The Neglect of Prescreening Information in Two-Stage Decision Tasks AMITAV CHAKRAVARTI CHRIS JANISZEWSKI GÜLDEN ÜLKÜMEN*

Transcription

1 1 The Neglect of Prescreening Information in Two-Stage Decision Tasks AMITAV CHAKRAVARTI CHRIS JANISZEWSKI GÜLDEN ÜLKÜMEN*

2 * Amitav Chakravarti is Assistant Professor of Marketing, New York University, 44 West Fourth Street, Suite 9-83, New York, NY (Phone: ; Fax: ). Chris Janiszewski is Faricy Professor of Marketing, University of Florida, P.O. Box , Gainesville, FL (Phone: Ext. 1240#; Fax: ). Gülden Ülkümen is a doctoral student, New York University, 44 West Fourth Street, Suite 9-175, New York, NY (Phone: ; Fax: ). This paper benefited from workshop participants at Duke University, HKUST, INSEAD, the National University of Singapore, the University of Chicago, the University of Colorado, and the University of Florida. All three authors contributed equally to this work. 2

3 3 Abstract Several studies show that information used to screen alternatives becomes less important relative to information acquired latter in the search process simply because it was used to screen. Experiment 1 shows that the tendency to deemphasize prescreening information leads to systematically different choices for decision makers who screen alternatives, in comparison to decision makers who do not screen alternatives. Additional studies show that screening encourages decision makers to place an inordinate amount of emphasis on the postscreening information (experiment 2). Prescreening information is deemphasized owing to the perceptual categorization that occurs when people create a consideration set of retained alternatives (experiments 3 and 4). Together, the results show that a brand s strength of consideration (i.e., how highly an option ranks on screening criteria) may have little influence on the likelihood it is chosen in a post-screening choice process.

4 4 When consumers encounter a large number of options, they often use a two-stage decision process (Beach 1993; Beach and Mitchell 1987; Bettman and Park 1980; Billings and Marcus 1983; Crow, Olshavsky, and Summers 1980; Howard and Sheth 1969; Nedungadi 1990; Payne 1976; Roberts and Lattin 1991). In stage one, options are screened to create a consideration set. In stage two, the remaining options are compared to make a final choice. This decision strategy is popular because it reduces the decision maker s workload by paring down the number of options to be examined at the final choice stage. A reduced number of options at choice allows the decision maker to consider more information about each option, which should lead to higher quality choices (e.g., Alba et al. 1997; Haubl and Trifts 1999; Lynch and Ariely 2000; Roberts and Nedungadi 1995). Thus, decision tools (e.g., web-based screening tools) that allow consumers to structure and organize their decision environments have been strongly endorsed by both the popular press and consumer decision researchers (e.g., Alba et al. 1997; Diehl, Kornish, and Lynch 2003; Häubl and Trifts 2000; Lynch and Ariely 2000; Widing and Talarzyk 1993). In contrast to this prevailing sentiment, there is a small body of research suggesting that there may be situations where screening choice sets leads to lower decision quality (Diehl et al. 2003; Van Zee, Paluchowski, and Beach 1992; Wright and Barbour 1977). For example, Diehl et al. (2003) argue that sorting alternatives on one attribute (e.g., quality) encourages people to make decisions using nonsorting attributes. If the nonsorting attribute is an important variable, like price, then decision quality should improve, as Diehl et al. show. If the nonsorting attribute is a less important variable, then decision quality may suffer. Placing more emphasis on the less important attribute may change the decision outcome relative to a situation in which all

5 5 information about a brand is considered concurrently (i.e., situations in which there is no screening of alternatives). In this paper, we show that the act of screening alternatives (e.g., during consideration set formation) encourages a person to neglect the prescreening information, and emphasize the postscreening information, in a subsequent choice (experiment 1). We argue that this tendency to discount the prescreening information may be a consequence of two processes. First, the act of screening is often a noncompensatory process in which alternatives that pass a cutoff are retained (i.e., conjunctive rule) or alternatives that fail to meet a cutoff are rejected (i.e., elimination-byaspects rule). To the extent the noncompensatory choice encourages a person to view prescreening information as already noted, the prescreening information may be ignored, even if it is the most diagnostic information. As a consequence, there should be an increased emphasis on the post-screening information (experiment 2). Second, the act of screening encourages a person to create a category of retained alternatives. Categorization encourages a person to view alternatives within a category as more similar (Goldstone, Lippa, and Shiffrin 2001). As a consequence, the person may neglect the prescreening information owing to its perceived homogeneity (experiments 3 and 4). BACKGROUND A large body of research has demonstrated the contingent nature of decision making (Beach and Mitchell 1978; Einhorn and Hogarth 1981; Newell and Simon 1972; Payne 1976, 1982; Svenson 1979). The selection of information processing strategies is shown to be highly contingent on the characteristics of the task, environment, and person characteristics (for an

