Jack of All Trades or Master of One? Product Differentiation and Comensatory Reasoning in Consumer Choice ALEXANDER CHERNEV* This research examines consumer reactions to two common ositioning strategies: a secialized-ositioning strategy in which an otion is described by a single feature, and an all-in-one strategy in which an otion is described by a combination of features. The emirical data reorted in this article demonstrate that a roduct secializing on a single attribute is erceived to be suerior on that attribute relative to an all-in-one otion, even when this attribute is exactly the same for both otions. It is further shown that the observed devaluation of the all-in-one otion can be mitigated by introducing another attribute on which the all-in-one otion is inferior to the secialized otion. When designing their roducts, comanies often choose one of two ositioning strategies: a narrower, secialized ositioning in which roducts are described by a single attribute and a broader, all-in-one ositioning in which roducts are described by a combination of attributes. To illustrate, Era is ositioned by Procter and Gamble as the detergent with owerful stain removal, Cheer romises to hel rotect against fading, Gain offers great cleaning ower, and, finally, Tide combines all of the above features. Combining these ositioning strategies raises the question of how consumers evaluate secialized otions in the resence of an all-in-one otion, and vice versa. From a concetual standoint, combining secialized and all-in-one otions raises several issues: (1) whether and how the erceived erformance of the attributes differentiating an all-in-one otion would change in the resence of otions secialized on each of these attributes (e.g., the change in the erceived stain-removal and fading-revention erformance of Tide in the resence of Era and Cheer); (2) whether and how the erceived erformance of the differentiating attribute of a secialized otion would change in the resence of an all-in-one otion (e.g., the change in the erceived stain-removal erformance of Era and the fading-revention *Alexander Chernev is associate rofessor of marketing, Kellogg School of Management, Northwestern University, Evanston, IL 60208 (ach@ northwestern.edu). The author thanks Gregory Carenter, Pierre Chandon, Ryan Hamilton, the editor, associate editor, and the three anonymous reviewers for their constructive comments. This research has benefited from seminars at the University of Pennsylvania and Cornell University. Dawn Iacobucci served as editor and Durairaj Maheswaran served as associate editor for this article. Electronically ublished January 22, 2007 430 erformance of Cheer in the resence of Tide); and (3) whether and how the erceived erformance of nondifferentiating attributes of a secialized otion would change in the resence of an all-in-one otion (e.g., the change in the erceived fading-revention erformance of Era and the stain-removal erformance of Cheer in the resence of Tide). Desite their concetual imortance and ractical relevance, these issues have not been exlicitly addressed in the marketing literature. This research examines consumer reactions to secialized and all-in-one roduct ositioning strategies. In articular, it is argued that when a secialized otion is comared with an all-in-one otion, the erceived erformance of the secialized otion on its focal attribute is enhanced relative to when it is considered in isolation. In contrast, the erceived erformance of the all-in-one otion is likely to decrease when it is comared with a secialized roduct relative to when it is considered in isolation. The logic for this rediction and the emirical analyses are discussed in more detail in the following sections. COMPENSATORY REASONING IN CONSUMER CHOICE The concet of comensation has been used in sychology, decision making, and marketing in different contexts. In sychology literature, the term comensation has been used to describe a sychological mechanism by which an individual attemts to make u for some ersonal deficiency by develoing or stressing another ability (e.g., Adler 1924). In decision research, comensation has been used in reference to the nature of the decision rocesses underlying an individual s choice strategies, articularly involving the 2007 by JOURNAL OF CONSUMER RESEARCH, Inc. Vol. 33 March 2007 All rights reserved. 0093-5301/2007/3304-0002$10.00
PRODUCT DIFFERENTIATION AND COMPENSATORY REASONING IN CHOICE 431 ability of an otion s strength on one attribute to make u for a deficiency on another (Johnson and Meyer 1984; Payne, Bettman, and Johnson 1993). In this context, comensatory decision strategies, reresented by multiattribute utility models (Keeney and Raiffa 1976), imly that an otion s strong erformance on one attribute can comensate for oor erformance on another, whereas in noncomensatory strategies, such as elimination by asects (Tversky 1972), a good value on one attribute cannot make u for a oor value on another. Most recently, the notion of comensation has been advanced in a somewhat different context: as a secific inference-making mechanism based on consumers intuitions of market efficiency (Chernev and Carenter 2001). In this context, market efficiency reflects the belief that offerings are riced at value arity, such that the benefit-cost tradeoffs are constant across otions; as a result, higher riced roducts are also of better quality, and vice versa. To illustrate, a consumer considering two equally riced alternatives might infer that the otion suerior on the observed attributes is likely to be inferior on some of the unobservable attributes. Thus, the concet of market efficiency reflects the notion that firms behave rationally, resulting in offerings that are value equivalent. Comensatory inferences derived from consumers market efficiency beliefs aim to resolve a salient incongruity between the intuitively exected and the observed relationshis among otions in a given choice set, whereby extreme values on one dimension are associated with inferior erformance on another. Concetually, comensatory inferences can be linked to consumers covariation beliefs, which reflect their intuitions about the degree of relationshi between two or more factors. Two tyes of covariation beliefs can be identified: attribute-secific covariation and context-deendent covariation. Most of the extant research has focused on attribute-secific covariation that is confined to airs of articular factors such as rice and quality (Baumgartner 1995; Bettman, John, and Scott 1986; Raghubir and Corfman 1999; Shiv, Carmon, and Ariely 2005), brand name and quality (Allison and Uhl 1964; Janiszewski and Van Osselaer 2000; Rao and Monroe 1989; Wernerfelt 1988), and reliability and warranty (Dick, Chakravarti, and Biehal 1990). To illustrate, consumers might believe that higher quality roducts are also more exensive (Lichtenstein and Burton 1989) and that more reliable roducts are likely to offer a longer warranty (Broniarczyk and Alba 1994). Unlike attribute-based covariation, in which an otion s erformance on a articular attribute is inferred from its erformance on other attributes, in the case of context-based covariation an otion s values are derived from the decision context defined by the other alternatives in the set. Examles of such context-based covariation are given by rior research on reference formation, documenting that consumer inferences of roduct erformance are derived from the characteristics of the other alternatives in the set (Prelec, Wernerfelt, and Zettelmeyer 1997; Wernerfelt 1995). In the same vein, rior research has documented that consumer resonse to marketing rograms and romotional offers is a function of the degree to which these activities fit the references of target customers better than of the tyical customer in the oulation (Kivetz and Simonson 2003). Building on rior findings, this research extends the notion of comensatory reasoning beyond attribute-secific covariation to a scenario in which inferences are drawn from the decision context defined by the other alternatives in the set, rather than from this otion s erformance on a articular attribute. These context-based comensatory inferences stem from a consumer s belief that otions in a given choice set are balanced in a way that advantages on one dimension are comensated for by disadvantages on another, even in the absence of rior attribute-secific covariation beliefs. Thus, an otion that excels on a articular attribute can be inferred to be inferior on some of the other attributes