Although advertising persuades through overt appeals to reason or emotion, we focus on the indirect process by which advertising influences the interpretation ...
Ambiguity, Processing Strategy, and Advertising-Evidence Interactions YOUNG-WON HA STEPHEN J. HOCH* Although advertising persuades through overt appeals to reason or emotion, we focus on the indirect process by which advertising influences the interpretation of objective product evidence. We investigate how two factors moderate advertisingevidence interactions: the ambiguity of the evidence and consumer information processing strategies. We provide a theoretical account of ambiguity, identifying structural characteristics that render evidence about product quality open to either one or multiple interpretations. In our first experiment, the ambiguity of a decision environment played a key role in determining the effect of advertising on product quality perception. In our second experiment, different information processing strategies influenced advertising's effects on interpretation of the evidence.
M
OSt experimental research on advertising has followed the classic persuasion tradition, examining how factors like source credibility and various message variables directly influence the persuasiveness of an appeal. Some recent research (Deighton 1984; Hoch and Ha 1986) has focused on the indirect effects of advertising, examining how advertising influences the way consumers interpret objective evidence about product quality. Research on advertising-evidence interactions (Deighton 1984) has shown that even when un persuasive in isolation, advertising can alter the interpretation of product evidence through confirmatory or sufficiency hypothesis testing (Klayman and Ha 1987). However, Hoch and Ha (1986) found that advertising-evidence interactions were limited to cases in which the objective evidence was ambiguous, that is, open to multiple interpretations. Advertising had no effect on the interpretation of unambiguous evidence. In this article, we offer an information-theoretic account of ambiguity and identify the structural characteristics that render evidence concerning product quality open to either one (unambiguous) or multiple (ambiguous) interpretations. We use this framework to construct multiattribute stimuli varying in ambi-
guity and show that both ambiguity and type of information processing strategy moderate advertising-evidence interactions.
AMBIGUITY AND INFORMATION STRUCTURE
Determinants of Product Ambiguity We define ambiguity as "the potential for multiple interpretations" of overall product quality (Hoch and Ha 1986). When making global evaluations of a set of alternatives, we assume that people go through a three-stage process: (1) identifying relevant attributes for consideration, (2) evaluating the level of each attribute, and (3) combining this information to form an overall evaluation of each alternative. Each stage can foster ambiguity. If consumers selectively encode different subsets of attributes from occasion to occasion, whether because oflittle product expertise or excessive information load, multiple interpretations are possible (Shepard 1964). For many products, consumers may have difficulty evaluating attribute levels, either because the attribute is fuzzy or subjective (Sabini and Silver 1982) or entangled with other attributes. Our empirical work focuses on the third stage. Even when faced with alternatives that are already partitioned into attributes with clearly specified levels (.e.g ..' information display boards), ambiguity may anse If consumers cannot apply a consistent information integration strategy (Shepard 1964). Ambiguity at the integration stage is most likely to arise when the consumer faces uncorrelated attributes. Positive attribute collinearity reduces the di-
'Young-Won Ha is Assistant Professor of Marketing at the College of Business Administration, Sogang University, Seoul 121, Korea. Stephen J. Hoch is Professor of Marketing and Behavioral Science at the Graduate School of Business, Center for Decision Research, University of Chicago, Chicago, IL 60637. The research was supported in part by the Bozell, Jacobs, Kenyon, & Eckhardt Endowment Fund at the University of Chicago Graduate School of Business.