6 6 extensive review, see Ford et al. 1989). Among task characteristics, task complexity has been most widely investigated with respect to its effect on information processing. Task Complexity The literature investigating the influence of task complexity on choice processes is extensive. Studies show that the selection, evaluation, and integration of information are all affected by task complexity. Of particular interest for this investigation is the complexity created by having to consider a large number of alternatives. Process tracing studies find that large choice sets are often reduced using noncompensatory screening strategies (Billings and Marcus 1983; Johnson and Meyer 1984; Olshavsky 1979; Onken, Hastie, and Revelle 1985; Payne 1976; Payne and Braunstein 1978; Wright and Barbour 1977). For example, Payne (1976) finds that when consumers encounter a two-alternative choice situation, they employ compensatory decision strategies. However, in the context of a more complex, multi-alternative decision task, consumers attempt to reduce the number of alternatives using a conjunctive or elimination by aspects rule. Similarly, Timmermans (1993) finds that consumers become more likely to form a consideration set as the number of available alternatives increases from three (21%) to six (31%) to nine (77%). Increases in task complexity also increase the variability of decision rules used within the same decision task. Montgomery and Svenson (1976) suggest that different decision rules may be applied in succession in the same decision situation. In support of this proposition, there is general agreement that the proportion of intra-alternative processing increases with each additional stage of a decision process (Payne, Bettman, and Johnson 1993). Correspondingly, verbal protocols in Payne (1976, Experiment 1) illustrate that participants often began a task using simplifying strategies. Once they had reduced the number of alternatives and dimensions

7 7 using less cognitively demanding strategies, the participants switched to a more elaborate, compensatory process in order to evaluate the remaining alternatives. Subsequent studies reported similar findings regarding the phased nature of decision making (Bettman and Park 1980; Billings and Marcus 1983; Crow, Olshavsky, and Summers 1980). These results clearly indicate that when decision makers confront a complex choice task, they often employ a two-stage choice process. In the first stage, a noncompensatory process is used to reduce the number of alternatives to a set of acceptable candidates. Subsequently, a compensatory process is used to select one alternative from among the remaining alternatives (Beach and Mitchell 1987). The first stage of this process is called screening, and it can be defined as the process that governs the admission of options to the consideration set (Beach 1993; Nedungadi 1987). Screening Historically, the screening stage of a choice process has been viewed as involving a relatively simple act that merely reduces the amount of information to be considered in the more complex and elaborate choice stage (e.g., Bettman 1979; Huber and Kline 1991; Johnson and Payne 1985). Outside of the marketing literature, the emphasis is often on how certain screening procedures (e.g., screen alternatives by inclusion / exclusion) influence the size of the resulting consideration set (see Levin et al for discussion). Within the marketing literature, research has been largely limited to the descriptive issues of (1) verifying whether consumers screen alternatives and create consideration sets (e.g., Roberts and Lattin 1991) and, (2) verifying the size of these consideration sets (e.g., Hauser and Wernerfelt, 1990). This focus on descriptive issues has come largely at the expense of process related issues. For example, does the act of screening systematically affect information processing at the final choice stage? Consequently,

8 8 do people who screen process information and make choices differently from those who make their decisions without screening? There are three pieces of evidence that suggest consideration set formation may alter the manner in which prescreening information is considered at choice. First, Wright and Barbour (1977) show that people are reluctant to reconsider information that has previously been used to reduce the size of a choice set. Wright and Barbour (1977) asked respondents to use a dimensional conjunctive rule to choose one alternative from among 24 alternatives described on five attributes. Respondents were asked to apply the cutoffs conjunctively during the screening phase and then choose one of the remaining options post-screening. Some respondents were also provided with a suggested sequence in which the prescreening attributes should be examined. Unbeknownst to the subjects at the initiation of the screening task, the initial screening phase resulted in either all options being discarded or the sole remaining option being out of stock. When respondents were asked to return to the original set of options, they tended to restrict their analysis to alternatives they had retained in the last iteration of the sequential screening process. In other words, participants were extremely reluctant to go back to the options that were rejected in the earlier rounds of the screening process, even when it was made clear that the criterion for screening (attribute sequence) had been fairly arbitrary. According to Wright and Barbour (1977) arriving at the end of a decision making phase (i.e., a round of screening involving a single attribute) closes the phase and, consequently, the details of the information used in that phase become vague once the next phase is initiated. Second, Diehl et al. (2003) argue screening may also influence the manner in which consumers use attribute information to make a choice. Diehl et al. (2003) show that in a two attribute environment, a screening tool that sorts exclusively on quality increases sensitivity to