even in the absence of an inherent covariation among its attributes (e.g., static revention and stain removing in a laundry detergent). In this research, the decision strategy reflecting individuals belief that alternatives in a given choice set are balanced in their overall erformance is referred to as the zerosum heuristic. The term zero-sum heuristic is derived from the zero-sum-game assumtion in game theory, which imlies that the wins and losses in a game will add u to zero for each set of strategies chosen (Von Neumann and Morgenstern 1953). Simly ut, a zero-sum game imlies that one layer s winnings should equal the other layer s losses. The concet of zero-sum heuristic advanced in this article is concetually similar to that of a zero-sum game in that it imlies a closed system in which all otions are balanced; thus, for each of the otions, the advantages on one attribute must be comensated for by disadvantages on another. Thus, the zero-sum heuristic is not contingent on rior covariation beliefs and, instead, is a function of the disersion of attribute values across choice alternatives. In this context, when consumers who exect otions in a given choice set to be comarable in terms of their overall erformance are resented with a choice set in which the allin-one otion aears to dominate the others, they are likely to draw comensatory inferences devaluing the aarently more attractive all-in-one otion. The zero-sum heuristic can be related to the notion of trade-off consistency in choice (Simonson and Tversky 1992). The trade-off consistency of a given choice set is usually characterized by the rate of exchange between attributes, so that in sets with a constant rate of exchange between attributes the advantages and disadvantages of each otion are balanced. In this context, the zero-sum heuristic imlies that when evaluating sets comrising otions with varying rates of exchange between attributes, consumers are likely to interret ambiguous attribute values in a way that decreases the observed trade-off contrasts and equates the rate of exchange across attributes. This research argues that when evaluating sets comrising secialized and all-in-one otions, consumer reasoning is often guided by the zero-sum heuristic, which redicts that
432 JOURNAL OF CONSUMER RESEARCH the sum of advantages and disadvantages is constant across the choice otions. In this context, it is argued that consumers are likely to interret the values of the choice alternatives in a way that equates the overall erformance of the all-in-one and the secialized otions, leading to a comensatory devaluation of the erceived erformance of the all-in-one otion and enhancement of the erceived erformance of the secialized otions. This idea is further articulated in a series of research hyotheses in the following section. RESEARCH HYPOTHESES Consider a set of three alternatives, each described on two attributes as follows: otion A is differentiated by attribute 1, otion B is differentiated by attribute 2, and otion C is differentiated by both attributes. To illustrate, consider three shaving creams, one emhasizing its moisturizing effect, the second one romoting its skin-rotection effectiveness, and the third one claiming to be effective in both skin moisturizing and skin rotection. In the absence of comensatory inferences, consumers are likely to view otion C as suerior on both attributes (i.e., c1 1 b1 and c2 1 a 2The ), as illustrated in figure 1A. use of the zero-sum heuristic, however, is likely to lead to comensatory inferences that balance the overall attractiveness of otions A, B, and C, making these otions trade-off equivalent. In articular, comensatory reasoning is likely to lead to a devaluation of the all-in-one otion C (DC infig. 1B) and a olarization of attribute values of the secialized otions A and B (DA and DB). In this context, comensatory inferences can be reresented as having a threefold effect. First, the all-in-one otion is likely to be devalued, such that the erceived erformance of the attributes differentiating an all-in-one otion will decrease in the resence of otions secialized on these attributes (comensatory devaluation effect denoted by Dc 1 and Dc 2 in fig. 1B). In addition to discounting the erformance of the all-inone otion C, consumers might also draw inferences about the secialized otions A and B, such that the erceived erformance of the attributes of a secialized otion will become more olarized in the resence of an all-in-one otion. In articular, the erceived erformance of the differentiating attribute of a secialized otion will increase in the resence of an all-in-one otion (comensatory enhancement effect denoted by Da 1 and Db 2 in fig. 1B). At the same time, consumers might also discount the erformance of secialized otions A and B on their secondary attributes in the resence of an all-in-one otion, leading to a downshift in their erceived erformance (comensatory devaluation effect denoted by Da 2 and Db 1 in fig. 1B). An imortant assumtion of the zero-sum heuristic is that the choice roblem is viewed as fully identified, such that the number of attributes used to define choice alternatives is constant and all attribute values are readily available. This assumtion raises the question of how the resence of attributes on which otions erformance is not readily observable influences comensatory reasoning. Generally seaking, there are two ossibilities. First, consumers might infer that all otions have equal values on the unobservable attribute a scenario that should not influence consumers use of comensatory reasoning and their reliance on the zero-sum heuristic. The second ossibility is that consumers infer that otions vary in their erformance on the unobservable attribute. In this case, the strength of the zero-sum heuristic is likely to be a function of the attern of these inferences: it should be stronger when the inferred values are consistent with the readily available disersion of attribute values and should be weaker (or even disaear) when the inferred attribute values are inconsistent with the disersion of the values of the readily available attributes. To illustrate, consider the earlier scenario involving three shaving creams, one emhasizing the moisturizing effect (otion A), the second romoting skin-rotection effectiveness (otion B), and the third claiming to be effective for both skin moisturizing and skin rotection (otion C). Now imagine that an additional attribute (e.g., rice, skin conditioning) on which the otions erformance is unobservable is art of consumers evaluation of these otions. In this context, it can be argued that the otions erformance on the two differentiating attributes (moisturizing and rotection) is likely to be a function of their erceived erformance on the attribute on which the otions erformance is unobservable. In articular, consumers who erceive the all-in-one otion C to be inferior on the unobservable attribute will be less likely to infer a comensatory relationshi between the otions values on featured attributes than consumers who erceive otion C to dominate the others on the unobservable attribute. Overall, this research osits that when evaluating choice sets comrising both all-in-one and secialized alternatives, consumers are likely to adot a zero-sum heuristic, which equates the overall attractiveness of choice alternatives and evaluates the available information in a comensatory fashion. In articular, two tyes of comensatory effects can be identified: comensatory devaluation, which lowers the erceived erformance of the all-in-one otion, and comensatory olarization, which enhances the erceived erformance of the secialized otion on the differentiating attribute, while detracting from its erformance on the secondary attribute. It is further roosed that the strength of the comensatory inferences across the attributes differentiating choice alternatives is a function of the resence of salient attributes on which otions erformance is not readily observable. Thus, in the resence of a salient attribute on which otions values are unobservable, comensatory inferences are redicted to be less ronounced than in the case when otions values on all salient attributes are readily observable. Furthermore, when consumers are exlicitly asked to make inferences about otions erformance on unobservable attributes, comensatory effects are redicted to be less ronounced when the inferred erformance is inconsistent with the readily available attribute information. These redictions are tested in the following exeriment.
