Abstract With the ever-increasing number of options from which consumers can choose, many decisions are made in stages. Whether using decision tools to sort, screen, and eliminate options, or intuitively trying to reduce the complexity of a choice, consumers often reach a decision by making sequential, attribute-level choices. The current article explores how the order in which attribute-level choices are made in such multistage decisions affects how consumers mentally represent and categorize their chosen option. The authors find that attribute choices made in the initial stage play a dominant role in how the ultimately chosen option is mentally represented, while later attribute choices serve only to update and refine the representation of that option. Across 13 studies (six of which are reported in the supplemental online materials), the authors find that merely changing the order of attribute choices in multistage decision processes alters how consumers (1) describe the chosen option, (2) perceive its similarity to other available options, (3) categorize it, (4) intend to use it, and (5) replace it. Thus, while the extant decision-making literature has mainly explored how mental representations and categorization impact choice, the current article demonstrates the reverse: that the choice process itself can impact mental representations. multistage decisions, mental representations, categorization, decision tree, phased decisions, replacement choice ultistage decision processes are those in which the decision maker reaches the final choice by making a series of lower-level (often, attribute-level) choices. With the ever-growing number of options available to consumers, such multistage decision processes are increasingly prevalent. For example, many decision aids help simplify decisions by guiding the consumer through the decision attribute-by-attribute. Websites commonly allow consumers to build and customize their preferred option or sort and screen the available options in an attribute-by-attribute process (e.g., Nike.com, Bestbuy.com, and most car manufacturers’ websites). Likewise, numerous restaurant chains, such as Chipotle and Domino’s, allow consumers to create their meal ingredient-by-ingredient. Even without the explicit use of such decision aids, in some cases, the decision context is multifaceted and individuals naturally break down the decision into a series of lower-level, attribute-based decisions (Bettman and Park 1980; Billings and Marcus 1983; Olshavsky 1979; Tversky and Sattath 1979; Wright and Barbour 1977). Such piecemeal decision processes help consumers simplify their decisions and reduce the cognitive load associated with considering all possible alternatives. It is often the case that when consumers are making such multistage decisions, the specific order in which they make attribute choices is arbitrary and can be easily changed (Kahn, Moore, and Glazer 1987; Levav, Reinholtz, and Lin 2012). The current article explores how a change in the order of attribute choices in multistage decisions—despite not influencing what will ultimately be chosen—impacts how consumers mentally represent and categorize the chosen option. We argue and demonstrate that such shifts in mental representations affect (1) the way consumers describe the chosen option, (2) the degree to which consumers perceive the chosen option to be similar to other available options, (3) how consumers categorize the chosen option, (4) how consumers intend to use the chosen option, and (5) what consumers choose to replace the chosen option with, if necessary. Thus, we show that the same product may be mentally represented and categorized by consumers differently depending on the order in which attribute choices were made. This article offers several unique contributions. First, from a theoretical perspective, while existing literature has mainly focused on how and why mental representations impact choice, this article is the first (as far as we know) to explore instances in which the choice process itself impacts mental representations and categorization. Thus, this article examines a novel interplay between decision making and mental categorization: even when the same option is ultimately chosen, its meaning and the consumers’ subsequent behavior substantially change due to the unique path leading to this choice. Further, the literature concerning multistage decision processes has mainly focused on decision structures in which earlier choices in the sequence determine the availability of options later in the sequence (i.e., path-dependent choice processes). In contrast, because the current article focuses on how multistage processes impact mental representations (and not the choice), we study path-independent multistage decision processes (i.e., the availability of options is not contingent on earlier choices in the sequence), allowing us to isolate and study pure order effects. Thus, we also add to the literature concerning multistage decision processes by focusing on a relatively understudied type of hierarchical decision process. Second, from an applied perspective, this research demonstrates that firms and policy makers can impact how consumers mentally represent their offerings in the marketplace by simply changing the order of attribute choices in multistage decision processes (which are often employed in sorting and configuration tools). As we demonstrate, such shifts in mental representations impact several important aspects of consumer behavior. Third, from a methodological perspective, our experimental designs and findings suggest that researchers studying topics related to mental categorization and its impact on consumer behavior could use this simple change in attribute-choice order to manipulate how participants mentally represent and categorize target stimuli. Next, we develop our theoretical framework and hypotheses. We then report seven studies (six additional studies are reported in the supplemental online materials), including studies that employ incentive compatible designs with consequential outcomes. We conclude with discussions on the theoretical, methodological, and applied contributions of this research, as well as the limitations and future directions. MULTISTAGE DECISION PROCESSES AND MENTAL REPRESENTATIONS Multistage decisions, sometimes referred to as phased decisions (Wright and Barbour 1977) or decision trees (Tversky and Sattath 1979), are those in which the decision maker reaches the ultimate choice by making a series of lower-level (often, attribute-level) choices. Unlike single-stage choices, in which consumers select a single option from a given assortment, in multistage decisions consumers reach their ultimate choice in a sequential, piecemeal (attribute-level) process (e.g., choosing the product’s design first, then its color, and so on). As mentioned above, many decision aids guide the consumer through the decision in stages, often in an attribute-by-attribute process. In other instances, the decision context is naturally multifaceted, and innately involves making a series of lower-level decisions. For example, choosing a vacation typically entails selecting a destination, hotel, room type, flights, car rental, and so on. In such cases, the consumer reaches the ultimate choice by making a series of lower-level decisions. Whether the choice involves an exogenously provided decision aid (Diehl, Kornish, and Lynch 2003) or innately comprises of a series of lower-level decisions, in many cases the order of attribute choices is malleable and can be easily manipulated. For example, most websites allow consumers to screen or configure products based on different attributes, and the order of attribute choices can be easily altered. Other types of decisions that are made offline naturally offer flexibility in terms of attribute ordering. Consumers may establish dining plans, for instance, by first choosing a location (e.g., dine in vs. take out) and then a food type (e.g., pizza vs. Chinese), or vice versa. Likewise, it is equally reasonable to first choose a movie theater and then a specific movie to view, or vice versa (at least in major metropolitan areas). Given the prevalence of multistage decision processes and the ease with which businesses can manipulate the order by which consumers choose attributes, it is important to study and understand the effect of altering the attribute-choice order. Existing research on multistage decisions has focused on, among other things, how such hierarchical choices impact consumers’ preferences and choices. For example, several articles have demonstrated that choices that are made after a screening process substantially differ from choices made without an initial screening stage. Specifically, decision makers were found to deemphasize the importance of screening attributes in later stages of the process (Chakravarti, Janiszewski, and Ülkümen 2006) and less likely to revisit information they used for screening (Wright and Barbour 1977). Closely related to the current research question, Kahn et al. (1987) examined how extrinsic factors that influence the order of attribute consideration (a situation termed constrained choice) impact which option consumers end up choosing. It is important to note that because the focus of existing research has been on how multistage decisions impact the ultimate choice, existing research has studied path-dependent hierarchical processes. That is, screening options on one attribute determined the subset of available levels of subsequent attributes (i.e., attribute levels across the different stages are correlated). For example, a consumer that screens options based on quality would face specific price levels in later stages of the decision process. One of the key aspects that distinguishes the current article from prior work on multistage decisions is that we examine the impact of such decision processes even when the ultimate choice remains the same. That is, we argue and demonstrate that the order by