9 9 price (i.e., a postscreening attribute). The authors maintain that sorting products into an ordered list results in over-sampling the more attractive options. To the extent these retained options are more homogeneous on quality, relative to a random draw of retained options, consumers should become more sensitive to price. The results of this study indicate that decision makers who screen are willing to put more weight on post-screening information, although it is not clear how much of this shift in attribute weight is a function of the act of screening and how much is a function of the increase in the objective homogeneity of the retained alternatives on the prescreening information. Third, Van Zee et al. (1992) examined the effect of screening information on the prechoice evaluations of the surviving options. Participants in Experiment 1 were presented with information regarding five apartments, described on five different attributes. They were then asked to find a suitable rental for a friend, based on the criteria that the friend had provided. Participants were first given cutoffs for three of the five attributes. Participants in the control group screened the options using the three criteria and then rated each remaining option s attractiveness. Participants in the four experimental groups screened alternatives using the three criteria, then received additional postscreening information (i.e., attribute values on the remaining two attributes) about their surviving options. These participants rated the attractiveness of the surviving options prior to making a final choice (i.e., prechoice). The authors tried to detect the differential impact of the prescreening and postscreening information on the groups prechoice attractiveness ratings by varying the correlation of the prescreening information relative to the postscreening information across the four experimental groups. The results indicated that the treatment group s prechoice ratings reflected the postscreening information closely, while the prescreening information was ignored. Van Zee et al. (1992) refer

10 10 to the observed phenomenon as the screening effect, and they indicated that this effect can lead to suboptimal choices. Together, the three studies suggest screening alternatives to create a consideration set can encourage people to ignore prescreening information and emphasize postscreening information. Van Zee et al. (1992) provide the closest test of this hypothesis, yet their demonstration does have some problems that limit our confidence in their conclusions. First, their conclusions relied on visual inspection of the data or on correlations of the ratings provided by the treatment and control groups, rather than on deductions from a more formal experimental design. Second, their prescreening and postscreening information presentation formats differed in that the prescreening procedure simply displayed the information, whereas the postscreening procedure asked participants to value each piece of postscreening information using an eight-point attractiveness scale. Rating the attractiveness of each piece of postscreening information, but not each piece of prescreening information, could have encouraged participants to place greater emphasis on the postscreening information. Third, and perhaps most importantly, their stimuli were constructed such that the retained alternatives often exhibited no variance on the prescreening attributes (e.g., prescreening rents of either $250/month or $300/month with a cutoff of $250/month). As Diehl et al. (2003) show, if the retained alternatives exhibit objective homogeneity on prescreening attribute values, then the prescreening attributes will be nondiagnostic and noninfluential in choice. Screening Effect Although there are findings in the literature that hint at a screening effect, the only explanation that supports the screening effect is that offered by Diehl et al. (2003), namely that the prescreening attributes become nondiagnostic owing to homogeneous values. We expect

11 11 there may be two additional reasons that prescreening information gets deemphasized when additional information is encountered post-screening. First, the act of screening is commonly a noncompensatory process. As argued by Wright and Barbour (1977), engaging in a noncompensatory process encourages a person to treat prescreening information as used. If information is viewed as used, it may have less influence when people subsequently shift to a compensatory strategy post-screening. A second reason for a screening effect is the process of screening may encourage people to perceive the prescreening information as more homogeneous, hence less diagnostic postscreening. Investigations into perceptual categorization suggest that alternatives that are grouped within a category are perceived as more similar as a consequence of the categorization, whereas alternatives that are grouped in different categories are perceived as less similar as a consequence of categorization. For example, Goldstone (1994) shows that categorizing simple figures (e.g., squares) as belonging to the same group decreases one s ability to discriminate among pairs of the figures using same/different judgments. Livingston, Andrews, and Harnard (1998) find that learning to classifying objects (e.g., animal body parts, microorganisms) into categories increases the perceived similarity of objects. Goldstone et al. (2001) show that categorizing objects into the same category increases their shared similarity with an unclassified object, a result that could only occur if the classed objects became perceptually more similar. Finally, Werker and Tees (1984) show that infants lose their ability to discriminate between speech sounds as a consequence of learning that the sounds come from the same phonetic category. These results suggest screening may not only produce increases in the objective similarity of the retained alternatives, as claimed by Diehl et al. (2003), but also increases in the subjective similarity of the retained alternatives as well. These increases in the subjective

12 12 similarity of the prescreening information may encourage people to deemphasize the prescreening information when considered along with postscreening information. Summary We hypothesize that screening will reduce the emphasis on prescreening information and increase the emphasis on postscreening information. We further hypothesize that this screening effect may occur for either of two reasons. First, the use of a noncompensatory choice strategy at prescreening discourages a person from reconsidering the prescreening information once additional postscreening information is encountered. Second, the act of screening encourages the formation of a category of retained alternatives and categorization increases the perceived similarity of alternatives within this category. If prescreening information is perceived to be homogeneous, then there is no reason to reconsider the information postscreening. In experiment 1, we investigate the screening hypothesis. If our hypothesis is correct, we should be able to observe this influence in an option set where prescreening and postscreening attributes are negatively correlated (cf. Widing and Talarczyk 1993). For example, suppose six brands are described on more differentiating prescreening attributes and less differentiating postscreening attributes. Brands A, C, and E perform best on the prescreening attributes and are most likely to be retained in the event a decision maker is asked to screen the options. If the three brands exhibit a preference pattern of A > C > E on the prescreening attributes, but E > C > A on the postscreening attributes, then shifts in the relative amount of emphasis placed on prescreening and postscreening attributes should influence choice shares. If decision makers choose without screening, and rely primarily on the more differentiating prescreening attributes to make this choice, they should prefer brand A. If decision makers initially screen, and screening reduces the emphasis placed on prescreening attributes and increases the emphasis