PRODUCT DIFFERENTIATION AND COMPENSATORY REASONING IN CHOICE 433 FIGURE 1 COMPENSATORY ENHANCEMENT AND DEVALUATION EFFECTS IN CHOICE NOTE. A and B are the secialized otions, and C is the all-in-one otion. A,B, and C reflect the erceived erformance of otions A, B, and C in the resence of comensatory inferences. Da 1 and Db 2 deict comensatory enhancement, whereby the value of otions A and B is enhanced by comensatory inferences. In contrast, Da 2, Db 1, Dc 1, and Dc 2 deict comensatory devaluation, whereby attribute values of otions A, B, and C are decreased by comensatory inferences. EXPERIMENT 1 The goal of this exeriment was to emirically examine how attribute erformance of an all-in-one otion is influenced by the resence of secialized otions, and vice versa. The secifics of the exerimental stimuli and research design are resented in more detail in the following sections. Method The resondents, 227 Northwestern University students, were recruited to articiate in a study on consumer decision making and were informed that the choice task involved making hyothetical urchase decisions. Five roduct categories were used as stimuli: laundry detergent, toothaste,
434 JOURNAL OF CONSUMER RESEARCH shaving cream, cold relief medicine, and vitamin sulements. Similar categories have been successfully used in rior research (Carenter and Nakamoto 1989; Simonson 1989). Choice sets consisted of either two or three otions, each described by either one attribute (in case of secialized otions) or two attributes (in case of all-in-one otions) as shown in the aendix, table A1. For all roduct categories, otions A and B were described by a single attribute (e.g., revents cavities), and otion C was described by a combination of features describing otions A and B (e.g., revents cavities and whitens teeth). Choice alternatives were organized into four scenarios: a trinary set ABC and three binary sets AB, AC, and BC. Resondents were randomly assigned to one of these scenarios, and each resondent was given five choice roblems, one er roduct category. For each roblem, resondents were shown the alternatives and were asked to rate their attribute erformance. To illustrate, in the toothaste category resondents were first asked to rate the erformance of the otions in the set on the first attribute (e.g., How would you rate the cavity-revention roerties of toothaste A? ) and then asked the same question with resect to the other otion(s) in the set. Next, resondents were asked to rate the otions erformance on the second attribute (i.e., teeth whitening). All ratings were collected using a nine-oint scale (1 low, 5 average, 9 high). Following the rating task, resondents were asked to make two choices, conditional on one of the attributes having rimary imortance. To illustrate, resondents were first asked to choose a toothaste assuming that cavity revention is the more imortant attribute (e.g., Which of the two roducts would you choose if your rimary concern is cavity rotection? ) and then to make a choice assuming that teeth whitening was the more imortant attribute. In addition to varying the comosition of the choice set, this exeriment also maniulated the salience of otions ricing. For that urose, rior to evaluating the otions attribute erformance, some of the resondents were asked to indicate how they would exect these otions to be riced. Resondents could choose between indicating that a articular otion was more exensive than the others or that all otions were equally riced ( All otions will be equally riced; otion A will be the most exensive one; otion B will be the most exensive one; or otion C will be the most exensive one ). The goal of this maniulation was to test the roosition that in the resence of an attribute on which otions erformance is not readily observable the strength of comensatory inferences is a function of otions erceived erformance on this attribute. Overall, there were eight scenarios corresonding to the conditions of the exerimental design (4 choice sets # 2 rice-inference conditions). The entire rocedure took 10 minutes, on average. At the end of the exeriment resondents were debriefed and aid $5.00 for articiating. Results Overview Each of the 227 resondents evaluated one set of alternatives in each of the five roduct categories, which yielded 1,135 sets. In each of these sets, resondents rated the erformance of each otion on both attributes, which yielded six attribute-secific ratings for each trinary set ABC and four attribute-secific ratings for each of the binary sets AB, AC, and BC. Thus, the total number of attribute ratings was 5,156 (four missing data oints). Resondents in each of the four scenarios (ABC, AB, AC, and BC) were further assigned into two grous. Some were asked to evaluate relative rices of the otions rior to rating their erformance on nonrice attributes; others were given the attribute-rating task without being asked to comare the otions on rice. Based on a random assignment, there were 114 resondents in the first condition (569 observations) and 113 in the second (565 observations). Of the 569 resonses, 380 indicated that the all-in-one otion was likely to be riced higher than the other otion(s), while the remaining 189 resondents indicated that either (1) one of the secialized otions was likely to be higher riced or (2) that all otions were likely to be at rice arity. This assignment of resondents into one of the two inferred-rice conditions was later used to examine how the resence of attributes on which otions erformance is not readily observable influences the strength of comensatory reasoning effects. A summary of the mean attribute ratings (averaged across roduct categories) is given in table 1. The attribute erformance of choice otions is consistent with the reresentation deicted in figure 1, such that otion A is differentiated by attribute 1 and otion B is differentiated by attribute 2. The significance of the data atterns shown in table 1 was validated by testing a model in which attribute ratings were given as a function of the choice set, inferredrice disersion, roduct category, and all interactions. Because not all observations were truly indeendent from one another (the same individual rated several roduct categories and also rated different otions within the choice set), the statistical analyses controlled for the within-subject nature of certain asects of the data (Winer, Brown, and Michels 1991). Because roduct category did not significantly interact with the deendent variables of interest, for resentation uroses the data are aggregated across categories. Due to the comlex nature of the focal effects and the ossibility of resonse biases and/or measurement errors, each of the effects was examined using multile emirical tests. The data analyses are resented in more detail in the following sections. Devaluation of the All-in-One Otion This research argues that the erceived erformance of the all-in-one otion will decrease in the resence of secialized otions (Dc 1, Dc 2 ). There are three strategies to test this roosition. The first test involves comaring the ratings of the all-in-one otion C to the ratings of the secialized otions A and B in the set ABC. The second strategy is to
PRODUCT DIFFERENTIATION AND COMPENSATORY REASONING IN CHOICE 435 TABLE 1 ATTRIBUTE RATINGS AS A FUNCTION OF THE COMPOSITION OF THE CHOICE SET AND PRICE INFERENCES Salience of unobservable attributes Choice otion Attribute Choice sets ABC AC BC AB Scenario 1: Otions evaluated only on featured attributes (no-inference condition) A 1 7.59 7.55 7.10 2 2.77 2.94 3.58 B 1 2.93 2.61 3.50 2 7.43 7.47 6.74 C 1 6.21 6.23 7.24 2 6.31 7.09 6.37 N 145 140 145 135 Scenario 2A: Otions evaluated on an unobservable attribute (rice) otion C exected to be the most exensive (high rice condition) A 1 7.72 7.48 2 2.34 2.48 B 1 2.60 2.50 2 7.87 7.35 C 1 7.18 6.81 6.92 2 7.31 7.06 6.98 N 122 124 134 Scenario 2B: Otions evaluated on an unobservable attribute (rice) otion C not exected to be the most exensive (equal/low rice condition) A 1 7.40 7.56 6.89 2 2.60 2.94 3.89 B 1 2.72 3.10 3.69 2 7.67 7.29 6.78 C 1 6.40 6.25 6.23 2 6.51 6.83 5.87 N 43 36 31 79 NOTE. Otions attribute erformance is consistent with the reresentation in fig. 1: otion A is differentiated by attribute 1; otion B is differentiated by attribute 2, and otion C is the all-in-one otion. Attribute values of choice otions are given in the aendix. Numbers in each cell reresent mean averages for all observations aggregated across categories and across resondents. The number of observations in each set is given by N. Cells in which observations could not be obtained by virtue of the exerimental design (e.g., the attractiveness of otion C could not be evaluated in context of the set AB) are left blank. comare the ratings of the all-in-one otion across the binary and trinary choice sets a strategy that comares the ratings of otion C in the set ABC to the ratings of otion C in sets AC and BC. Finally, the third strategy involves comaring the attribute ratings of the all-in-one otion across the two binary sets AC and BC. The data analyses corresonding to these three strategies are resented in more detail below. The first test of the rediction that the erceived erformance of the all-in-one otion will decrease in the resence of secialized otions comared the individual-secific attribute ratings of otion C to the attribute ratings of otions A and B in the set ABC. In articular, this rediction imlies that C1 ABC! A1ABC and C2 ABC! B2ABC, where C1 ABC is the rating of otion C on attribute 1 in the set ABC and C2 ABC is the rating of otion C on attribute 2 in the same set. The data show that the mean ratings were MC1(ABC) 6.21 versus MA1(ABC) 7.59 ( t 9.6;!.001) when comaring C1 ABC and A1 ABC and MC2(ABC) 6.31 versus MB2(ABC) 7.43 ( t 6.7;!.001) when comaring C2 ABC and B2 ABC. The difference in the attribute ratings of otion C was nonsignificant, indicating that the devaluation effect was symmetric across attributes ( MC1(ABC) 6.21 versus MC2(ABC) 6.31; 1.20). These data are consistent with the exerimental redictions. A further test of the imact of secialized otions on the all-in-one otion comared the ratings of the all-inone otion C in the trinary set ABC and its ratings in the binary sets AC and BC. Of articular interest are the attribute-rating comarisons across scenarios with and without a secialized otion: C1 ABC versus C1 BC and C2 ABC versus C2 AC. It was roosed that adding otion A to the set BC should lead to a devaluation of otion C on attribute 1 because otion A is secialized on this attribute, as well as a devaluation of otion C on attribute 2, on which otion B is secialized. The data show that otion C had lower attribute ratings in the trinary set than in either of the binary sets. The mean rating of otion C on attribute 1 in the set ABC was MC1(ABC) 6.21, significantly lower than its rating of MC1(BC) 7.24 in the set BC ( F(1, 166) 21.97;!.001). Similarly, the mean rating of otion C on the second attribute in the set ABC was MC2(ABC) 6.31, significantly lower than its rating of MC2(AC) 7.09 in the set AC ( F(1, 165) 13.27;!.001).