354 © JOURNAL OF CONSUMER RESEARCH. Vol. 16. December 1989
355
AMBIGUITY/ADVERTISING-EVIDENCE INTERACTIONS
mensionality of the information space, thereby decreasing information load and task difficulty. Any unreliability in attribute valuation (the second stage) is mitigated because there are multiple measures of the same underlying dimension. Decision conflict is minimized because no attribute trade-offs are required. Global evaluations also are relatively insensitive to the selection of different subsets of attributes and/or inconsistent weighting policies (Einhorn and Hogarth 1975). High cue redundancy (positive multicollinearity) results in a narrower range of interpretation of anyone brand, leading to less ambiguity. When attributes are uncorrelated or negatively correlated with each other, information processing demands will be much greater (Shugan 1980). Negative intercorrelations result in Pareto optimal choice sets; selecting an alternative on the efficient frontier requires multiple trade-offs. If the consumer does not exhaustively process the information or experiences slight fluctuations in attention (Reisberg and O'Shaughnessy 1984), changes in evaluations occur because of greater sensitivity to the magnitude of individual attribute weights. Figure A offers a simple illustration of how cue redundancy affects the potential for multiple interpretations in product sets consisting of three products described on two attributes. Utility is monotonically increasing in both attribute at and a2. In the ambiguous set, at and a2 are negatively correlated, while in the unambiguous set they are positively correlated. Imagine two different families ofindifference curves where points to the northeast are more preferred; at is more important given the solid line indifference relation, and a2 is more important given the dashed line. Further assume that exogenous factors (advertising) can influence the relative weighting of at and a2. In the ambiguous set, attribute weights really matter. With indifference curves parallel to the solid line, the order of the alternatives would be (C > B> A); indifference curves parallel to the dashed line would result in the reverse ordering (A > B> C). In the unambiguous set, both indifference relations would produce the same ordering of the alternatives (C > B> A). Ambiguity arising during attribute identification and attribute evaluation can also be incorporated into Figure A. In a case with only two attributes, selective perception of attributes (in the extreme) would result in subjects' relying solely on at or a2, with indifference curves parallel to either the x (A > B > C) or y (C > B > A) axes. Ambiguity arising in the attribute evaluation stage is represented by the elliptical distributions of attribute values around each of the brand points in Figure A, the shape and orientation of the ellipse capturing the relative fuzziness of the attributes. Depending on where the consumer locates the brand within the ellipse, multiple interpretations of quality could result.
FIGURE A A SIMPLIFIED REPRESENTATION OF AMBIGUOUS AND UNAMBIGUOUS STIMULUS SETS
Ambiguous product set Attribute 2
~.~~ u(a1) < u(a2)
••••••••••••
"".""
........ Attribute 1
Unambiguous product set Attribute 2
Attribute 1
356
THE JOURNAL OF CONSUMER RESEARCH TABLE 1 DIFFERENT FORMS OF EVIDENCE FOR THE 19-INCH, COLOR TV CATEGORY Adjacent channel rejection
Freedom from geometric distortion
Observed picture quality
Serviceability
Tone quality
Resolution"
Fringe VHF
Reception UHF
Warrantyb
Ambiguous evidence A (Magnavox) B (Panasonic) C(Quasar) D(RCA) E(Sharp) F(Sony) G (Zenith) Target (Samsung)
G G F VG F F VG G
G VG G G G G F F
VG G VG VG G VG VG VG
G VG G G F F G VG
G F G G G VG G G
260 300 275 270 300 320 300 300
VG F VG G VG G G G
F VG F F VG G F G
12/3/24 12/12/24 24/24/24 12/3/24 12/3/30 12/3/24 24/6/24 12/12/24
Unambiguous evidence A (Magnavox) B (Panasonic) C (Quasar) D(RCA) E (Sharp) F (Sony) G (Zenith) Target (Samsung)
P G F G F VG VG G
F F F G F VG G F
F G F VG F E VG VG
G G G VG G VG VG VG
F G F G F E G G
260 300 275 300 270 330 320 300
F G G G F VG VG G
P F F G F E VG G
12/3/24 12/6/24 12/3/24 12/12/24 12/3/24 24/24/24 12/12/24 12/12/24
Brand
• Resolution: lines/screen. b Warranty: parts/labor/picture tube (in months). NOTE: P = Poor; F = Fair; G = Good; VG = Very Good; E = Excellent.