which consumers consider and choose attribute levels has important implications even when the option that is chosen remains the same. To test pure order effects in such multistage decisions, and to distinguish it from other mechanisms, throughout the article we examine path-independent multistage processes—that is, decision processes in which the attribute levels available in later stages are not contingent on the choices made in earlier stages. To clearly emphasize the importance of the distinction between path-dependent and path-independent decision processes, and to demonstrate how the order of attribute choices in multistage decisions could impact consumers, consider the following simple example. Imagine a consumer choosing a coat based on two attributes: material (wool vs. leather) and design (stylish vs. casual). Further, assume that all four possible combinations of coats are available for purchase (i.e., wool-stylish, wool-casual, leather-stylish, and leather-casual). Because all four options are available for purchase, it is evident that which coat is eventually chosen should not be influenced by whether the consumer first chooses material and then design or first design and then material. The consumer in this case can freely choose any design regardless of the chosen material or freely choose any material regardless of the chosen design. Consequently, the consumer’s ultimate choice will typically not depend on the order in which attribute choices are made (assuming attributes do not interact to influence preferences—we limit our discussion and empirical examinations to contexts with no such interactions). The above case is a stylized example of path-independent decision structures, which are the focus of the current article. We speculate that previous research has not focused on such path-independent structures (where all combinations of attribute levels are available) because changes in attribute-choice order in such cases typically do not lead to shifts in choice shares. However, we argue that these path-independent structures are in fact important to study from both theoretical and applied perspectives. Even though the same coat will be selected regardless of which attribute choice is made first, we argue that, from the consumer’s perspective, this coat will be mentally represented differently based on the order of those choices. Specifically, if the consumer is first asked to select the material (and selects wool over leather) and is then asked to choose the design (and selects stylish over casual), this consumer will mentally represent the chosen coat as a wool coat that has a stylish design. That is, the coat will be primarily mentally represented and categorized on the basis of its material (wool) and differentiated from other wool coats on the basis of its design (stylish). However, the same coat will be mentally represented by the consumer very differently if the coat’s design was selected first and then its material. In this case the coat will be represented and categorized as a stylish coat made of wool. We argue that this seemingly meaningless difference is in fact meaningful and has important consequences for how consumers describe, use, and replace their chosen options. HOW MULTISTAGE DECISION PROCESSES INFLUENCE MENTAL REPRESENTATIONS Mental representations are generally defined as an “encoding of information, which an individual can construct, retain in memory, access, and use in various ways” (Smith 1998). A mental representation is an internal cognitive symbol that represents external reality and can be drawn upon to describe, evaluate, or make decisions with respect to certain stimuli (e.g., objects, experiences, people). The mental representation of stimuli typically occurs rapidly, spontaneously, unconsciously, and as information is encountered (Lakoff 1987; Sato, Schafer, and Bergen 2013). Accordingly, research has demonstrated that mental representations, and the process by which mental categories are formed, are influenced by context (Anderson and Ortony 1975; Barsalou 1982; Roth and Shoben 1983). The importance of mental representations to human perception and, consequently, numerous marketing-relevant variables, has been repeatedly demonstrated within the marketing literature. To wit, consumers’ mental representations have been shown to influence information search, inference, memory, and choice (Alba and Hutchison 1987; Cohen and Basu 1987; Henderson and Peterson 1992; Huber and McCann 1982; Loken and Ward 1990; Moreau, Markman, and Lehmann 2001; Parker and Lehmann 2014; Reinholtz, Bartels, and Parker 2015; Sujan and Deklava 1987; Weiss and Johar 2013; Yates, Jagacinski, and Faber 1978). Given the influence of mental representations on such a broad spectrum of consumption-related behaviors, it is unsurprising that researchers have also examined factors influencing mental representations such as language structure (Schmitt and Zhang 1998; Yorkston and De Mello 2005), self-construal (Jain, Desai, and Mao 2007), numerical list structure (Isaac and Schindler 2014), and goal salience (Barsalou 1983). However, while existing literature in judgment and decision making has mainly focused on how mental representations influence choice, the current article examines if and how the choice process itself impacts mental representations. As mentioned, context plays a role in the formation of mental representations. Directly related to the current research question, previous research suggests that the order in which information about a stimulus is encountered can impact how this stimulus is mentally represented, with earlier information playing a greater role (Markman 1987, 1989; Moreau, Markman, and Lehmann 2001). For example, Moreau et al. (2001) showed that participants relied more heavily on ads and labels that were encountered earlier when asked to categorize a new product they had never seen before. Thus, earlier information will likely dominate how a stimulus is mentally represented, with later information serving to refine the individual’s mental representation of the stimulus. This may be especially relevant in the context of multistage decision processes, where options are gradually “constructed” or “formed” in an attribute-by-attribute choice process. Our main hypothesis is that because multistage decisions are made in a sequential, piecemeal (attribute-level) process, and because mental representations are formed spontaneously as information is encountered, attribute choices that were made earlier will likely be the primary basis by which the soon-to-be-selected option will be mentally represented. Returning to the coat example, once consumers select the material (e.g., wool), they will use this information to attach meaning to, and thus mentally represent, the soon-to-be-selected coat. That is, even prior to finalizing the entire multistage decision process, consumers will already use the chosen material to mentally represent their impending choice as a wool coat. Then, after choosing the coat’s design (e.g., stylish), consumers will add this information to (i.e., refine) their mental representation of the wool coat, such that it is now a wool coat that has a stylish design. As illustrated by this example, the mental representation of the chosen option will therefore be sensitive to the order in which attribute choices are made. Reversing the attribute-choice order should lead consumers to represent the same coat as a stylish coat that is made of wool. This shift, we argue and demonstrate, impacts several important aspects of consumer behavior. In the first empirical section (studies 1a and 1b, and studies 1c–1e in the supplemental online materials [hereafter, SOM]), we directly test our main hypothesis—attribute-choice order in a multistage path-independent decision process alters how consumers mentally represent the chosen option—using four distinct and validated markers for shifts in mental representation. Specifically, we test how the attribute-choice order alters how consumers (1) describe the chosen option (studies 1a and 1c), (2) perceive the similarity of the chosen option to other available options (study 1b), (3) categorize their chosen option (study 1d), and (4) intend to use the chosen option (study 1e). Due to space constraints we report empirical evidence only for the first two markers for mental representation (i.e., description in study 1a and similarity in study 1b) in the main article. Evidence for the additional markers of shifts in mental representations (i.e., categorization and intended usage) are mentioned in the main text but fully reported in the SOM (studies 1c–1e). In the second empirical section (studies 2–5) we test and demonstrate that the order of attribute choices in the multistage decision process alters how consumers replace the chosen option, if necessary. Studies 3a and 3b extend our understanding of the underlying mechanism by exploring the roles of choice and agency in the documented effect. Consistent with the mental representation account, study 3a demonstrates that the focal order effect does not occur when consumers merely view the attribute information sequentially without choosing levels from those attributes (i.e., outside the context of an impending decision, and therefore without an impending choice to mentally represent). Consistent with our proposed account, and ruling out several rival accounts, study 3b demonstrates that the effect persists even without consumer agency (i.e., when consumers observe a third party choosing in a multistage decision process). Next, studies 4 and 5 further demonstrate the robustness of the effect and explore important boundary conditions. Study 4 demonstrates that the effect occurs spontaneously, even when the order of attributes is clearly determined randomly and when the attributes themselves offer no meaningful information about the alternatives, effectively ruling out inference-based accounts. Study 5 demonstrates that the effect is further attenuated when consumers already