436 JOURNAL OF CONSUMER RESEARCH The third test of the rediction that the erceived erformance of the all-in-one otion will decrease in the resence of secialized otions comared the attribute ratings of otion C across the two binary sets AC and BC. Consistent with the exerimental redictions, the resence of a secialized alternative in each of the sets AC and BC should lead to asymmetric devaluation of otion C on the attribute featured by the secialized otion. Thus, otion C s ratings on attribute 1 should be lower in the set AC than in the set BC, and vice versa for attribute 2. The data show that the mean rating of otion C on attribute 1 was significantly lower in the set AC than in the set BC (MC1(AC) 6.23 versus MC1(BC) 7.24; F(1, 159) 21.09;!.001). Similarly, the mean rating of otion C on the second attribute was significantly lower in the set BC than in the set AC ( MC2(BC) 6.37 vs. MC2(AC) 7.09; F(1, 159) 10.63;!.005). Considered together, the three tests offer converging emirical evidence in suort of the roosition that secialized alternatives lead to a comensatory devaluation of the allin-one otion. Polarization of the Secialized Otions This research argued that the attribute ratings of a secialized otion are olarized by the resence of an all-inone otion in a way that enhances the erceived erformance of the secialized otion on its differentiating attribute (Da 1, Db 2 ) and detracts from its erceived erformance on the secondary attribute (Da 2, Db 1 ). This roosition was tested by comaring the attribute ratings of the secialized otions A and B as a function of the resence of an all-in-one otion. The test of the comensatory enhancement effect comared the ratings of otions A and B on their differentiating attributes across sets ABC and AB. The data show that the mean rating of otion A in the set ABC was MA1(ABC) 7.59, which was significantly greater than this otion s ratings in the absence of the all-in-one otion MA1(AB) 7.10; ( F(1, 127) 4.11;!.05). Similarly, otion B was rated significantly higher on its differentiating attribute in the resence of the all-in-one otion C than when otion C was absent ( MB2(ABC) 7.43 vs. MB2(AB) 6.74; F(1, 127) 8.23;!.01). A further test of the comensatory enhancement effect comared the attractiveness of otions A and B across sets AB and AC/BC. In this context, the secialized otions A and B are redicted to be erceived as more attractive on their differentiating attributes in the resence of an all-in-one alternative. The data were consistent with this rediction: otion A was erceived to be more attractive on its differentiating attribute in the set AC than in the set AB ( MA1(AC) 7.55 vs. MA1(AB) 7.10; F(1, 120) 3.91;.05). Similarly, otion B was rated higher on its differentiating attribute in the set BC than in the set AB ( MB2(BC) 7.47 vs. MB2(AB) 6.74; F(1, 121) 7.99;!.01). The test of the comensatory devaluation effect comared the ratings of otions A and B on their nondifferentiating attributes across sets ABC and AB. The data show that otion A was erceived as less attractive on its nondifferentiating attribute in the set ABC than in the set AB ( MA2(ABC) 2.77 vs. MA2(AB) 3.58; F(1, 127) 9.55;!.005). Similarly, otion B was erceived as less attractive on its nondifferentiating attribute in the absence of the all-in-one otion than when the all-in-one otion was resent ( MB1(ABC) 2.93 vs. MB1(AB) 3.50; F(1, 127) 4.68;!.005). A further test of the comensatory devaluation effect comared the attractiveness of otions A and B across sets AB and AC/BC. The data show that the nondifferentiating attribute of otion A received lower ratings in the set AC than in the set AB ( MA2(AC) 2.94 vs. MA2(AB) 3.58; F(1, 120) 4.86;!.05). Similarly, the nondifferentiating attribute of otion B was rated less attractive in the set BC than in the set AB ( MB1(BC) 2.61 vs. M B1(AB) 3.50; F(1, 121) 9.95;!.005). Overall, the exerimental data suort the roosition that adding an all-in-one otion to a choice set comrising secialized otions is likely to lead to a comensatory enhancement of the erformance of the secialized otions on differentiating attributes and a comensatory devaluation of the erformance of the secialized otions on nondifferentiating attributes. Inferring Otions Performance on Unobservable Attributes This research argued that the strength of comensatory inferences is a function of the erceived erformance of the otions on unobservable attributes. In articular, the devaluation of the all-in-one otion in the resence of secialized alternatives was redicted to be less ronounced when the all-in-one otion is erceived to be inferior on the unobservable attribute. This rediction was tested by making one of the unobservable attributes (rice) salient and examining the imact of consumers rice ercetions on the strength of the comensatory inferences. In this context, it was exected that resondents who exected the all-in-one otion to be higher riced would be less likely to devalue this otion than resondents who exected the all-in-one otion to be riced at arity or lower than the secialized otions. Recall that resondents were randomly assigned to either the no-inference condition or the rice-inference condition. Based on their resonses, individuals in the rice-inference condition were further assigned to two grous: a high rice grou made u of resondents who inferred that the all-inone otion was likely to be the most exensive otion in the set and a low rice grou made u of resondents who exected the all-in-one otion to be riced at arity or less than the other otions. In this context, the imact of rice inferences on the strength of the comensatory reasoning effects was tested by comaring the disersion of attribute ratings across the control and high rice conditions, whereby the comensatory devaluation of the all-inone otion should be greater in the control than in the high rice condition. The strategy for testing this rediction was similar to the strategy used in testing the devaluation of the
PRODUCT DIFFERENTIATION AND COMPENSATORY REASONING IN CHOICE 437 all-in-one otion: the imact of rice inferences on the disersion of attribute ratings of otions was evaluated (1) within the set ABC, (2) across sets ABC and AC/BC, and (3) across sets AC and BC. A comarison of attribute ratings of the all-in-one otion C between the high rice and no-inference conditions shows that in the set ABC otion C was rated higher on both attributes when it was erceived to be higher riced ( MC1(ABC) 7.18 and MC2(ABC) 7.31) than when no rice inferences were drawn ( MC1(ABC) 6.21 and MC2(ABC) 6.31). The difference in attribute evaluations of otion C was significant for both attributes ( F(1, 86) 19.37;!.001 for attribute 1 and F(1, 86) 16.93;!.001 for attribute 2). Similarly, in sets AC and BC, otion C had higher attribute ratings when inferred to be higher riced ( MC1(AC) 6.81 and MC2(BC) 6.98) than when there was no rice inference ( MC1(AC) 6.23 and MC2(BC) 6.37; F(1, 79) 6.18;!.05 and F(1, 80) 5.55;!.05). Furthermore, inferred-rice ranking influenced not only the ratings of the all-in-one otion C but also its relative attractiveness vis-à-vis the secialized otions A and B. Thus, the difference in attractiveness of otions A and C was DMA1 C1(ABC).54 in the high rice condition, comared with DMA1 C1(ABC) 1.38 in the no-inference condition ( F(1, 86) 12.56;!.005). Similarly, the difference in attractiveness of otions B and C was DMB2 C2(ABC).56 in the high rice condition, comared with DM B2 C2(ABC) 1.12 in the no-inference condition ( F(1, 86) 4.32;!.05). These data show that the devaluation of the all-in-one otion was mitigated by the inferred rice-inferiority (i.e., higher rice) of this otion a finding consistent with the exerimental redictions. The imact of rice inferences on the strength of the comensatory reasoning effects was further tested by comaring the attribute ratings of the all-in-one otion C in the set ABC to its ratings in sets AC and BC across the high rice and the no-inference conditions. The data show that the difference in attractiveness of otion C across sets ABC and AC was DMC2(ABC) C2(AC).26 in the high rice condition, comared with DMC2(ABC) C2(AC).78 in the noinference condition ( F(1, 165) 9.85;!.005) a finding consistent with the exerimental redictions. Similarly, the difference in attractiveness of otion C across sets ABC and AB was DMC1(ABC) C1(BC).26 in the high rice condition, comared with DMC1(ABC) C1(BC) 1.02 in the no-inference condition ( F(1, 86) 18.04;!.001). The imact of rice inferences on the strength of the comensatory reasoning effect was also tested by comaring the attribute ratings of otion C in sets AC and BC across the high rice and the no-inference conditions. The data show that the difference in attractiveness of otion C across sets AC and BC was DMC1(AC) C1(BC).12 in the high rice condition, comared with DMC1(AC) C1(BC) 1.01 in the no-inference condition ( F(1, 159) 