13 13 placed on postscreening attributes, they should prefer brand E. Thus, the act of screening should systematically shift preferences from the brand that performs best on prescreening attributes to the brand that performs best on postscreening attributes. EXPERIMENT 1 The goal of experiment 1 was to demonstrate that systematic differences in choice shares can arise when consumers make choices without screening versus with screening. Participants in the control conditions were encouraged to choose a brand from among six brands described on six attributes. Participants in the screening condition were encouraged to reduce the size of the choice set using four prescreening attributes, then consider the complete set of attribute information when choosing among the remaining brands. It was expected that screening condition participants would neglect prescreening attributes during a subsequent choice task and rely on the postscreening attributes to make their choice. Method Stimuli. The stimuli were six brands of microwave popcorn described on six attributes adopted from Zhang and Markman (2001) (see appendix). The first two prescreening attributes were cost per serving and level of sodium and had identical performance levels across the six brands (i.e., they were common attributes). The next two prescreening attributes were alignable attributes. A pretest had shown that crunchiness and calories per serving were the two most desirable attributes among a large array of potential attributes (see Zhang and Markman 2001). The brands performances on these attributes were graded so that the brands preference rank was likely to be A, C, E, B, D, F. The final two postscreening attributes for any one brand were nonalignable attributes, hence were unique to each brand (i.e., there were 12 additional

14 14 attributes). These postscreening attributes were assigned to brands so as to negatively correlate with the top three brands performance on the prescreening attributes (i.e., brand preference based on only the postscreening attributes was expected to be E, C, A, B, D, F). Design and Procedure. The design included two choice groups and a screening group. The partitioned choice group and screening group followed a three-part procedure that varied only in the screening activity (see appendix). At time one, participants were seated at a computer and asked to rank the desirability of the best level of each prescreening attribute (e.g., crunchiness lasts 3.5 hours, calories equal to less than a slice of bread) plus the 12 postscreening attributes. At time two, participants were told they needed to purchase some microwave popcorn and that partial descriptions (i.e., the prescreening attributes) for the six available brands were listed in the middle of the screen. Then participants in the partitioned choice group were told After you have gone through these descriptions to your satisfaction, we will provide you with more information on these six brands on the next screen. Please do not make up your mind about your final choice yet; simply take a careful look at the information provided to you so far. Participants in the screening group were told After you have gone through these descriptions to your satisfaction, as a first step towards picking a brand of your choice, please shortlist three brands that you would consider more seriously for purchase. We will provide you with more information on these three brands on the next screen. Please do not make up your mind about your final choice yet; simply select (i.e., shortlist) three brands that you think warrant further attention by clicking on the appropriate buttons below.

15 15 At time three, the postscreening attribute information was added to the stimulus array and participants were asked to make a choice using all of the attribute information. Partitioned choice group participants chose from among six brands. Screening group participants were also presented with six brands, but were asked to choose only from among the three brands in their consideration set (e.g., Now, from the three brands that you had short listed earlier, please choose one brand as your final choice by clicking the appropriate button below ). The postscreening attributes were assigned so that a participant s third and fourth most desired attributes (i.e., the two most desired postscreening attributes) were paired with brand E (i.e., the third most desirable brand on the prescreening attributes), fifth and sixth most desired postscreening attributes were paired with brand C, and seventh and eighth most desired postscreening attributes were paired with brand A. The remaining postscreening attributes were assigned to the remaining brands as illustrated in the appendix. Thus, for each individual the prescreening attributes were negatively correlated with the postscreening attributes for the target brands of interest (i.e., A, C, E). The experimental design also included a free choice group. Free choice group participants were asked to rate the desirability of the attributes at time one, then were given all of the information (i.e., all six attributes) about all of the brands and asked to make a choice. The postscreening attribute information was assigned to the brands in the same manner as in the other two conditions. Thus, free choice group participants simply skipped part two of the procedure, where only the prescreening attribute information was presented. The free choice group participants were meant to experience a more ecologically valid procedure in which consumers review information and make a choice. We note that the free choice group and partitioned choice group are conceptually equivalent control groups, but that the free choice group is a less