have an established mental representation of the stimulus class. Figure 1 summarizes the dependent measures employed in the studies and the predicted results (using the aforementioned stylized coat example). Six additional studies reported in the SOM further demonstrate the robustness of the effect across domains, rule out choice-based accounts, and address additional methodological concerns. FIGURE 1 View largeDownload slide HYPOTHESIZED IMPACT OF ATTRIBUTE-CHOICE ORDER ACROSS STUDIES * Studies 1c–1e are fully reported in the SOM. FIGURE 1 View largeDownload slide HYPOTHESIZED IMPACT OF ATTRIBUTE-CHOICE ORDER ACROSS STUDIES * Studies 1c–1e are fully reported in the SOM. STUDIES 1A AND 1B: THE IMPACT OF MULTISTAGE DECISIONS ON HOW CONSUMERS MENTALLY REPRESENT THEIR CHOICE Study 1a: Descriptions of the Chosen Option Previous research has repeatedly demonstrated the strong link between mental representations and language (Lakoff 1987; Sato et al. 2013; Schmitt and Zhang 1998; Yorkston and De Mello 2005). Indeed, much of the seminal work in psychology relating to mental representations and categorization has focused on text comprehension (Roth and Shoben 1983; Rosch et al. 1976). Further emphasizing the link between mental representations and language, in a cross-cultural study, Schmitt and Zhang (1998) showed that speakers of a specific language, with its unique linguistic labels that classify the world into categories, perceive and categorize external stimuli differently than speakers of a different language. Building on this strong link between representations and language, we expect that if the order of attribute choices in multistage decisions influences how decision makers mentally represent and categorize their chosen option, then attribute-choice order should influence the way they later describe it. Thus, we predicted that individuals would be more likely to use the attribute in the initial stage as the primary descriptor of the chosen option, and the attribute in the subsequent stage as a secondary descriptor. Method Eighty-four paid participants (58% male, median age 33), recruited from the Amazon Mechanical Turk online panel (hereafter, AMT), were asked to imagine that they were planning to purchase a sofa set. Participants were told that the sofa set with their preferred design was available in three colors (white, brown, or black) and three fabric materials (linen, cotton, or wool). Participants were then told that the order in which they would make the attribute choices would be randomly determined by the computer (we elaborate on this randomization procedure and its purpose in the discussion section of studies 1a and 1b). Participants were then randomly assigned to one of two conditions. In the material-first condition, participants first chose their preferred material and then their preferred color. Participants in the color-first condition made these attribute choices in the reverse order. After making their choice of color and material, or material and color, participants were asked to imagine that a friend called them and asked which sofa they decided to purchase. Participants were then asked to write, in an open-end format, how they would describe the sofa to their friend over the phone. These responses were coded by two research assistants, blind to the hypothesis and conditions. Results and Discussion The frequency of choosing any specific material-color combination did not differ significantly by the attribute-choice order manipulation (χ2 (8) = 5.2, p > .74). Thus, the following results cannot be attributed to shifts in choices across conditions. Note that in all subsequent studies reported in the article and SOM, the choice shares of the option that was initially chosen did not differ significantly across the attribute-choice order conditions. For brevity, we do not repeat this analysis in the main text for each of the studies but rather summarize these chi-square analyses as well as multinomial logistic regressions testing for order effects on choice shares in appendix A. Two research assistants coded participants’ responses. For each response, the coders were instructed to indicate which, if any, of the two attributes (color or material) was used as the main descriptor for the sofa. The level of agreement between the two coders was high (88.1%) and the coders resolved disagreements on the remaining 10 responses through discussion. Overall, seven responses were coded by both coders as not clearly indicating the use of one attribute over the other as the main descriptor. Analyzing the data with and without these seven responses does not substantially changes the pattern of results. As predicted, the majority of participants (58.5%) assigned to the color-first condition used color as the main descriptor of the sofa. In contrast, only 18.6% of participants assigned to the material-first condition used color as the main descriptor. A binary logistic regression confirmed that this difference between conditions was significant (β = 1.82, Wald-χ2 (1) = 13.0, p < .001). Supporting our hypothesis that attribute-choice order in multistage decisions influences the mental representation of the chosen options, the manner in which participants described their chosen options varied systematically with the order in which they had made their attribute choices. An additional study reported in the SOM (study 1c) replicated these results using a different response mode that rules out issues related to sentence structure. Study 1b: Similarity Judgments A fundamental aspect of how individuals mentally represent and categorize stimuli is how those cognitive processes influence similarity judgments. Stimuli are perceived as more similar when they are considered members of the same category (Goldstone, Lippa, and Shiffrin 2001; Harnad 1987; Levin and Beale 2000; Livingston, Andrews, and Harnad 1998). Accordingly, if multistage decisions influence how consumers mentally represent their chosen option, then this should be revealed in their perceptions of similarity. That is, if consumers use the first attribute in the sequence as the primary attribute when forming their representations, then options that share the same level of that attribute should be perceived as more similar (compared to options that share levels of attributes chosen later in the sequence). Method Ninety-seven paid AMT participants (70% male, median age 29) were asked to imagine that they had recently won a Starbucks workplace raffle and that they would be able to choose a free drink. The drink options varied on two attributes: type (coffee vs. green tea) and temperature (hot vs. iced). After learning about the attributes and levels, participants were randomly assigned to either first choose a drink type and then temperature (the type-first condition), or vice versa (the temperature-first condition). After choosing, participants indicated which of the two other drinks in the set that shared one attribute level with the originally chosen option was most similar to their chosen drink. Specifically, participants were shown an image of the drink they had chosen, and the images of two other drinks that were available but not chosen: either the drink of (1) the same type but different temperature, or (2) a different type but the same temperature. They were asked to choose the drink that was the most similar to their original choice. Results and Discussion As predicted, 60.4% of participants that first chose the drink’s type (type-first condition) indicated that the drink of the same type (but different temperature) was the most similar to their chosen drink. In contrast, only 36.7% of temperature-first participants made the same similarity judgment. A binary logistic regression confirmed that this difference between conditions was significant (β = .97, Wald-χ2 (1) = 5.34, p = .02). Further supporting shifts in mental representations, participants judged the chosen option to be more similar to another option that shared an attribute level they chose earlier (vs. later) in the decision process. A direct extension of the findings in study 1b is that attribute-order in multistage decisions should also impact how participants group (i.e., categorize) the options available in the original set. That is, if changing the order of attribute choices changes how consumers mentally represent the stimuli, then one would expect that attributes that appear earlier in the sequence will serve as the primary dimension on which consumers group the available options in the set. We report an additional study in the SOM (study 1d) that directly examines this proposition. As predicted, study 1d shows that the attribute level chosen earlier in the sequence is the main dimension on which the options are subsequently grouped. Further, study 1e (also reported in the SOM) demonstrates how shifts in mental representations, due to the order of attribute choices, impact how consumers intend to use their chosen option. Discussion: Studies 1a–1e Taken together, the results of studies 1a and 1b (and SOM studies 1c–1e) provide consistent support for our main hypothesis. Merely changing the attribute-choice order in multistage decision processes significantly changes how consumers mentally represent their chosen option. Consumers rely more heavily on the first attribute that they select as they gradually form their mental representations of the complete product they end up choosing. Thus, as hypothesized, even though the order of attribute choices did not change which option participants eventually chose, it did have a significant effect on how participants described, categorized, and intended to use their chosen option, as well as how similar they judged it to be relative to other options in the set. While these results provide direct evidence for a shift in mental representation, it is possible that consumers may directly infer the relative importance of the different attributes based on the exogenously determined order in which they appear in the sequence. Because