8.32;!.005). Similarly, the difference in attractiveness of otion C across sets ABC and BC was DMC2(BC) C2(AC).08 in the high rice condition, comared with DMC2(BC) C2(AC).71 in the no-inference condition ( F(1, 165) 3.32;!.10). The otential artifacts of the rice-inference maniulation were also tested by examining the strength of the comensatory reasoning effects across the two rice-inference conditions. In this context, this research argued that the comensatory devaluation effect will be weaker for resondents who rated otion C as being the highest riced than for resondents who did not infer otion C to be the most exensive. The data show that otion C was rated higher in the high rice condition ( MC1(ABC) 7.18 and MC2(ABC) 7.31; F(1, 86) 5.65;!.05) than in the equal/low rice condition ( MC1(ABC) 6.40 and MC2(ABC) 6.51; F(1, 86) 6.99;!.05). Further analysis shows that the difference in the attractiveness of otion C and otions A and B was also less ronounced for resondents in the high rice condition than for those in the equal/low rice condition. Thus, in the high rice condition the difference in attribute evaluations across otions A and C was DMA1 C1(ABC).54 and DMB2 C2(ABC).56 across otions B and C. In contrast, in the equal/low rice condition the difference in attribute evaluations was greater: DMA1 C1(ABC) 1.00 and DMB2 C2(ABC) 1.16. The combined effect of the differences in the devaluation effect was significant ( F(2, 86) 3.84;!.05), lending further suort to the roosition that the devaluation of the all-in-one otion is mitigated by the inferred inferiority of this otion on rice. Considered together, the above findings suort the roosition that the comensatory devaluation of the all-in-one otion will be less ronounced when this otion is erceived to be inferior on an attribute with not readily observable values. This rediction was further tested by analyzing resondents choices across exerimental conditions. Recall that resondents were asked to make two contingent choice decisions (one assuming attribute 1 to be most imortant, and one assuming attribute 2 to be most imortant). The attern of disersion of choice shares across the exerimental conditions was consistent with the attribute ratings shown in figure 1, and the relevant differences across the exerimental conditions were significant, relicating the ratings data. These findings suort the exerimental redictions and further document comensatory effects across different reference-elicitation rocedures (ratings and choice). Discussion The data furnished by exeriment 1 suort the roosition that consumers evaluate the secialized and all-in-one otions in a comensatory fashion. In articular, the attractiveness of the attributes differentiating the all-in-one otion was found to decrease in the resence of secialized otions. The data further show that comensatory reasoning also leads to olarization of the attributes describing secialized otions: it enhances the attractiveness of the differentiating attribute of a secialized otion and devalues this otion on its secondary attributes. Moreover, the strength of comensatory inferences was shown to be a function of consumer inferences about the otion s erformance on unobservable
438 JOURNAL OF CONSUMER RESEARCH attributes. Thus, merely asking resondents to infer otions erformance on an unobservable attribute resulted in a significant decrease in comensatory devaluation effects. In this context, the comensatory effects were ronounced when the inferred erformance was inconsistent with the readily available attribute information than when the inferred erformance was consistent with the available information. A further test of the comensatory reasoning hyothesis can be offered by examining the disersion of attribute ratings across choice alternatives as a function of rice. Prior research on rice-quality inferences has shown that increasing the rice of an otion is also likely to increase its erceived erformance on nonrice attributes (Huber and McCann 1982; Johnson 1987; Johnson and Levin 1985; Lichtenstein and Burton 1989). Building on these findings, consumers reliance on the zero-sum heuristic can be further tested by examining the imact of a rice increase on the relative attractiveness of the all-in-one and the secialized otions. The logic of the argument that varying rice is likely to have a differential imact on the erceived erformance of the secialized and all-in-one otions is derived from the concavity of the value function, which imlies that an increase in an object s value on a articular attribute is associated with a decrease in this attribute s marginal utility (Bernoulli 1738; Kahneman and Tversky 1979; Tversky and Kahneman 1991). The rincile of diminishing sensitivity, imlied by the concavity of the value function, is consistent with the sychohysical rincile that sensitivity to changes along a articular attribute is reduced as the magnitude of that attribute increases (Torgerson 1958). In marketing, the diminishing sensitivity argument has been demonstrated in the context of new roduct develoment, whereby new roduct features have been shown to contribute more value to relatively inferior roducts than to relatively suerior roducts (Nowlis and Simonson 1996; see also Chernev and Hamilton 2007). The differential imact of varying rice on the erceived erformance of the secialized and all-in-one otions is illustrated in figure 2. In the absence of comensatory inferences, both otions A and C are erceived to have equal ratings on attribute 1, as shown in figure 1A. Therefore, in the absence of comensatory enhancement and devaluation effects, the imact of raising rice on otions erceived erformance should be the same for the all-in-one and the secialized otions, as shown in figure 2A. In contrast, comensatory reasoning and the associated devaluation and enhancement effects are likely to make the secialized otion A more attractive than the all-in-one otion C on attribute 1 (see fig. 1B). If this indeed is the case, then ricing each of these otions at a remium should have a differential imact on their erceived erformance. In articular, the diminishing sensitivity argument redicts that the increase in the relative attractiveness of the devalued otion C will be greater than the increase in the enhanced otion A (see fig. 2B). The essence of the above argument is that if the zerosum heuristic leads to a olarization of an otion s erformance on differentiating attributes then increasing the rice of each of these otions by the same amount should have an asymmetric imact on the otions erformance on nonrice attributes; this, in turn, will lead to a greater imrovement in attribute erformance of the all-in-one alternative relative to secialized otions. Because the difference in otions attribute erformance is a rerequisite for the redicted asymmetric imact of rice, the resence of a significant asymmetric imact of rice can be viewed as evidence suorting the zero-sum heuristic. The validity of this rediction is tested in the following exeriment. EXPERIMENT 2 The goal of this exeriment was to further test the comensatory reasoning hyotheses and, in articular, to test the roosition that the erceived attractiveness and choice share of the all-in-one otion can be increased by ricing this otion at a remium in relation to the secialized otions. This rediction was tested by emloying four exerimental conditions: a base condition in which all otions were riced at arity and three conditions in which one of the otions was riced higher than the others. The exerimental method and emirical results are resented in more detail in the following sections. Method Resondents, 74 Northwestern University students, were recruited to articiate in a study on consumer decision making and were informed that the choice task involved making hyothetical urchase decisions. Four roduct categories were used as stimuli: laundry detergent, shaving cream, toothaste, and vitamin sulements. Choice alternatives were described on either one (for secialized otions) or two (in case of the all-in-one otion) attributes and riced as shown in the aendix, table A2. Resondents were randomly assigned to one of the four exerimental conditions that differed in terms of the distribution of otions rices. All sets were made u of three otions: two secialized otions (A and B) and an all-inone otion (C). Otions in one of the conditions (control condition) were all riced at arity, while in the other three conditions one of the otions was riced higher than the others (ABC, A + BC, AB + C, ABC +, with the higher riced otion marked with a +). Each resondent made four choices, one in each roduct category. For each roduct category, resondents were asked to choose one of the three otions and then rate all the otions erformance on each of the two attributes on a nine-oint scale (same as in the first exeriment). At the end of the exeriment, resondents were debriefed and aid for articiating.