16 16 internally valid control. The free choice group information presentation procedure (i.e., entire stimulus array at once) varied from the procedure used in the screening condition (i.e., a time two presentation of six brands described on four attributes followed by a time three presentation of six brands described on six attributes). Thus, the free choice group was our ecologically valid control group and the partitioned choice group was our internally valid control group. Results One hundred twenty participants from an undergraduate subject pool participated in the experiment. It was expected that participants in the choice (i.e., control) groups would prefer brand A, C, or E and that the majority would prefer brand A owing to its superior performance on the most desirables attributes. The choice shares in the partitioned choice condition (πˆ A =.610, πˆ C =.268, πˆ E =.049) and the free choice condition (πˆ A =.590, πˆ C =.231, πˆ E =.103) were consistent with this expectation (see table 1). Second, it was expected that screening condition participants would select brands A, C, and E as their consideration set. All of the participants did so. Finally, if screening encourages participants to deemphasize prescreening attributes during the choice task, then the choice share should shift from brand A to brand E since brand E has better postscreening attribute performance than brand A. The choice shares in the screening condition (πˆ A =.350, πˆ C =.275, πˆ E =.375) were consistent with this prediction. The choice share of brand A was lower in the screening group (πˆ A =.350) than in the partitioned choice group (πˆ A =.610; z = 2.43, p <.05) or the free choice group (πˆ A =.590; z = 2.20, p <.05). 1 The choice share of brand E was higher in the screening group (πˆ E =.375) than in the partitioned choice group (πˆ E =.049; z = 3.90, p <.05) or the free choice group (πˆ E = 1 All hypotheses involving choice shares are directional. One-tail z-tests were used for all comparisons of choice shares.

17 17.103; z = 3.00, p <.05). Choice shares among the partitioned choice and free choice conditions did not differ for brand A (z = 0.18, p >.05) or brand E (z = 0.91, p >.05). ************************* Place Table 1 about here ************************* Discussion Experiment 1 shows that screening options using a subset of attribute information makes the prescreening attributes less important in a subsequent choice. Participants preferred the brand that performed best on the prescreening attributes when they did not engage in screening, but were less likely to prefer this brand if they initially screened using these desirable attributes. Preference shifted to a brand that performed better on attributes that were introduced postscreening, even though these postscreening attributes were not as desirable. In general, people seem to ignore the magnitude of differences between the three surviving brands on the prescreening attributes. It is important to recognize that the design and stimuli used in experiment 1 rule out processes that might masquerade as a screening effect. First, it is not the case that the prescreening information was unimportant, and hence, ignored by the subjects in the final choice stage. Subjects rated the prescreening attributes as more attractive than the postscreening attributes and yet, deemphasized the prescreening information during final choice. Second, it is not the case that a small amount of brand differentiation on the prescreening attributes was overwhelmed by a large amount of brand differentiation on the postscreening attributes. Participants in the choice conditions also saw the postscreening information. These participants preferred the brand that performed best on the prescreening attributes, suggesting that the

18 18 prescreening attributes had more influence on choice than the postscreening attributes when screening was not required. Third, the shift in choice shares in the screening condition was not a function of an increased salience of the postscreening attributes owing to their introduction just before choice. The partitioned choice group also had postscreening attributes introduced just prior to choice, yet exhibited a pattern of choice shares that differed from the screening group. Thus, it appears that the act of screening itself, rather than the partitioned nature of the task, initiated a decision process that was responsible for the observed shift in choice shares. Thus far, we have argued that the screening effect occurs because people screen on attributes that contribute the most utility to the product, and then deemphasize these attributes at choice, relying on differences in less desirable, less attractive postscreening attributes. If people are assigning more weight to the postscreening attributes as a consequence of screening, then we should be able to show that people are sensitive to changes in the relative performance of the brands on the postscreening attributes. Moreover, people should be most sensitive to changes in the relative performance of the brands on the postscreening attributes when they engage in screening. EXPERIMENT 2 The goal of experiment 2 was to demonstrate that people become more sensitive to postscreening attribute information as a consequence of screening. Two stimulus sets were used to test this hypothesis. The first stimulus set had the same structure as the stimulus set used in experiment 1. Recall the postscreening attributes were assigned so that a participant s third and fourth most desired attributes were paired with brand E, fifth and sixth most desired attributes were paired with brand C, and seventh and eighth most desired attributes were paired with brand