Constructing Ambiguous and Unambiguous Stimuli Using this information-theoretic approach, we can design product categories that vary in ambiguity (i.e., that have the potential to support multiple interpretations of product quality) by pairing multiattribute alternatives that. either contain dominated alternatives (unambiguous evidence) or all fall on an efficient frontier (ambiguous evidence). The product category selected was 19-inch color televisions (see Table 1). Unambiguous evidence was created by manipulating the levels of the attributes to establish a clear dominance hierarchy among the brands (i.e., F > G > D and so on). Dominance produces high levels of cue redundancy (average attribute intercorrelations of more than 0.8), providing an information environment in which the consumer is not required to make any attribute trade-offs (see Table 2). With high multicollinearity, all attributes provide virtually the same information (at least ordinally) about quality. It should be noted that dominance does not necessarily imply an inefficient market. With inclusion of an attribute like price that is negatively correlated with all other attributes, a Pareto optimal market could be preserved. Ambiguous evidence was created by manipulating the attribute ratings so that cue redundancy was very low. No alternative strictly dominated another (Pareto optimal), and alternatives varied little on impor-
tant attributes (e.g., observed picture quality). Table 2 shows that attribute intercorrelations were less than 0, and pairwise comparisons of all brands required multiple trade-offs between attributes. Although the alternatives were similar in terms of overall quality, each brand offered a distinctive attribute profile, providing different strengths and weaknesses. Multiple interpretations of overall quality were possible, depending on how subjects weighted individual attributes. The evidence consisted of eight brands described on nine attributes each, a total of 72 data points. The attribute information was displayed separately for each brand. To check whether the manipulation of ambiguity was successful, a "blindfold" test was conducted. In a pretest, subjects (n = 32) rated the overall quality of each brand using a 0 (low) to 10 (high) scale; the only procedural difference from the main experiment was that brand names were disguised (e.g., "Brand F" rather than Sony) to preclude the influence of prior opinions about the brands. After a one-week delay, the same subjects reevaluated the same materials. To measure the potential for multiple interpretations of quality, we calculated inter-judge and intrajudge reliabilities (Hoch and Ha 1986). As shown in Table 2, subjects showed much higher reliability when inspecting unambiguous compared to ambiguous evidence (both ps < 0.001). Subjects perceived clear quality differences between brands given unam-
357
AMBIGUITY /ADVERTISING-EVIDENCE INTERACTIONS TABLE 2 STIMULUS CHARACTERISTICS AND PRETESTING RESULTS FOR THE AMBIGUOUS AND UNAMBIGUOUS SETS OF EVIDENCE
Type of evidence Ambiguous Unambiguous
Average attribute intercorrelation"
-.11 .83
-.11 .87
Number of attribute trade-offs required
Intrajudge
335 0
.25 .91
Range of mean Interquality judge ratings
Reliability
-.03 .90
.71 6.65 b
• Cue redundancy; Pearson Spearman. b MANOVA F (5,80) = 123.0, P < 0.00001.
biguous evidence (p < 0.0001), but no differences given ambiguous evidence.
EXPERIMENT 1
Method This study examined how ambiguity in information structure moderates advertising-evidence interactions. We expected an advertising effect only when subjects were exposed to ambiguous evidence. With ambiguous evidence, we expected subjects to rely on top-down processing, assimilating the evidence with either their prior opinions or ad-induced expectations. With unambiguous evidence, we expected bottom-up processing to dominate; the clarity of the objective evidence would dominate any expectations. Both of these predictions follow from the work of Hoch and Ha (1986). The procedure was a pre-post design measuring changes in perceptions of quality. After providing pretest ratings, subjects saw storyboard copy test ads, then inspected product evidence (information on nine attributes for each of eight brands), and then reassessed product quality. Storyboard Ads. Print ads showed a black and white picture of the product and logo and 100 to 120 words of copy that stressed quality. The slogan for the target (advertised) brand was "To find the TV set that's right for you . . . Compare." The target was Samsung, at the time a relatively new Korean import. It was the lowest rated brand in a pretest (X = 5.5 compared to Sony X = 8.7), a stereotypical underdog. The first paragraph of the target ad emphasized the need for comparison without making explicit brand comparisons. "Buying a TV is one of the most important purchase decisions you'll make. And one way to make sure of making the right choice is to compare before you buy. Take a good look at Samsung and
take a good look at the competition. . . ." The remaining text was attribute-oriented but mentioned none ofthe attributes later provided as evidence. Product Evidence. The evidence consisted of eight brands, each described on nine attributes. Deighton (1984) provided subjects with an enormous amount of data, effectively forcing subjects to engage in selective processing. Because our theoretical development of ambiguity is based on the structure of information without recourse to the amount of information, we took a middle ground. Too much information might render any evidence ambiguous regardless of underlying structure. Subjects had two minutes to process 72 attributes. The attribute information appears in matrix form in Table 1, but subjects actually saw each brand separately. The position of the target brand was counterbalanced across subjects.