PRODUCT DIFFERENTIATION AND COMPENSATORY REASONING IN CHOICE 439 FIGURE 2 COMPENSATORY ENHANCEMENT AND DEVALUATION EFFECTS IN CHOICE: COMPENSATORY REASONING AS A FUNCTION OF PRICE INFORMATION NOTE. The grah reflects otions erformance on one of the two attributes differentiating choice alternatives (attribute 1 in fig. 1). Otion A is the secialized otion and otion C is the all-in-one otion. Otions A + and C + denote the higher riced versions of otions A and C. Results The Imact of Price Disersion on Attribute Inferences. Each of the 74 resondents evaluated one set of alternatives in each of the four roduct categories, which yielded 296 sets. For each set, resondents rated the erformance of each otion on both attributes, yielding a total of 1,776 observations (six attribute-secific ratings for each set). A summary of the mean attribute ratings (averaged across roduct categories) is given in table 2. The significance of the data atterns shown in the table was validated by testing a model in which attribute ratings were given as a function of rice disersion, otion tye (secialized vs. all-in-one), roduct category, and their interactions. The data analyses are resented in more detail in the following sections. Consistent with the comensatory reasoning argument, the data show that when otions were riced at arity, re-
440 JOURNAL OF CONSUMER RESEARCH TABLE 2 ATTRIBUTE RATINGS AS A FUNCTION OF PRICE DISPERSION (EXPERIMENT 2) Otion/attribute Price Equal C + A + B + A 1 7.53 7.38 7.97 7.04 A 2 2.73 3.14 3.12 3.11 B 1 2.91 2.95 2.96 3.42 B 2 7.42 7.62 7.09 7.91 C 1 5.99 6.89 5.72 6.5 C 2 6.14 7.01 6.39 5.72 NOTE. Otion A is differentiated by attribute 1 (A 1 ), otion B is differentiated by attribute 2 (B 2 ), and otion C is the all-in-one otion differentiated by both attributes (C 1 and C 2 ). Numbers in each cell reresent mean averages for all observations aggregated across categories and across resondents. In each set, N 74. sondents were likely to devalue the all-in-one otion relative to the secialized otions. In articular, the comensatory reasoning argument imlies that C 1! A1 and C 2! B 2, where C 1 is the rating of otion C on attribute 1 and C 2 is the rating of otion C on attribute 2. Consistent with this rediction, the data show that MC1 5.99 versus MA1 7.53 ( t 10.07;!.001) when comaring C 1 and A 1, and MC2 6.14 versus MB2 7.42 ( t 7.78;!.001) when comaring C 2 and B 2. The data further show that higher rice was also associated with an increase in the erceived erformance of the all-in-one otion on both attributes (MC1(ABC) 5.99 vs. MC1(ABC+) 6.89; F(1, 219) 28.34;!.001, and MC2(ABC) 6.124 vs. MC2(ABC+) 7.01; F(1, 219) 19.03;!.001). A similar increase in erceived erformance occurred when the higher riced otion was one of the secialized alternatives. Thus, resondents were more likely to rate the secialized otion A as suerior on the differentiating attribute A 1 when it was riced at a remium than in a scenario where all otions were riced at arity ( MA1(ABC) 7.53 vs. MA1(ABC+) 7.97; F(1, 219) 6.94;!.01). In the same vein, otion B was rated suerior on the differentiating attribute B 2 when riced at remium than when riced at arity ( MB2(ABC) 7.42 vs. MB2(ABC+) 7.91; F(1, 219) 5.80;!.05). More imortant, the comensatory reasoning argument redicts that the enhancement from remium ricing is likely to be more ronounced for the all-in-one otion than for either of the secialized otions. This rediction is tested by comaring the magnitude of the increase in attractiveness of the all-in-one and the secialized otions. For that urose, for each resondent the increase in the erceived attractiveness of the remium-riced otion on the differentiating attribute was comared across the secialized and the all-in-one otion. In articular, the average increase in the attractiveness of the secialized otion differentiated by attribute 1 was less ronounced than the increase in the attractiveness of the all-in-one otion on the same attribute when this otion was higher riced ( MA1(A+BC).44 vs. MC1(ABC+).90). The corresonding effect for the second attribute dislayed a similar attern ( MB2(AB+C).49 vs. MC2(ABC+).88), and the asymmetric nature of the enhancement effect was significant across the two attributes ( F(1, 219) 7.06;!.01). This finding rovides further suort for the roosed comensatory reasoning mechanism of the observed effects. The Imact of Price Disersion on Choice. Each of the 74 resondents selected one of the alternatives in each of the four sets, which yielded 296 observations. The choice shares of otions A, B, and C in each of the four conditions are summarized in table 3. The data show that the choice share of the all-in-one otion was greater when one of the otions was riced at a remium than when all otions were riced at arity. In articular, the choice share of the all-inone otion C was greater when its rice exceeded that of the secialized otions than when it was riced at arity 2 with the other otions (54.1% vs. 74.3%; x (1) 6.47;.01). This finding is consistent with the notion that the attractiveness of the all-in-one otion on the differentiating attributes is likely to increase when this otion is riced at a remium. More imortant, the increase in the choice share of the all-in-one otion (relative to the rice-arity condition) was significantly greater when this otion was the highest riced otion than when either of the secialized otions was highest riced (54.1% vs. 74.3% vs. 60.8% vs. 2 58.1%; x (1) 6.05;.01). Furthermore, the imact of arity ricing on choice shares of the alternatives was asymmetric for the all-in-one and the secialized otions. In articular, while increasing the rice of the all-in-one otion resulted in an increase in its choice share, the imact of increasing the rice of the secialized otions was less consistent, leading to a minor decrease in the share of the higher riced otion A (25.7% vs. 24.3%) and an increase in the share of the higher riced otion B (20.3% vs. 25.7%). In this context, the increase in the choice share of the all-in-one otion was significantly greater than the increase in the share of the secialized otions 2 ( x (1) 15.64;!.001), lending further suort for the exerimental redictions. Discussion The data furnished by this exeriment show that the erceived attribute erformance of the all-in-one otion is likely TABLE 3 CHOICE SHARE AS A FUNCTION OF PRICE DISPERSION (EXPERIMENT 2) Otion Price Equal C + A + B + A 25.7 14.9 24.3 16.2 B 20.3 10.8 14.9 25.7 C 54.1 74.3 60.8 58.1 NOTE. Percentages in each of the cells indicate the choice share of the corresonding otion. For all cells, N 74.