19 19 A. The second stimulus set assigned a more hedonically dispersed set of postscreening attributes to brands E, C, and A. Postscreening attributes were assigned so that a participant s third and fourth most desired attribute were paired with brand E, seventh and eighth most desired attribute were paired with brand C, and eleventh and twelfth most desired attribute were paired with brand A. It was expected that participants would be sensitive to the change in postscreening attribute hedonic dispersion, especially in the screening condition. In other words, the screening effect should become larger when the desirability of the postscreening attributes assigned to the consideration set brands becomes more dispersed. Method The experiment was a three (task format: partitioned choice / free choice / screening group) by two (hedonic dispersion of postscreening attributes: low / high) between-subject design. Except for the changes noted above, the stimuli were same as in experiment 1. The procedure was identical to the procedure used in experiment 1. Results Two hundred thirty-nine participants from an undergraduate subject pool participated in the experiment. Collapsing across the preference dispersion factor, the results replicate the results of experiment 1 (see table 2). Participants in the partitioned choice condition (πˆ A =.530, πˆ C =.217, πˆ E =.181) and the free choice condition (πˆ A =.577, πˆ C =.211, πˆ E =.141) were more likely to prefer brand A to C to E. Participants in the screening condition exhibited a shift in preference from brand A to brand E (πˆ A =.294, πˆ C =.224, πˆ E =.447). The choice share of brand A was lower in the screening group (πˆ A =.294) than in the partitioned choice group (πˆ A =.530; z = 3.20, p <.05) or the free choice group (πˆ A =.577; z = 3.69, p <.05). The choice share of brand E was higher in the screening group (πˆ E =.447) than in the partitioned choice

20 20 group (πˆ E =.181; z = 3.88, p <.05) or the free choice group (πˆ E =.141; z = 4.50, p <.05). Choice shares among the partitioned choice and free choice conditions did not differ for brand A (z = 0.59, p >.05) or brand E (z = 0.68, p >.05). Owing to the lack of difference between the partitioned choice and free choice conditions, the groups were collapsed for the hedonic dispersion analysis. The first key prediction of the experiment was that participants would be sensitive to the change in the hedonic dispersion of the postscreening attributes. This means that there should be a greater shift in market share in the high hedonic dispersion condition than in the low hedonic dispersion condition. To test this prediction, we did three calculations using the market shares for brands A and E. First, we confirmed there was a shift in choice shares in the low hedonic dispersion condition ( 2 (2) = 13.03, p <.05). In the choice groups, the share for Brand A was 58.3% and the share for brand E was 13.1%. In the screening group, the share for Brand A was 36.7% and the share for brand E was 34.7%. Thus, the shift in market share was 43.2% ([ ] - [ ] = 43.2) for the low hedonic dispersion condition. Then, we confirmed there was a shift in market shares in the high hedonic dispersion condition ( 2 (2) = 15.78, p <.05). In the choice groups, the share for Brand A was 51.4% and the share for brand E was 20.0%. In the screening condition, the share for Brand A was 19.4% and the share for brand E was 58.3%. Thus, the shift in market share was 70.2% ([ ] - [ ] = 70.2) for the high hedonic dispersion condition. Finally, we confirmed that the shifts in market share in the low and high hedonic dispersion conditions were different. A log-linear test for the three-way interaction was significant (G 2 (7) = 34.96, p <.01). This means the shift in choice shares in the high hedonic dispersion condition (70.2%) was greater than the shift in choice shares in the low hedonic dispersion condition (43.2%).

21 21 The second key prediction of the experiment was that hedonic dispersion should exert a greater influence in the screening conditions. For screening condition participants, brand A was less preferred in the high hedonic dispersion condition (πˆ A =.194) than in the low hedonic dispersion condition (πˆ A =.367; z = 1.82, p <.05). For the choice participants, brand A was equally preferred in the high hedonic dispersion condition (πˆ A =.514) and the low hedonic dispersion condition (πˆ A =.583; z = 0.86, p >.05). For screening condition participants, brand E was more preferred in the high hedonic dispersion condition (πˆ A =.583) than in the low hedonic dispersion condition (πˆ A =.347; z = 2.21, p <.05). For the choice participants, brand E was equally preferred in the high hedonic dispersion condition (πˆ A =.200) and the low hedonic dispersion condition (πˆ A =.131; z = 1.14, p >.05). Thus, the hedonic dispersion manipulation exerted its influence primarily in the screening conditions. ************************* Place Table 2 about here ************************* Discussion We have argued that screening prior to choice encourages decision makers to put less emphasis on the attributes they used to screen, and more emphasis on new attribute information, at choice. Experiment 2 provides evidence for the shift in the relative importance of prescreening and postscreening attributes during the choice task by showing that decision makers are sensitive to the hedonic dispersion of the postscreening attribute information. In addition, the hedonic dispersion of postscreening attributes exerts its influence primarily in the screening conditions. This finding provides direct support for the hypothesis that the act of screening encourages a decision maker to discount prescreening attribute information and increase the emphasis on

22 22 postscreening attribute information. The findings also suggest that the act of screening does not simply encourage people to use a noncompensatory choice process post-screening (c.f., Johnson and Meyer 1984). STUDY 3 The results of the first two studies suggest that screening alternatives encourages people to neglect prescreening information in the subsequent choice. Earlier, two processes were proposed to account for this effect. First, the process of screening may encourage the use of a noncompensatory process that subsequently discourages the consideration of the prescreening information at choice, assuming additional postscreening information is encountered. Second, the process of screening may encourage people to create a category of retained alternatives (i.e., the consideration set). Categorization encourages people to perceive within-group alternatives as more similar, hence put less emphasis on this information once postscreening information is encountered. In study 3, we tried to differentiate between these two potential sources of the screening effect. The challenge of study 3 was to develop a manipulation that allowed participants to screen the alternatives using a noncompensatory process, but to not think of the retained alternatives as a group. To this end, a screen-by-rejection condition was developed. In the screen-by-rejection condition, participants were asked to remove the alternatives that they did not want in the consideration set. Thus, similar to the screening condition, the screen-byrejection condition encouraged the use of a noncompensatory screening process, but was exclusionary instead of inclusionary. We expected that a consideration set created by rejecting undesirable alternatives should discourage participants from perceiving the remaining

