic Responses to New Product Designs," unpublished dissertation, Marketing Department, College of Business Administration, University of Florida, Gainesville, FL 32611.

Ward, Thomas B. and James Scott (1987), "Analytic and Holistic Modes of Learning Family-Resemblance Concepts," Memory & Cognition, 15 (1), 42-54.

Whitney, Daniel E. (1988), "Manufacturing by Design," Harvard Business Review (July-August), 83-90.

Winer, B.J., Donald R. Brown, and Kenneth M. Michels (1991), Statistical Priciples in Experimental Design, New York: McGraw-Hill.

Wohlwill, Joachim F. (1981), "A Conceptual Analysis of Exploratory Behavior: The 'Specific-Diverse' Distinction Revisited," in Advances in Intrinsic Motivation and Aesthetics, ed. Hy I. Day, New York: Plenum, 341-364.

Woll, Stanley and Arthur Graesser (1982), "Memory Discrimination for Information Typical and Atypical of Person Schemata," Social Cognition, 1 (42), 287-310.

Zeithaml, Valarie A. (1988), "Consumer Perceptions of Price, Quality, and Value: A Means-End Model and Synthesis of Evidence," Journal of Marketing, 52 (July), 2-22.

Robert W. Veryzer, Jr., is assistant professor of marketing at Rensselaer Polytechnic Institute, Troy, NY 12180. J. Wesley Hutchinson is associate professor of marketing at the University of Pennsylvania, Philadelphia, PA 19104. Experiments 2-4 were conducted as part of the first author's dissertation at the University of Florida. The authors thank Eric Bradlow, Aimee Drolet, Eric Greenleaf, Steve Hoch, Chris Janiszewski, Richard Lutz, John Lynch, Jr., David Mick, Brian Sternthal, and the reviewers for their comments and advice.

[Expanded Picture] [Expanded Picture] [Expanded Picture] [Expanded Picture] [Expanded Picture]