consumers often attend first to attributes they consider—a priori—more important (Tversky 1972), they might also infer that attributes that appear earlier in the decision process are more important. Similar inferential processes or forms of Gricean norms (e.g., “This attribute must be more important. Otherwise, why was it presented to me first?”) could also be suggested as rival theoretical accounts. However, as mentioned in the method section of study 1a, we employed an overt randomization procedure to address this concern. Participants were informed prior to engaging in the multistage process that the order of attribute choices in the sequence would be determined randomly by the computer. To make this as salient as possible, we showed participants a screen with a loading image that lasted 2 seconds and a message stating that the computer was randomly selecting the first attribute that would appear. An attention check at the end of the study confirmed the vast majority of participants (95.2%) recognized that the order of attribute choices was determined randomly by the computer. Analyzing the results with or without the few respondents that failed to acknowledge this randomization procedure does not substantially change the pattern of results. We employed the exact same overt randomization procedure in several other studies that we report in the article and in the SOM (i.e., studies 1d, 3a, and 4). In addition, studies 3a and 4 further rule out this inferential account in additional ways that are discussed later. Finally, other than supporting the proposed shifts in mental representations, the dependent variables employed in studies 1a–1e also demonstrate important and relevant behavioral consequences. Specifically, shifting how consumers describe, intend to use, and categorize chosen options is of substantive importance with practical implications. The following studies examine another conceptually and substantively important dependent variable that also taps into consumers’ mental representations of their chosen options: replacement choices. STUDY 2: THE IMPACT OF MULTISTAGE DECISIONS ON REPLACEMENT CHOICES The studies reported thus far provide converging evidence that attribute-choice order in multistage decisions influences how consumers mentally represent their chosen options. However, one may call into question the behavioral implications of these findings by arguing that if consumers ultimately choose the same option, regardless of attribute-choice order, then the implications of these results may be of limited practical importance. Contrary to this perspective, we argue that the shift in how consumers describe and use their chosen option (studies 1a, 1c, and 1e), resulting from a subtle and easy-to-implement shift in attribute-choice order, is both conceptually and practically important. Further, we contend that although the chosen option is objectively identical (from the perspective of an outside observer), it is meaningfully different from the consumer’s perspective. In this and subsequent studies we demonstrate that shifts in how consumers mentally represent their chosen option can actually impact consumers’ preferences and choices when they make subsequent replacement decisions. Replacement decisions are important in their own right, particularly given their prevalence. In the marketplace, there are numerous instances in which consumers cannot have their first choice of a product or service (due to unexpected stock outs, legal restrictions, bidding processes, etc.). In such cases, the consumer is likely to choose a replacement option (Boland, Brucks, and Nielsen 2012). Assuming the initially chosen option was the consumer’s most preferred option, it is likely that the consumer will choose the most similar replacement option available in the set. Because we hypothesized and found that the order of attribute choices in multistage decisions influences the perceived similarity of the other remaining options (study 1b), we predicted that individuals would be more likely to choose replacement options that share the attribute level selected in the initial (as opposed to the subsequent) stage of the multistage decision. Returning to the coat example, the consumer that first chose a stylish design, then the wool material, would be more likely to choose the other stylish coat (i.e., stylish-leather) as a replacement coat. In contrast, the replacement choice would be the casual-wool coat if the consumer first chose the material of the coat and then its design (see figure 1). In order to empirically test this hypothesis and extend it to actual behavior, the current study involved consequential (nonhypothetical) replacement choices of bags that participants actually received. Method One hundred fifty-eight student participants (32% male, median age 20) completed this incentive compatible study as part of a larger battery of studies in a behavioral lab at a major northeastern university. At the end of the session, to ostensibly thank them for their participation, we gave each participant a university-branded drawstring bag of his or her choosing. Participants could choose the logo design (two distinct designs were available) and material color (white vs. gray) of the bag (see figure 2 for a picture of the actual bags used in this study). Participants were randomly assigned to one of two conditions. In the logo-first condition, participants first chose their preferred logo and then their preferred color. In the color-first condition, the sequence was reversed and participants first chose their preferred color and then their preferred logo. After making their choices, participants were told that their chosen bag was unexpectedly unavailable. They were then offered the opportunity to choose a replacement bag that either had the same logo but different color or had the same color but a different logo. FIGURE 2 View largeDownload slide BAGS USED IN THE MULTISTAGE PROCESS ACROSS CONDITIONS IN STUDY 2 FIGURE 2 View largeDownload slide BAGS USED IN THE MULTISTAGE PROCESS ACROSS CONDITIONS IN STUDY 2 Results and Discussion As hypothesized, 65.8% of participants assigned to the logo-first condition chose a replacement bag that had the same logo as their most preferred bag, but a different color. In contrast, only 40.5% of participants assigned to the color-first condition made this same replacement choice. A binary logistic regression confirmed that this difference between conditions was significant (β = 1.04, Wald-χ2 (1) = 9.94, p < .001). Consistent with the perceived similarity results reported in study 1b, study 2 demonstrates that consumers’ consequential replacement choices were significantly influenced by the order in which they made attribute choices in the multistage decision. Taken together, studies 1a, 1b, and 2 used three distinct dependent variables to demonstrate that merely changing the order of attribute choices in multistage decisions can significantly impact how consumers mentally represent and categorize their chosen options. The next study broadens the scope of the investigation by testing the roles that choice and agency may have on the proposed effect. STUDIES 3A AND 3B: THE ROLE OF CHOICE AND AGENCY A question that remains unanswered is whether there is something unique about the gradual selection process of attribute levels in the multistage decision process that affects mental representations, or whether the effect is mainly driven by simple differences in the salience of the attributes due to the order in which they appear. Stated differently, would merely presenting the information about the attributes in a sequential manner without an impending choice produce the same results? As alluded to earlier, a mental representation is an internal cognitive symbol that represents a certain stimulus (objects, concepts, people, etc.). Thus, within the psychological process involved in forming mental representations, there needs to be a specific focal stimulus to be mentally represented. In the absence of such a focal stimulus, mental representations do not form. In the context of multistage decision processes, this focal stimulus is the option that is being gradually selected (or “formed” as attribute levels are selected). That is, as consumers gradually advance through the selection process, they assign meaning to and begin to mentally represent their soon-to-be-chosen option using the attribute choices they make in this gradual process. However, we argue that when consumers merely view the attribute information in a sequential manner, without the need to choose, a focal stimulus does not exist and therefore mental representations will not form. Accordingly, outside the context of an impending choice—where no focal stimulus exists in consumers’ mind—shifts in attribute order will not impact consumers’ mental representations. More specifically, while the attribute location in the sequence may impact the degree of attribute salience (i.e., engender primacy or recency effects), because no focal object exists or is being formed in the consumer’s mind, attribute order should not influence the consumer’s mental representations. We directly test this hypothesis in study 3a. Although the proposed mental representation account suggests that a choice process must occur for attribute order to impact mental representations, this account does not require agency. That is, based on the mental representation account, even if we observe someone else choosing a product sequentially, attribute ordering will impact how we (as observers) mentally represent the option that someone else chose. That is, we would expect to see the same shift in behavior (e.g., shifts in replacement options) even when someone else made the sequential attribute decisions for the consumer. In sum, we hypothesize that a selection process is necessary to produce shifts in mental representations, but that agency (i.e., choosing for oneself) is not required. Studies 3a and 3b directly examine these hypotheses using