23 23 alternatives as a group, and hence any screening effect attributable to categorical perception should be mitigated. This is in keeping with findings from past research that show that choice sets formed under exclusionary instructions tend to (a) be larger, (b) contain more ambiguous items, (c) require less effort, and (d) create more loosely held group perceptions (e.g., see Levin et al. 2001). In contrast, if the screening effect observed in experiments 1 and 2 is a consequence of a noncompensatory screening process, we should observe a screening effect in both the screening and screen-by-rejection groups. Method The experiment was a four cell (partitioned choice, free choice, screen-by-inclusion, screen-by-rejection) between-subject design. The stimuli were same as in experiment 1. The procedure for the partitioned choice, free choice, and screen-by-inclusion groups was identical to the procedure used in experiment 1. The procedure for the screen-by-rejection group mimicked the procedure of the other groups for the attribute ranking task performed at time one, and was then modified as follows. At time two, the screen-by-rejection group was told, After you have gone through these descriptions to your satisfaction, as a first step towards picking a brand of your choice, please reject three brands that you would not consider more seriously for purchase. We will provide you with more information on the three surviving brands on the next screen. Please do not make up your mind about your final choice yet; simply reject (i.e., throw away) three brands that you think do not warrant further attention by clicking on the appropriate buttons below. The screen-by-rejection group then followed a procedure identical to the screen-by-inclusion group.

24 24 Results Two hundred fifty participants from an undergraduate subject pool participated in the experiment. Choice shares for the four experimental groups are shown in table 3. The results for the partitioned choice, free choice, and screen-by-inclusion groups replicated the results of experiment 1. Participants in the partitioned choice condition (πˆ A =.594, πˆ C =.219, πˆ E =.125) and the free choice condition (πˆ A =.644, πˆ C =.237, πˆ E =.102) were more likely to prefer brand A to C or E. Participants in the screen-by-inclusion condition exhibited a shift in preference from brand A to brand E (πˆ A =.385, πˆ C =.215, πˆ E =.338). The choice share of brand A was lower in the screen-by-inclusion group (πˆ A =.385) than in the partitioned choice group (πˆ A =.594; z = 2.43, p <.05) or the free choice group (πˆ A =.644; z = 2.99, p <.05). The choice share of brand E was higher in the screen-by-inclusion group (πˆ E =.338) than in the partitioned choice group (πˆ E =.125; z = 2.97, p <.05) or the free choice group (πˆ E =.102; z = 3.34, p <.05). Choice shares among the partitioned choice and free choice conditions did not differ for brand A (z = 0.57, p >.05) or brand E (z = 0.40, p >.05). The critical analysis involved the screen-by-rejection group. Participants in the screenby-rejection group exhibited choice shares (πˆ A =.581, πˆ C =.194, πˆ E =.194) that were similar to the partitioned choice and free choice groups (i.e., showed a lack of a screening effect). The choice share of brand A in the screen-by-rejection group (πˆ A =.581) did not differ from the partitioned choice group (πˆ A =.594; z = 0.15, p >.05) or the free choice group (πˆ A =.644; z = 0.71, p >.05). The choice share of brand E in the screen-by-rejection group (πˆ E =.194) did not differ from the partitioned choice group (πˆ E =.125; z = 1.06, p >.05) or the free choice group (πˆ E =.102; z = 1.44, p >.05). The choice share of brand A in the screen-by-rejection group (πˆ A =.581) was higher than the choice share of brand A in the screen-by-inclusion group (πˆ A =

25 25.385; z = 2.25, p <.05). The choice share of brand E in the screen-by-rejection group (πˆ E =.194) was lower than the choice share of brand E in the screen-by-inclusion group (πˆ E =.338; z = 1.86, p <.05). One additional analysis was performed. It was possible that the failure to find a screening effect in the screen-by-rejection group was a consequence of fewer members of this group retaining brand E in their consideration set, as compared to members in the screen-by-inclusion group. If this were so, the choice share of brand E would be artificially suppressed in the screenby-rejection group. The percentage of participants including brand E in their consideration set was equivalent in the screen-by-rejection (73%) and the screen-by-inclusion (71%) groups (z = 0.02, p >.05). ************************* Place Table 3 about here ************************* Discussion Experiment 3 provides evidence that perceptual categorization contributes to the screening effect. When people screened alternatives by selecting brands to retain in their consideration set, a screening effect occurred. When people screened alternatives by rejecting brands from retention in their consideration set, a screening effect did not occur. It appears as if the active creation of a consideration set, with an emphasis on the prescreening information of the retained alternatives, encourages people to view the prescreening information as less relevant in a subsequent choice. Forming consideration sets in a less active manner, with an emphasis on rejecting alternatives, does not lead to a reduced emphasis on the prescreening information. Thus, using information to actively categorize reduces its importance when additional postscreening information is encountered prior to a choice.