Procedure Subjects were 60 undergraduate and graduate students at the University of Chicago who were paid for their participation. There were two independent variables, manipulated according to a 2 X 3 between-subjects factorial: advertising (no ad versus ad) and ambiguity of evidence (no evidence control, ambiguous, unambiguous). Subjects were given the experimental materials in booklet form and told that the research was being conducted for a downtown Chicago marketing research firm interested in evaluating various advertising copy-testing and product-testing procedures. Subjects were told that they would engage in four separate tasks. First, to control for heterogeneity in prior attitudes, subjects provided II-point pretest quality ratings (0 = very low, 10 = very high). To limit unnecessary attention, the TV evidence was embedded among three other categories. Subjects were told about attributes that Consumer Reports considered important. Sub-
jects rated six out of the eight different brands constituting each category to minimize response overload. Second, subjects saw five different storyboard ads for 60 seconds each. In the conditions in which subjects saw the target ad, it appeared in the third position surrounded by filler ads for products not related to the experiment. To make the cover story consistent, subjects were told to evaluate each ad in terms of informativeness, believability, and copy flow. Third, subjects went into an adjacent room to inspect the evidence (eight brands by nine attributes) for two different product categories. Subjects in the no evidence control condition saw evidence for two unrelated categories, but subjects in the ambiguous and unambiguous conditions saw the TV evidence in Table 1 and evidence for one other category. They
358
had two minutes to inspect each set of evidence. In both the ambiguous and unambiguous conditions, the target brand was represented by the same profile-the evidence rated fourth best in the unambiguous condition blindfold pretest. Each profile was labeled with the name of a commercially available brand. Fourth, subjects returned to the other room and rerated the quality of the six brands in each of the four pretest categories. The entire procedure took about 45 minutes.
THE JOURNAL OF CONSUMER RESEARCH FIGUREB NET BELIEF CHANGE IN EXPERIMENT 1 FOR VARIOUS COMBINATIONS OF ADVERTISING AND EVIDENCE
Net belief change (AO,' AOc) Ambiguous evidence
3 2
---"7£...---____
Unambiguous evidence
Results and Discussion The six brands were partitioned into a two-level variable, posttest quality ratings for the target brand versus the competition (average quality of the five nontargets). The data were analyzed using a 3 X 2 X 2 (ambiguity by advertising by brand) multivariate analysis of covariance (MANCOY A), where pretest target and competition ratings served as covariates. A reanalysis using all six levels of the brand variable revealed the same conclusions. Although the data were analyzed using this full model, focusing on the simple contrast between the target and the competition facilitates presentation. Graphically, we show the overall difference between post-pre change scores for the target (AQt = Qt2 - Qt1) less the competition (AQc = Qc2 - QcI)' The competition remained fairly stable, so these AQt - AQc difference scores reflect changes in the target due to advertising and/or evidence. First, there was a significant effect of evidence, F(2,53) = 11.64, p < 0.001. Both the ambiguous and unambiguous forms of evidence increased the position of the target relative to the no evidence control. This is not particularly surprising since prior expectations for the target (Samsung) were not high. The unambiguous evidence revealed that Samsung was the fourth best brand, while the ambiguous evidence suggested that Sam sung was of comparable quality to the competition. As shown in Figure B, however, this main effect was qualified by a significant advertising by ambiguity interaction, F(2,53) = 3.96, p = 0.025. Simple main effects tests of advertising within levels of ambiguity indicated that the interaction was due to a large increase in quality ratings when an ad was coupled with ambiguous evidence. When confronted with ambiguous evidence, advertising significantly boosted subjects' evaluations of the target compared to when no ad was present, F(1,54) = 6.07, p = 0.017. Without exposure to product evidence, no belief change occurred, with or without an ad, indicating that the ad by itself was not persuasive. With unambiguous evidence, ratings for the target increased and ratings of
o ~idence
••- - - - - - - - - - - .