replacement options as the dependent variables. In study 3a we test whether the mere sequential presentation of information is sufficient to produce shifts in mental representations or whether an actual selection process is required. In study 3b (and SOM study 7) we test whether the same shifts in mental representations would occur even when participants observe someone else making the sequential decision process for them (i.e., without agency). Study 3a: The Role of Choice Method Two hundred eighty-six paid AMT participants (58% male, median age 31) took part in this study, which used coffee mug sets as stimuli. Participants were randomly assigned to one of four between-subjects conditions in a 2 (choice vs. view) × 2 (attribute-choice order: material-first vs. color-first) design. In the choice condition, participants were asked to imagine that they were offered a choice of a free coffee mug set and that the sets came in two colors and two styles. Participants were informed that they would be required to choose their preferred color and style. Participants assigned to the view condition were given the same information about the coffee mug sets varying on color and style, but were not put in a context of an impending choice. Participants were told that they would be asked to view the different colors and styles but that they would not be asked to select their preferred levels. To ensure that participants in this condition paid attention to the information that would be displayed, they were asked to memorize the attribute information. Subsequent recall measures, as well as time spent on choosing or viewing, were collected across both conditions. Participants in both the choice and view conditions were then told that the computer would randomly determine the order by which they would either choose or view the attributes, respectively. This overt randomization procedure, also employed in studies 1a, 1d, and 4, was designed to eliminate inferences from the selection order. The second factor manipulated the order of attributes. Specifically, participants were assigned to either the color-first condition (first choosing/viewing the mugs’ color, and then style) or the style-first condition (first choosing/viewing the mugs’ style, and then color). The color attribute was presented with two color swatches, and the style attribute was presented with the mugs’ silhouettes. Unlike participants in the choice condition, those assigned to the view condition learned about the choice context (choosing a free coffee mug set) only after viewing these attributes in the same sequential manner. Only then were they presented with the entire choice set of four coffee mugs and asked to choose their preferred mug. On the next screen, participants in all four conditions viewed a picture of their chosen mug and were informed that it was unavailable. They were then told that they could select any of the three remaining mugs as a replacement (pictures of the three remaining mugs appeared below the selected mug). Accordingly, participants could choose a replacement mug set that shared the same color with their original mug choice, shared the same style, or was different on both attributes. Finally, two attention-check measures were used to ensure that participants paid attention to both the randomization of attribute choices procedure and the attribute information. As in studies 1a and 1e, the vast majority of participants across all conditions (95.5%) acknowledged that the computer randomly determined the order in which the attributes were selected from or viewed, and this did not significantly differ across the manipulated factors or their interaction (all ps > .55). Results and Discussion As mentioned, in order to ensure that participants in the view condition paid attention, they were explicitly asked to memorize the attribute information. Indeed, participants’ recall of the available colors and styles (by checking a list of four color swatches and four style silhouettes) was very high (97.9%) and did not differ across the manipulated factors or their interaction (all ps > .9). Excluding or including the six participants that failed these attention checks does not substantively change the results. Thus, participants in the choice and view conditions equally attended to the attribute information presented to them. Further, the average time participants spent choosing from (M = 5.35 seconds) versus viewing (M = 6.20 seconds) the attributes did not significantly differ (F(1, 284) = 2.32, p = .13). Hence, it is unlikely that any differences found between the choice and view conditions on the primary dependent measures can be attributed to participants’ paying less attention to information in the viewing condition. Participants’ replacement choices were analyzed via a logistic regression including both factors (attribute-order and choice vs. view) and their interaction as independent variables. The analysis revealed a significant interaction (β = 1.37, Wald-χ2 (1) = 5.81, p = .016) with no significant main effects. The choice shares in the choice condition replicated the proposed attribute-choice order effect. Specifically, participants assigned to the color-first condition were more likely to replace their initially chosen option with an option that shared the same color (39.1%) compared with participants in the style-first condition (14.7%; Wald-χ2 (2) = 10.55, p < .005). However, this effect did not hold for participants who merely viewed the attributes in a sequential manner outside the context of an impending choice. Specifically, in the view conditions, color-first-condition participants were as likely to replace their initially chosen option with an option that shared its color (23.6%) as were style-first-condition participants (24.7%: Wald-χ2 (2) = .29, p > .86; please refer to appendix B for a table summarizing the choice shares of replacement options across conditions in studies 2–5). Note that the choice proportions of replacement options sharing no attribute levels with the participants’ original choice were very low (3.8%) and did not significantly differ across conditions (both the main effects and interaction were nonsignificant; all ps > .33). Analyzing the data with or without these 11 participants does not substantively change the results. Study 3b: The Role of Agency Method One hundred forty-five paid AMT participants (67% male, median age 30) were asked to imagine that they had received a gift card from a friend for a pen from a well-known prestigious brand. The available pens varied on material (two levels) and finish color (five levels). Participants were randomly assigned to one of four between-subjects conditions in a 2 (agency: yes vs. no) × 2 (attribute-choice order: material-first vs. color-first) design. Participants in the agency conditions chose their preferred level of each attribute. Participants in the no-agency conditions were told that the friend who had given them the gift card had selected the attribute levels for them. Whether participants chose the initial pen themselves or had the choice made for them by a friend, it was always chosen via a multistage decision. That is, participants either chose their preferred level of each attribute in a sequential manner (in the agency condition) or saw each attribute in a sequential manner with their friend’s selected level marked on the screen (no-agency condition). Thus, the difference between the conditions was that participants themselves completed the multistage decision in the agency condition, while those in the no-agency condition merely observed the sequence of attribute choices that their friend had made for them. Participants in the material-first condition made (or observed their friend making) the material choice first and then the finish color choice. Participants in the color-first conditions made (or observed their friend making) these choices in the opposite order. The pen ostensibly chosen by the friend in the no-agency condition was randomly determined for each participant. Next, participants were told that the chosen pen was unavailable because it was out of stock, but that all other combinations of material and color were available. All participants then indicated whether they would keep the same material but change the finish color or, alternatively, change the material but keep the same finish color. Results Participants’ replacement choices were analyzed via a binary logistic regression including both factors (attribute-choice order and agency) and their interaction as independent variables. As predicted, attribute-choice order had a significant main effect on participants’ replacement choices (β = .93, Wald-χ2 (1) = 3.83, p = .05); participants assigned to the material-first condition were significantly more likely to choose the replacement pen with the same material (70.4%) than those assigned to the color-first condition (50.0%). Neither the main effect of agency nor its interaction with attribute order was statistically significant (β = .65, Wald-χ2 (1) = 1.93, p > .16; β = –.09, Wald-χ2 (1) = .16, p = .90, respectively). Thus, the influence of attribute-choice order on participants’ replacement choices persisted even when participants did not make the initial choice themselves. Discussion: Studies 3a and 3b Studies 3a and 3b provide two important insights into the influence of attribute-choice order on mental representations of chosen options—both of which are consistent with our theoretical framework. First, study 3a demonstrates the role of the attribute selection process over and above saliency effects. Merely viewing the available attributes sequentially, outside the context of choice, does not affect consumers’ mental representations of their subsequently chosen option. While viewing the attributes in a specific order might make an attribute more salient (e.g., via primacy or recency effects), it does not involve the gradual formation of the chosen option, and therefore does not trigger mental representations. One