26 26 The results of experiment 3 suggest that actively categorizing retained alternatives into a group (a consideration set) is a necessary precondition for the screening effect. However, the results do not provide insight into what aspects of this active categorization process is responsible for the screening effect. For example, it may be that inclusionary screening encourages more intensive consideration of the prescreening information that describes the retained alternatives and exclusionary screening encourages more intensive consideration of the prescreening information that describes the rejected alternatives. In other words, the neglect of prescreening information that we observe may simply be a function of the amount of effort put into its initial consideration, and that the act of forming a category is not critical to obtaining the screening effect. We refer to this as the greater-consideration-of-information hypothesis. Alternatively, it may be the case that inclusionary screening leads to the screening effect precisely because it encourages people to create a category using the prescreening information, and not simply because it encourages greater consideration of the prescreening information. We refer to this as the perceptual categorization hypothesis. Experiment 4 addresses these two explanations. EXPERIMENT 4 The goal of experiment 4 was to differentiate between the perceptual categorization and the greater-degree-of-consideration accounts of the screening effect. To investigate this issue, we used the partitioned choice, free choice, and screening groups from experiment 1 plus two additional groups the screen & rate group and the rate & screen group. In the screen & rate group, participants first went through the screening step, then they rated the brands on the prescreening information, and finally they chose a brand. In the rate & screen group, participants

27 27 first rated the brands on the prescreening information, then they went through the screening step, and finally they chose a brand. These two new groups allowed us to directly test two predictions that could differentiate the perceptual categorization account from the greater-consideration-ofinformation account of the screening effect. First, recall that the act of categorization tends to increase the perceived similarity of within-group members (Goldstone, Lippa, and Shiffrin 2001). If screening is encouraging categorization, then the ratings of the retained brands should be less dispersed after screening (as in the screen & rate group), as compared to before screening (as in the rate & screen group). Second, if screening encourages the perception of retained alternatives as a group, and this grouping is responsible for the screening effect, then anything that negates the perception of this grouping should remove the screening effect. For example, if screening encourages a person to view retained alternatives as a group and rating encourages a person to perceive alternatives individually, then rating followed by screening should create a screening effect, whereas screening followed by rating should not. In other words, inserting a rating task between the screening stage and the final choice stage, as in the screen & rate group, should weaken or switch off the screening effect. In contrast, the greater-consideration-of-information hypothesis predicts there should be a similar screening effect in a screen & rate group and rate & screen group because the amount of consideration of the prescreening information on the retained alternatives should be equivalent in each group. Method The experiment was a five cell (partitioned choice, free choice, screening, screen & rate, rate & screen groups) between-subject design. 2 The stimuli were same as in experiment 1. The 2 We also ran a rating group to confirm our assumption that rating encourages people to view alternatives individually and, thus, negates the screening effect. Participants first rated the brands on the prescreening

28 28 procedure for the partitioned choice, free choice, and screening groups was identical to the procedure used in experiment 1. The procedure for the screen & rate and rate & screen groups mimicked the procedure of the other groups with respect to the attribute ranking task performed at time one, and was then modified as follows. In the screen & rate group, respondents were asked to screen using a procedure identical to the screening group and then were told, You just shortlisted three brands (i.e., {list of brands}). However, before you make your final choice, we want to understand your evaluation of these three brands relative to the three brands that you did not shortlist. Please rate each of the six brands using the following scales. We will provide you with more information on these brands on the next screen. Please do not make up your mind about your final choice yet; simply rate the brands by clicking on the appropriate buttons below. After rating the six brands on a nine-point scale, the screen & rate group moved to the choice step and was told, Now, from the three brands that you had short listed earlier, please choose one brand as your final choice by clicking the appropriate button below. This instruction was identical to the instruction in the screening group. The rate & screen group followed a similar procedure. The rating task instructions were Before you begin choosing a microwave popcorn, first we want to understand your evaluation of these six brands. Please rate each of the six brands using the following scales. Please do not make up your mind about your final choice yet; simply rate the brands by clicking on the appropriate buttons below. After rating the brands on a nine-point scale with endpoints extremely unattractive and extremely attractive, the rate & screen group was asked to screen the alternatives using a information and then chose a brand from among their top three rated brands. Rating brand shares (πˆ A =.673, πˆ C =.231, πˆ E =.096) did not differ from either control group (all p >.25) suggesting there was no screening effect.