PRODUCT DIFFERENTIATION AND COMPENSATORY REASONING IN CHOICE 441 to be greater when this otion is riced at a remium relative to secialized otions than when it is riced at arity. More imortant, the increase in the erceived attribute erformance associated with remium ricing was greater for the all-in-one otion than for secialized otions. This asymmetric nature of the imact of rice on erformance ratings and choice suggests that the observed effects cannot be accounted for solely by rice-quality inferences. Indeed, in the absence of devaluation of the all-in-one otion, it is difficult to account for the asymmetric increase in the attractiveness of the all-in-one versus the secialized otions. Overall, this data attern is consistent with the roosition that increasing the rice of the all-in-one otion introduces a dimension on which this otion is deficient, which, in turn, eliminates the basis for comensatory reasoning and decreases the likelihood of consumers devaluing that otion. An imortant consideration raised by the ricing maniulation is that the disersion of otions rices might have changed the nature of the trade-offs made by resondents. Indeed, when all otions are riced at arity and otion C is erceived to be inferior on the attributes dominated by secialized otions, consumers are faced with a trade-off between maximizing the erformance on a articular attribute (choosing one of the secialized otions) and settling for moderate quantities of each attribute (choosing the allin-one otion). In contrast, when the all-in-one otion is riced at a remium (and hence is less likely to be devalued), it will be erceived as combining the good features of the secialized otions, which then eliminates the need for trading off the two erformance-related attributes. Instead, the decision now involves a trade-off between erformance and rice. The exected increase in the share of the all-in-one otion in this case is consistent with the research documenting consumer reference for erformance (quality) over rice (Carmon and Simonson 1998; Hardie, Johnson, and Fader 1993; Simonson, Nowlis, and Lemon 1993; Simonson and Tversky 1992). In this context, choosing the all-in-one alternative eliminates the need to trade off two erformancerelated attributes and instead ositions the all-in-one otion as the rice-based comromise. It is imortant to note, however, that because the actual attribute values of the secialized and the all-in-one otions are identical, the roosition that varying rice might have changed the nature of the trade-offs is meaningful only in the resence of comensatory inferences. GENERAL DISCUSSION This research examines consumer reactions to two common ositioning strategies: a secialized strategy in which an otion is differentiated by a single feature, and an allin-one strategy in which an otion is differentiated by a combination of features. It was roosed that when making choices from sets comrising both secialized and all-inone otions, consumers are likely to adot a zero-sum heuristic, in which they equate the overall attractiveness of choice alternatives and evaluate ambiguous information in a comensatory fashion. Two emirical studies reorted in this article demonstrate that the erceived attractiveness of the attributes differentiating the all-in-one otion tends to decrease in the resence of secialized otions. In contrast, adding an all-in-one otion to a set made u of secialized otions tends to olarize the erceived erformance of these otions: it enhances the attractiveness of the attributes differentiating each of the secialized otions, while devaluing the erformance of these otions on their secondary, nonfeatured attributes. These comensatory enhancement and devaluation effects were shown to be a function of otions erceived erformance on unobservable attributes, such that these effects were less ronounced when the otions erformance on unobservable attributes was inferred to be inconsistent with their erformance on the readily observed attributes. To account for the reorted effects, this research introduced the zero-sum heuristic as a general mechanism for drawing comensatory evaluations in choice. Here, the zerosum heuristic is defined as a context-based inference strategy in which an otion s values are derived from the other alternatives in the set. Thus, the zero-sum heuristic reflects consumers belief that otions in a given choice set are balanced, such that advantages on one dimension are likely to be comensated for by disadvantages on another. As a result, when an all-in-one alternative is embedded among secialized otions, its attribute erformance is likely to be devalued so that it matches the overall erformance of these otions. Similarly, the erformance of secialized otions on their rimary attributes is bolstered and their secondary attributes are devalued so that their overall erformance matches that of the all-in-one otion. Research reorted in this article contributes to the decision literature by alying the notion of comensatory inferences to exlain the devaluation and enhancement effects associated with evaluating secialized and all-in-one otions. Building on the findings reorted by Chernev and Carenter (2001), who examine market efficiency inferences as a function of rice, the research resented in this article documents a different tye of comensatory inference that is not necessarily contingent on the availability of rice information. Indeed, because drawing inferences about overall value is difficult in the absence of rice information, the availability of rice information is a necessary condition for market efficiency intuitions to occur. In contrast, the zero-sum heuristic reorted in this research offers a more general decision strategy that is not necessarily contingent on the availability of ricing information. Documenting the existence of comensatory reasoning without exlicitly available ricing information is an imortant contribution to the research on consumer decision making and choice. It is imortant to note that the observed effect of adding an all-in-one otion to a set comrising secialized otions (and vice versa) is concetually different from the range effects described by rior research (Janiszewski and Lichtenstein 1999; Lynch, Chakravarti, and Mitra 1991). Indeed, in the case of range effects, adding an otion to the choice set changes the attribute-value scale describing choice al-
442 JOURNAL OF CONSUMER RESEARCH ternatives. In contrast, in the case of the comensatory effects reorted in this research, adding either a secialized or an all-in-one otion does not change the scale describing choice alternatives because both the secialized and the allin-one otions are described by identical attribute values (e.g., both offer cavity revention ). The findings reorted in this research also imly that the widely used strategy of ricing secialized and all-in-one otions at arity might, in fact, be subotimal. Thus, the data show that the choice share of the all-in-one otion is likely to be greater when this otion is riced higher than when it is riced at arity with the secialized otions. It is imortant to note, however, that this increase in the share of the all-in-one otion when riced at a remium is not a universal rule; it is contingent on a variety of other factors, such as the rice sensitivity of consumers. Thus, even though increasing the rice of the all-in-one otion might mitigate the devaluation effect, in cases when consumer rice sensitivity is high, the overall choice share might nevertheless decrease. An imortant issue raised by this research concerns identifying factors that facilitate the use of the zero-sum heuristic vis-à-vis alternative decision heuristics. In this context, three imortant reconditions for the zero-sum heuristic can be identified. The first one is that otions in the choice set are erceived as being a art of the same subset of alternatives for examle, a art of a comany s roduct line or a art of the assortment offered by a retailer. The second recondition is the belief that all relevant information is available and choice alternatives can be readily evaluated. Finally, the third recondition is that consumers believe that otions in the choice set are balanced in terms of their overall attractiveness. Such balance might be due to the erceived market efficiency (in cases involving rice and/or rice romotions; Chernev and Carenter 2001; see also Simonson, Carmon, and O Curry 1994), technological constraints (such as the relationshi between size and weight; Hauser and Shugan 1983), or commonly observed correlations (such as reliability and warranty; Broniarczyk and Alba 1994). The research resented in this article focused on documenting the secific effects associated with the use of the zero-sum heuristic and identifying some of its moderating factors, without exlicitly examining the influence of all the above reconditions on the strength of the zero-sum heuristic. Identifying antecedents and the boundary conditions of the zerosum heuristic is an imortant and fruitful area for further research. APPENDIX TABLE A1 STIMULI OVERVIEW (EXPERIMENT 1) Product Otion A Otion B Otion C Cold relief medicine Relieves chest congestion Clears nasal stuffiness Relieves chest congestion and clears nasal stuffiness Laundry detergent Prevents static Removes stains Prevents static and removes stains Shaving cream Moisturizes Protects Moisturizes and rotects Toothaste Prevents cavities Whitens teeth Whitens teeth and revents cavities Vitamin sulements Imrove memory Reduce stress Imrove memory and reduce stress TABLE A2 STIMULI OVERVIEW (EXPERIMENT 2) Product Otion A Otion B Otion C Price* Laundry detergent Protects color Removes stains Protects color and removes stains $7.99 ($9.99) Shaving cream Moisturizes Protects Moisturizes and rotects $2.99 ($3.99) Toothaste Prevents cavities Protects from tartar Prevents cavities and rotects from tartar $1.99 ($2.99) Vitamin sulements Ginseng Ginkgo biloba With ginseng and ginkgo biloba $6.99 ($8.99) *Premium rice is given in arentheses.