might challenge the null effect in the view condition by attributing it to lower involvement in the viewing (as opposed to choice) task. However, no differences in participants’ recall of attribute information, nor in the average time spent choosing from or viewing the attributes, were observed between the choice and view conditions. Hence, the absence of an attribute-choice order effect in the view condition is unlikely to have occurred due to participants’ lower involvement or attentiveness. Second, study 3b further demonstrated that although a selection process is a necessary condition to influence mental representations of the chosen option, agency is not. That is, consistent with our theoretical account, merely observing someone else sequentially choosing attribute levels influences mental representations of the chosen option based on attribute order. These results are notable, as attribute order causes significant shifts in replacement choices even for products that consumers did not originally pick themselves. These results are also inconsistent with various choice-based accounts such as internal consistency (Festinger 1957), self-perception (Bem 1967), post-choice reasoning and justification (Shafir, Simonson, and Tversky 1993), and choice closure (Wright and Barbour 1977). All these accounts fundamentally relate to how consumers’ initial (active) choice impacts their subsequent behavior through different motivations. However, the attribute-choice order effect persists even when the initial choice was not made by the consumers themselves. Finally, these findings are also inconsistent with the aforementioned rival account pertaining to inferences about attribute weights drawn from the order in which these attributes appear. Such inferences, to the extent that people form them, should occur regardless of whether people choose from or merely view the attributes sequentially. The absence of an attribute-choice order effect in the view condition in study 3a casts further doubt that inferences were the main driver for the observed effect. We further test the inferential account in the next study. STUDY 4: MENTAL REPRESENTATIONS VERSUS INFERENCES The main goal of study 4 was to further test whether inferential processes—as opposed to mental representations—drive the observed effect. In order to directly test this account, study 4 asked participants to choose which task, out of four available tasks, they wanted to complete. To make potential inferences about the attributes describing the tasks less plausible, we used meaningless labels (i.e., letters and colors) to describe the four possible tasks. Further, we also manipulated between subjects the ostensible procedure by which the order of attribute choices was determined: random versus nonrandom. By our account, although these procedures should make inferences about attribute weights immaterial, the order of attribute choices should still impact how individuals mentally represent the stimuli. That is, because mental representations are formed spontaneously and automatically (Lakoff 1987; Markman 1987, 1989; Sato et al. 2013), we expected the effect to persist even when it was made explicit that the attribute order was determined randomly and when the attribute levels did not convey any meaningful information. Finally, in order to further broaden the scope of these findings, study 4 used a preference elicitation mode more consistent with screening decisions that are commonly used by online decision (or search) aids. That is, participants observed all possible task combinations and were asked to screen these options in stages based on the available attributes. Method Two hundred one paid AMT participants (63% male, median age 30) were asked to choose the task they would ostensibly complete from four different tasks. The tasks were described on two meaningless dimensions—letter (H and M) and color (blue and red)—and a 2 × 2 table presenting all four combinations was shown to the participants in advance. Intentionally, no information about the content or characteristics of the tasks was provided, and thus participants could not make meaningful inferences about the importance of the attributes or their levels. Participants were randomly assigned to one of four conditions in a 2 (choice-order determination: random vs. nonrandom) × 2 (attribute-choice order: letter-first vs. color-first) between-subjects design. Participants assigned to the random conditions were given the same randomization instructions and procedure as was employed in studies 1a, 1d, and 3a. Participants assigned to the nonrandom conditions were simply told they would make their screening decisions in a prescribed order and then proceeded to the next screen. On the choice screen, the participants saw a 2 × 2 table containing all four combinations of letters and colors, which ostensibly represented the four tasks they might complete. Participants assigned to the letter-first condition screened out tasks by first choosing a letter and then a color, while those assigned to the color-first condition made these choices in the opposite order. When the participant chose a level for a specific attribute (e.g., H), the options containing the other level (M) faded to a darker color, signifying that they had been screened from consideration. After choosing a task, participants were informed that the chosen task had enough workers and was currently unavailable. Participants were then asked to choose a replacement task from the three remaining tasks. Results Participants’ replacement choices were analyzed via binary logistic regressions. Our primary dependent measure was the percentage of participants choosing a replacement task with the same color as but a different letter than their initially selected task. A binary logistic regression revealed a significant main effect of attribute-choice order (β = .56, Wald-χ2 (1) = 14.74, p < .001). Specifically, 64.7% of participants in the color-first condition chose the replacement task with the same color but different letter, compared to only 37.3% of participants in the letter-first condition. Importantly, making it explicit that the order of attributes was determined randomly had no influence on replacement choices (β = –.11, Wald-χ2 (1) = .54, p > .46), nor did this factor interact with the attribute-choice order manipulation (β = –.02, Wald-χ2 (1) = .02, p > .90). Note that a few participants did choose the replacement task that differed from their first chosen task on both attributes, but the percentage of such responses was very low (highest percentage across all conditions reached 9.1%). Moreover, the rate of these responses was not influenced by attribute-choice order manipulation (β = .20, Wald-χ2 (1) = .44, p > .50), the order-randomness manipulation (β = –.28, Wald-χ2 (1) = .81, p > .36), or the interaction of these two manipulations (β = .00, Wald-χ2 (1) = .00, NS). Thus, the pattern of results cannot be explained by the percentage of participants that opted to replace their chosen option with a completely dissimilar option. Discussion The results of study 4 replicated our key order effect and are clearly inconsistent with an inferential account. Even after we made it salient to participants that the order of attributes was randomly determined by a computer, replacement choices were still influenced by the order of attribute choices in the multistage decision process. Further, the effect persisted in a context in which the attributes themselves (letters and colors) could not meaningfully signal actual qualities of the available options. Study 5 explores an important boundary condition for these shifts in mental representations. STUDY 5: THE MODERATING ROLE OF ESTABLISHED MENTAL REPRESENTATIONS If the reported order effect of multistage decisions is indeed driven by shifts in how consumers mentally represent their chosen option, then this effect should attenuate when consumers already have established or “fixed” mental representations of the target stimulus class. The current study tests this boundary condition. Method Three hundred five paid AMT participants (57% male, median age 32) were recruited for this study, which asked them to choose a coffee mug via a multistage decision process. This study was composed of four conditions in a 2 (pre-established mug representation: fixed vs. control) × 2 (attribute-choice order: color-first vs. design-first) between-subjects design. To manipulate the pre-established mug representation, we had participants first complete a randomly assigned task before completing the multistage decision. In the fixed mental-representation condition, participants were given a picture of two coffee mugs that varied in color and design (neither the colors nor designs were used in the subsequent multistage decision) and were asked to choose their preferred mug. After indicating their choice, participants were asked to describe their chosen mug such that a person who could not see the option would be able to visualize it. The purpose of this condition was to make participants form their mental representations of the target class of stimuli (coffee mugs) on the two dimensions of interest: color and design. In so doing, participants would have to rely on the two attributes and should presumably establish the hierarchy of attributes in their mental representation of coffee mugs. Regardless of which attribute hierarchy was established by this manipulation, this task should mitigate the impact of the attribute-order effect in the subsequent multistage decision. In contrast, the control mental-representation condition asked participants to do the same description task after choosing between two options of chairs. Although this task would fix these participants’ mental representations of chairs, it should not influence their mental representation of coffee mugs (the target product used in the subsequent multistage decision). In the next stage, participants in both conditions were asked to imagine that they were looking