No ad
Ad Advertising
the competition decreased regardless of the presence of advertising. The unambiguous evidence was sufficiently convincing in and ofitselfto override any additional influence due to advertising. Using a theoretically motivated measure of ambiguity, we replicated Hoch and Ha (1986), demonstrating that ambiguity moderates the interaction between advertising and evidence. Also, we were able to isolate where in the evaluation process the ambiguity occurred. Our use of decomposed evidence meant that ambiguity could arise only during the information-integration stage. This is, of course, both a strength (high internal validity) and weakness (low external validity). In most product evaluation settings, ambiguity also would arise during the other stages.
EXPERIMENT 2 In Experiment 1, we assumed that subjects engaged in top-down processing if evidence inspection was guided by either prior expectations about the brands (no ad) or by advertising for the target (ad). Whereas the unambiguous evidence was clear enough to override any expectations, the ambiguous evidence was sufficiently equivocal to support expectations. To test this processing assumption, we explicitly manipulated information processing strategy in Experiment 2. Lichtenstein and Srull (1987) have shown that different processing goals (e.g., recall versus impression-formation instructions) affect the use of either bottom-up and top-down processing. We expected advertising to have its biggest influence on the evalua-
AMBIGUITY jADVERTISING-EVIDENCE INTERACTIONS
359
tion of ambiguous evidence when subjects engaged in impression-formation (top-down) processing.
its information structure and (2) consumer information processing strategies. We provide an information-theoretic framework for thinking about the ambiguity of evidence and show that evaluation sets differing in cue redundancy (and hence dominance) vary systematically in their potential to support multiple interpretations of product quality. Experiment 1 reaffirms that ambiguity is an important determinant of whether advertising shapes the interpretation of objective product evidence. Although the experimental procedures were removed from actual consumption experience, these data coupled with the results of Hoch and Ha (1986) suggest that ambiguity is necessary for observing what Wells (1984) has termed the transformational effects of advertising. George Lois (as quoted in the Washington Post, Suplee 1987) has aruged that, "When advertising is great advertising, it fastens on the myths, signs, and symbols of our common experience and becomes quite literally a benefit of the product. . . . As a result of great advertising, food tastes better, clothes feel snugger, cars ride smoother." We agree that advertising can shape how consumers interpret ambiguous product evidence. We disagree that advertising can alter the interpretation of unambiguous experience. Even "great" advertising is not going to make soyburgers taste better, double-knit leisure suits fit snugger, or Yugo cars ride smoother. These products provide clearcut evidence not open to multiple interpretations. Besides the structural determinants of ambiguity, this research shows that consumers' information processing strategies affect the likelihood of advertising-evidence interactions. When subjects were externally motivated to engage in exhaustive, bottom-up processing, the interpretation of a structurally ambiguous decision environment was rendered less susceptible to influence by advertising. Alternatively, when subjects were instructed to engage in top-down processing rather than focus on individual attribute information, advertising systematically influenced how subjects evaluated the ambiguous evidence. Recently, Hoch and Deighton (1989) offered a framework for understanding how consumers learn from experience. They discuss three factors that influence experientiallearning: the consumer's familiarity with the domain, the consumer's motivation to learn, and the ambiguity of the information environment. Our research provides additional evidence concerning the importance of information structure and motivation (in the present case, externally provided processing goals) on the consumer learning process. [Received January 1988. Revised May 1989.]