PRODUCT DIFFERENTIATION AND COMPENSATORY REASONING IN CHOICE 443 REFERENCES Adler, Alfred (1924), The Practice and Theory of Individual Psychology, 2nd rev. ed., London: Kegan Paul. Allison, Ralh I. and Kenneth P. Uhl (1964), Influence of Beer Brand Identification on Taste Percetion, Journal of Marketing Research, 1 (August), 36. Baumgartner, Hans (1995), On the Utility of Consumers Theories in Judgments of Covariation, Journal of Consumer Research, 21, 10. Bernoulli, Daniel (1738), Secimen Theoriae Novae de Mensura Sortis, Commentarii Academiae Scientiarum Imerialis Petroolitanae, 175 92; English translation in Bernoulli, Daniel (1954), Exosition of a New Theory on the Measurement of Risk, trans. Louise Sommer, Econometrica, 22, 23 36. Bettman, James R., Deborah Roedder John, and Carol A. Scott (1986), Covariation Assessment by Consumers, Journal of Consumer Research, 13 (December), 316 26. Broniarczyk, Susan M. and Joseh W. Alba (1994), The Role of Consumers Intuitions in Inference Making, Journal of Consumer Research, 21 (December), 393 407. Carmon, Ziv and Itamar Simonson (1998), Price-Quality Trade- Offs in Choice versus Matching: New Insights into the Prominence Effect, Journal of Consumer Psychology,7(4), 323 43. Carenter, Gregory S. and Kent Nakamoto (1989), Consumer Preference Formation and Pioneering Advantage, Journal of Marketing Research, 26 (August), 285 98. Chernev, Alexander and Gregory S. Carenter (2001), The Role of Market Efficiency Intuitions in Consumer Choice: A Case of Comensatory Inferences, Journal of Marketing Research, 38 (August), 349 61. Chernev, Alexander and Ryan Hamilton (2007), Too Much of a Good Thing? Otion Attractiveness and Assortment Choice, working aer, Kellogg School of Management, Northwestern University, 2001 Sheridan Road, Evanston, IL 60208. Dick, Alan, Diankar Chakravarti, and Gabriel Biehal (1990), Memory-Based Inferences during Consumer Choice, Journal of Consumer Research, 17 (June), 82 93. Hardie, Bruce G. S., Eric J. Johnson, and Peter S. Fader (1993), Modeling Loss Aversion and Reference Deendence Effects on Brand Choice, Marketing Science, 12, 17. Hauser, John R. and Steven Shugan (1983), Defensive Marketing Strategies, Marketing Science, 4 (Fall), 319 60. Huber, Joel and John McCann (1982), The Imact of Inferential Beliefs on Product Evaluations, Journal of Marketing Research, 19 (August), 324 33. Janiszewski, Chris and Donald R. Lichtenstein (1999), A Range Theory Account of Price Percetion, Journal of Consumer Research, 25 (March), 353 68. Janiszewski, Chris and Stijn M. J. Van Osselaer (2000), A Connectionist Model of Brand-Quality Associations, Journal of Marketing Research, 37, 331 50. Johnson, Eric J. and Robert J. Meyer (1984), Comensatory Choice Models of Noncomensatory Processes: The Effect of Varying Context, Journal of Consumer Research, 11 (June), 528 41. Johnson, Richard D. (1987), Making Judgments When Information Is Missing: Inferences, Biases, and Framing Effects, Acta Psychologica, 66 (1), 69 82. Johnson, Richard D. and Irwin P. Levin (1985), More Than Meets the Eye: The Effect of Missing Information on Purchase Evaluations, Journal of Consumer Research, 12 (Setember), 169 77. Kahneman, Daniel and Amos Tversky (1979), Prosect Theory: An Analysis of Decision under Risk, Econometrica, 47 (March), 263 92. Keeney, Ralh L. and Howard Raiffa (1976), Decisions with Multile Objectives : Preferences and Value Tradeoffs. New York: Wiley. Kivetz, Ran and Itamar Simonson (2003), The Idiosyncratic Fit Heuristic: Effort Advantage as a Determinant of Consumer Resonse to Loyalty Programs, Journal of Marketing Research, 40 (February), 454 67. Lichtenstein, Donald R. and Scot Burton (1989), The Relationshi between Perceived and Objective Price-Quality, Journal of Marketing Research, 26 (November), 429 43. Lynch, John G., Diankar Chakravarti, and Anusree Mitra (1991), Contrast Effects in Consumer Judgments: Changes in Mental Reresentations or in the Anchoring of Rating Scales? Journal of Consumer Research, 18 (December), 284 97. Nowlis, Stehen M. and Itamar Simonson (1996), The Effect of New Product Features on Brand Choice, Journal of Marketing Research, 33 (February), 36 46. Payne, John W., James R. Bettman, and Eric J. Johnson (1993), The Adative Decision Maker. New York: Cambridge University Press. Prelec, Drazen, Birger Wernerfelt, and Florian Zettelmeyer (1997), The Role of Inference in Context Effects: Inferring What You Want from What Is Available, Journal of Consumer Research, 24 (June), 118 25. Raghubir, Priya and Kim Corfman (1999), When Do Price Promotions Affect Pretrial Brand Evaluations? Journal of Marketing Research, 36, 211 22. Rao, Akshay R. and Kent B. Monroe (1989), The Effect of Price, Brand Name, and Store Name on Buyers Percetions of Product Quality: An Integrative Review, Journal of Marketing Research, 26, 351 57. Shiv, Baba, Ziv Carmon, and Dan Ariely (2005), Placebo Effects of Marketing Actions: Consumers May Get What They Pay For, Journal of Marketing Research, 42 (November), 383 93. Simonson, Itamar (1989), Choice Based on Reasons: The Case of Attraction and Comromise Effects, Journal of Consumer Research, 16 (Setember), 158 74. Simonson, Itamar, Ziv Carmon, and Suzanne O Curry (1994), Exerimental Evidence on the Negative Effect of Product Features and Sales Promotions on Brand Choice, Marketing Science, 13 (1), 23 40. Simonson, Itamar, Stehen Nowlis, and Katherine Lemon (1993), The Effect of Local Consideration Sets on Global Choice between Lower Price and Higher Quality, Marketing Science, 12 (4), 357 77. Simonson, Itamar and Amos Tversky (1992), Choice in Context: Tradeoff Contrast and Extremeness Aversion, Journal of Marketing Research, 29 (August), 281 95. Torgerson, Warren S. (1958), Theory and Methods of Scaling. Oxford: Wiley.
444 JOURNAL OF CONSUMER RESEARCH Tversky, Amos (1972), Elimination by Asects: A Theory of Choice, Psychological Review, 79 (October), 281 99. Tversky, Amos and Daniel Kahneman (1991), Loss Aversion in Riskless Choice: A Reference-Deendent Model, Quarterly Journal of Economics, 106 (November), 1039 61. Von Neumann, John and Oskar Morgenstern (1953), Theory of Games and Economic Behavior, 3rd ed., New York: Wiley. Wernerfelt, Birger (1988), Umbrella Branding as a Signal of New Product Quality: An Examle of Signalling by Posting a Bond, Rand Journal of Economics, 19, 9. (1995), A Rational Reconstruction of the Comromise Effect: Using Market Data to Infer Utilities, Journal of Consumer Research, 21 (March), 627 33. Winer, B. J., Donald R. Brown, and Kenneth M. Michels (1991), Statistical Princiles in Exerimental Design, 3rd ed., New York: McGraw-Hill.