to buy a new set of coffee mugs and that they would choose their preferred mug attribute-by-attribute. The available mugs varied on two attributes—design (two levels: contemporary or rustic) and color (two levels: yellow or blue)—yielding four possible combinations. The second between-subjects factor was the attribute-choice order. Participants in the color-first condition first chose the mug’s color and then its design. In the design-first condition, the order of these decisions was reversed. At each stage of the decision, participants viewed either pictures of the mug’s design (color was removed from these images, creating a grayscale silhouette) or a color swatch indicating the mug’s color. After choosing a mug, participants across all conditions were told that their chosen mug was out of stock, and were asked to choose a replacement mug. Note once again that the mugs’ colors and designs were distinct from those used in the first stage of the study. Results and Discussion As predicted, and replicating our key effect, the choice of replacement option significantly differed as a function of attribute-choice order in the control condition, where participants did not have pre-established representations of mugs. Specifically, 60.8% of participants in the color-first condition chose the replacement mug with the same color compared to only 43.2% of participants in the design-first condition (β = –.71, Wald-χ2 (1) = 4.53, p < .04). However, the effect was fully attenuated in the fixed condition, where participants had pre-established mental representations of mugs: 37.2% of participants in the color-first condition and 40.5% of participants in the design-first condition chose the replacement mug with the same color (β = .14, Wald-χ2 (1) = .18, p = .67). The interaction of the two factors was marginally significant (β = –.85, Wald-χ2 (1) = 3.30, p = .069). These results demonstrate that once the mental representation of a class of options is set (via visualization, in this study), the attribute-choice order in a multistage decision does not impact how the chosen option is construed. This provides further support to the proposed mental representation account. GENERAL DISCUSSION The current research examines how the order of attribute choices in multistage decisions impacts consumers’ mental representations and categorization processes as well as subsequent behavior. Using multiple established paradigms, we find that consumers mentally represent chosen options based on the attribute that is selected earlier in the multistage decision process. Accordingly, we find that the attribute-choice order changes how consumers (1) describe the chosen option, (2) judge its similarity relative to other available options, (3) categorize it, (4) intend to use it, and (5) replace it. Theoretical Contribution The literature concerning mental representations and decision making has primarily focused on how mental representations impact choice. In contrast, this article demonstrates that the choice process itself can impact how consumers mentally represent and, consequently, categorize their chosen option. Our findings further show a range of consequences of such shifts in mental representations. Related work by Chakravarti et al. (2006) also proposed a type of a categorization effect in multistage decision processes. However, their proposed categorization process is distinct from the one explored in the current article. In particular, Chakravarti et al. (2006) explored how information and attributes used for prescreening alternatives become less important relative to information acquired at later stages. These authors argued that after an option has passed the screening process, it is categorized as an acceptable option (i.e., as part of the final consideration set). In the current article, however, we demonstrate a cognitive process that does not relate to whether an option has passed a certain threshold, but rather relates to a shift in the mental representation of the actual option and its meaning. Further, because the extant literature on multistage decision processes has explored how such processes impact choice, it has largely focused on path-dependent processes (including the aforementioned Chakravarti et al. 2006). However, we focus on path-independent multistage processes, where all options are available regardless of choices made earlier in the process. We find that, although consumers choose the same option regardless of the order in which attribute choices are made, the chosen option is subjectively different from the consumer’s perspective. As demonstrated, such shifts in subjective perception can have important behavioral implications. Accordingly, the current article focuses on understudied, path-independent decision processes, which allows us to explore pure order effects and disentangle such effects from other mechanisms explored in the past. Relatedly, Boland et al. (2012) report an attribute carryover effect that seems inconsistent with some of the results we report. In particular, Boland et al. (2012) examine consumers’ choice of replacement options after consumers engaged in a screening process and learned that their top choice was unavailable. The authors report instances in which consumers, instead of choosing their previously stated runner-up option (which met the initial screening criteria), tend to choose an option that does not meet the initial screening criteria but has another desirable feature shared with their top choice. We speculate that the seemingly inconsistent results may arise from several distinct features in the paradigms and procedures employed by Boland et al. (2012) that may have made it hard for participants to recognize the hierarchical process of their attribute selection (to the extent there was one). This may have hindered participants from mentally representing their chosen option based on a certain attribute-selection process. We elaborate further in the SOM on the specific differences of the experimental paradigms. Applied and Methodological Contributions Beyond its conceptual contribution, this article also demonstrates that a relatively simple-to-implement change in the decision process could enable firms and policy makers to influence how individuals mentally represent their offerings in the market. Such shifts in mental representations are shown to impact several important behavioral dependent variables including preferences and subsequent decisions (e.g., replacement decisions). Further, given the importance of understanding consumers’ reactions to within-category versus cross-category substitutes (Huh, Vosgerau, and Morwedge 2016), shifting consumers’ perceptions of what constitutes within- or cross-category substitutes by a simple change in attribute-choice order may be beneficial. Finally, it seems reasonable and worthwhile to test whether companies can influence consumers’ perceived set of competitive options in the marketplace as well as impact their brand’s perceived positioning by such shifts in mental representations. Additionally, because mental representation is a construct often proposed and identified as a mechanism underlying different behavioral phenomena (Henderson and Peterson 1992; Lynch, Chakravarti, and Mitra 1991; Reinholtz et al. 2015; Weiss and Johar 2013), researchers often seek to manipulate how participants categorize and mentally represent stimuli. Given the results of the current article, researchers investigating such mechanisms may find that manipulating attribute ordering in multistage decisions could serve as a simple and subtle manipulation of mental representations. Alternative Explanations One potential rival account for the observed pattern of results is that consumers infer the importance of the attributes based on their location in the sequence. That is, consumers believe that important attributes typically appear early in the multistage decision process. In this article we empirically address this alternative explanation in multiple ways. Studies 1a, 1d, 3a, and 4 used an explicit and overt manipulation informing participants that the order of attribute choices in the sequence was randomly determined by a computer. Further, study 4 tested the effect in a context that rendered the attributes (and therefore their importance) meaningless. Lastly, in study 3a we find that merely viewing the attributes in a sequential manner does not change consumers’ mental representations although inferential processes about attribute importance should have still occurred. Thus, our data is inconsistent with a more deliberative inference-making account. That said, we fully acknowledge that one could, in principle, still argue that such inferential processes occur automatically and would therefore apply also when people are explicitly made aware that the attribute order is determined randomly. Although it is not clear why such automatic inferential processes would not apply in the view condition in study 3a, additional research could more thoroughly test different types of inferential processes. Study 3b is also inconsistent with other rival accounts, such as internal consistency (Festinger 1957), self-perception (Bem 1967), post-choice reasoning and justification (e.g., Shafir et al. 1993), and choice closure (Wright and Barbour 1977). All these alternative accounts fundamentally relate to how consumers’ initial (active) choices impact their subsequent behavior through different motivations. However, we find that the effect persists even when the initial choice was made randomly by a friend (study 3b) or by a computer (study 7 in the SOM). Thus, our findings cast doubt on these various choice-based accounts. Finally, throughout the studies we systematically demonstrate that the observed pattern of results is not driven merely by differences in the option that consumers initially chose. That is, we neither expected nor observed shifts in the initial choice shares when employing path-independent multistage processes with different