Method The four-part procedure was similar to that used in the previous study except that all subjects saw ambiguous evidence. We elected not to use unambiguous evidence, since in the first study subjects' processing ofthat evidence appeared data-driven even under the same conditions most conducive to top-down processing of ambiguous evidence. Before inspecting the evidence, half the subjects saw the target ad used in Experiment 1 embedded among other ads and half the subjects saw only unrelated ads. We manipulated processing strategy by providing subjects with explicit processing goals. "Impression" subjects examined the evidence with the top-down goal offorming an overall evaluation of each alternative and were given instructions that were similar to those used in Experiment 1: "Your task is to form a judgment about the overall quality of each brand." "Recall" subjects were provided with a bottom-up goal and were told to "try and remember as much information as you can from the product information tables" for a later recall test. As in Experiment 1, subjects in both the recall and impression conditions had two minutes to inspect the evidence. Subjects were 64 undergraduate and graduate students at the University of Chicago who were paid for their participation. The two main independent variables were advertising (no ad versus ad) and processing goal (impression formation versus recall).
Results and Discussion The data were analyzed using a 2 (advertising) X 2 (processing goal) X 2 (brand) repeated measures MANCO VA. Again, the primary dependent variable of interest was the repeated measures contrast of the target relative to the competition. There was a significant crossover interaction between advertising and processing goal, F(1,S9) = S.S8,p < O.OS. Advertising influenced opinion revision (~Qt - ~Qc) only when subjects were instructed to pursue a top-down strategy of impression formation (ad = 1.83, no ad = 0.33). Advertising had no influence when subjects were engaged in exhaustive, bottom-up processing (~Qt - ~Qc ad = 0.66, no ad = 0.74).
CONCLUSIONS We have focused on how a marketer-controlled source of information, advertising, can influence the interpretation of product evidence. We showed that two factors moderate advertising-evidence interactions: (1) ambiguity of the evidence as represented by
REFERENCES Deighton, John (1984), "The Interaction of Advertising and Evidence," Journal o/Consumer Research, 11 (December),763-770.
360 Einhorn, Hillel J. and Robin M. Hogarth (1975), "Unit Weighting Schemes for Decision Making," Organizational Behavior and Human Performance, 13, 171192. Hoch, Stephen J. and John Deighton (1989), "Managing What Consumers Learn from Experience," Journal of Marketing, 53 (April), 1-20. - - and Young-Won Ha (1986), "Consumer Learning: Advertising and the Ambiguity of Product Experience," Journal ofConsumer Research, 13 (September), 221-233. Klayman, Joshua and Young-Won Ha (1987), "Confirmation, Disconfirmation, and Information in HypothesisTesting," Psychological Review, 94 (2), 211-228. Lichtenstein, Meryl and Thomas K. Srull (1987), "Processing Objectives as a Determinant of the Relationship Between Recall and Judgment," Journal ofExperi mental Social Psychology, 23,93-118.
THE JOURNAL OF CONSUMER RESEARCH
Reisberg, D. and M. O'Shaughnessy (1984), "Diverting Subjects' Attention Slows Figural Reversals," Perception, 13, 461-468. Sabini, John and Maury Silver (1982), "Some Senses of Subjective," in Explaining Human Behavior, ed. P.F. Secord, Beverly Hills, CA: Sage, 71-91. Shepard, Roger N. (1964), "On Subjectively Optimum Selection Among Multi-Attribute Alternatives," in Human Judgments and Optimality, eds. M.W. Shelley and G.L. Bryan, New York: John Wiley, 257-28l. Shugan, Steven M. (1980), "The Cost of Thinking," Journal of Consumer Research, 7 (September), 99-111. Suplee, Curt (1987), "In Search of More Perfect Persuasion," The Washington Post, (January 18/0utposts Section), C3. Wells, William (1984), "How Advertising Works," DDB Needham Worldwide, Inc., 303 East Wacker Drive, Chicago, IL 60601.