attribute-choice orders. It is important to emphasize that what distinguishes the multistage decision process is the gradual, piecemeal “formation” of the object (either through consumers’ choice, or choices made by a third party). Merely providing consumers the complete set of available attributes and levels in advance (or even in a sequential manner outside the context of choice; see study 3a) does not influence their mental representations. Indeed, in several of the studies reported in this article and in the SOM (studies 1a, 1b, 1e, 2, 3b, 4, and 7), the complete set of attributes and levels were given to participants in advance, yet the actual attribute-choice order in the multistage process influenced the way by which chosen options were mentally represented. Boundary Conditions, Limitations, and Future Directions Given the process we propose, we do not expect the effect to occur in situations where consumers’ mental representations of the stimuli are already well established and relatively fixed. Accordingly, in study 5 we experimentally induced the formation of prior representations of the focal class of stimuli and found that the effect attenuates. Similarly, we expect that such shifts in mental representations are limited to situations in which the attributes involved in the hierarchical process are reasonably representative of the category. For example, we do not expect that consumers will represent their chosen car based on its upholstery when other, more basic attributes such as brand name or the car’s class appear in the sequence. However, we do suggest that once such “basic-level” attributes are controlled for—for example, in configuring the internal design of a BMW sedan—the order of more comparable attributes in the sequence (such as the upholstery’s color and pattern) could shift consumers’ mental representations of the car’s design. More broadly, we suspect that the degree to which the attributes in the sequence are perceived as basic-level descriptors of the category, as well as their level of abstractness, would impact the degree to which their order might actually impact consumers’ mental representations (see Rogers and Patterson 2007 on the basic-level effect). Furthermore, our investigation focused on short time horizons, and we do not have evidence for how long the shifts in mental representations will last. Still, in many situations, even local shifts in mental representations could have important, possibly long-term, consequences, especially in cases where consumers’ mental representations are still malleable (e.g., novel categories, products, and attributes). Examining the long-term impact of these effects is an important and potentially fruitful avenue for future research. We also limited our investigation to a relatively small number of dimensions. This provided us a simple and strong test of our predictions. In the marketplace, however, multistage decision processes may sometimes involve a greater number of both attributes and levels. Examining the focal effect in settings with a larger set of dimensions seems important and could inform companies and policy makers about the extent to which this effect may be employed. Additionally, whether the mere number of levels of a certain attribute (controlling for its position in the sequence) changes the degree to which it impacts mental representations seems interesting and important from both theoretical and applied perspectives. In sum, this article demonstrates that merely changing the order in which consumers select attribute levels in a multistage decision alters how they mentally represent the chosen option. Although the initial choice remains the same, this option is conceptualized very differently from the decision maker’s own perspective. We argue more broadly that this finding may highlight a class of instances that, in our view, is relatively underexplored in the judgment and decision-making literature. Specifically, the core focus of research in the field of decision making is to explore how and why certain actions and interventions may lead to shifts in choice shares. Such shifts in choice shares are essentially defined and measured by decision makers’ choice of an objectively different option from the set. However, the current article suggests that perhaps broadening the scope of what constitutes a “different” choice may benefit the field and lead to additional important research questions. After all, if the same option is construed differently and therefore described, used, and replaced differently, perhaps it meaningfully ceases to be the same option. DATA COLLECTION INFORMATION Study 1a: The first author supervised the collection of data using Amazon Mechanical Turk panel in spring 2017. The first author analyzed these data (two coders coded the data). Study 1b: The second author supervised the collection of data using Amazon Mechanical Turk panel in spring 2015. The first, second, and fourth authors analyzed these data. Study 1c (SOM): The first and second authors supervised the collection of data using Amazon Mechanical Turk panel in fall 2015. The first, second, and fourth authors analyzed these data. Study 1d (SOM): The first and second authors supervised the collection of data using Amazon Mechanical Turk panel in spring 2017. The first and second authors analyzed these data. Study 1e: The second author supervised the collection of data using Amazon Mechanical Turk panel in summer 2015. The first and second authors analyzed these data. Study 2: The fourth author supervised the collection of data by research assistants at the University of Pennsylvania in spring 2015. The first, second, and fourth authors analyzed these data. Study 3a: The first authors supervised the collection of data using Amazon Mechanical Turk panel in spring 2017. The first author analyzed these data. Study 3b: The first and second authors supervised the collection of data using Amazon Mechanical Turk panel in spring 2015. The first, second, and fourth authors analyzed these data. Study 4: The first and second authors supervised the collection of data using Amazon Mechanical Turk panel in spring 2015. The first, second, and fourth authors analyzed these data. Study 5: The first author supervised the collection of data using Amazon Mechanical Turk panel in fall 2016. The first author analyzed these data. Study 6 (SOM): The first and second authors supervised the collection of data using Amazon Mechanical Turk panel in fall 2014. The first, second, and fourth authors analyzed these data. Study 6b (SOM): The first and second authors supervised the collection of data using Amazon Mechanical Turk panel in fall 2014. The first, second, and fourth authors analyzed these data. Study 7: The first and second authors supervised the collection of data using Amazon Mechanical Turk panel in fall 2014. The first, second, and fourth authors analyzed these data. INITIAL CHOICE ANALYSIS ACROSS ALL STUDIES Likelihood ratio tests *p-values for change in shares of each specific option using multinomial logistic regression Chi-square df Sig Study 1a **(.92; 1; .71; .83; .92; 1) 5.20 8 0.74 Study 1b (.67; 1; .64) 0.71 3 0.87 Study 1c (SOM) (.08; .73; .84) 3.32 3 0.34 Study 1d (SOM) (.72; .44; .85) 1.44 3 0.69 Study 1e (SOM) (.22; .87; .27) 2.78 3 0.43 Study 2 (.57; .46; .69) 0.59 3 0.89 Study 3a (.88; .56; .81) 1.67 3 0.64 Study 3b **(.82; .36; .37; .87; .78; .16; .51; .68) 5.45 9 0.79 Study 4 (.66; .34; .73; .41; .18; .34; .75; .46; .92) 0.08 3 0.99 Study 5 (.92; .48; .43; .92; .19; .48; .38; .19; .60) 3.13 3 0.37 Study 6 (SOM) (.29; .83; .93; .52; .75; .19; .93; .22; .21; .43; .62; .61; .16; .94; .45; .22; .85; .66) 15.17 18 0.65 Study 6b (SOM) (.41; .58; .90; .73; .69; .56; .21; .82; .80) 3.74 9 0.92 Likelihood ratio tests *p-values for change in shares of each specific option using multinomial logistic regression Chi-square df Sig Study 1a **(.92; 1; .71; .83; .92; 1) 5.20 8 0.74 Study 1b (.67; 1; .64) 0.71 3 0.87 Study 1c (SOM) (.08; .73; .84) 3.32 3 0.34 Study 1d (SOM) (.72; .44; .85) 1.44 3 0.69 Study 1e (SOM) (.22; .87; .27) 2.78 3 0.43 Study 2 (.57; .46; .69) 0.59 3 0.89 Study 3a (.88; .56; .81) 1.67 3 0.64 Study 3b **(.82; .36; .37; .87; .78; .16; .51; .68) 5.45 9 0.79 Study 4 (.66; .34; .73; .41; .18; .34; .75; .46; .92) 0.08 3 0.99 Study 5 (.92; .48; .43; .92; .19; .48; .38; .19; .60) 3.13 3 0.37 Study 6 (SOM) (.29; .83; .93; .52; .75; .19; .93; .22; .21; .43; .62; .61; .16; .94; .45; .22; .85; .66) 15.17 18 0.65 Study 6b (SOM) (.41; .58; .90; .73; .69; .56; .21; .82; .80) 3.74 9 0.92 * Across all of these studies we also tested, using a multinomial logistic regression, whether the choice shares of each specific option were affected by the order manipulation (or its interaction with other manipulated constructs). Overall, across all studies, none of the 83 tests revealed a significant impact of the order manipulation on the alternatives’ choice shares. ** Note that two options in study 1a and one option in study 3b had 0% choice frequency. We excluded these three options from our multinomial logistic regressions, as these cannot be estimated due to the separation problem. The authors are grateful for the financial support of the Wharton Behavioral Lab and the Wharton Dean’s Research Fund. They also thank Robert Meyer and Christophe Van den Bulte for their helpful comments and suggestions received on previous versions of this article. Supplementary materials are included in the web appendix accompanying the online version of this article. APPENDIX A APPENDIX B CHOICE SHARES OF REPLACEMENT OPTIONS ACROSS CONDITIONS IN STUDIES 2–5 Study 2 (n = 158): Backpacks Conditions Logo first Color first % choosing replacement option with the same logo 65.8% 40.5% Study 3a (n = 286): Coffee mugs Choice conditions View conditions Color first Style first Color first Style first % choosing replacement option with the same color 39.1% 14.7% 23.6% 24.7% Study 3b (n = 145): Pens Agency conditions No-agency conditions Material first Color first Material first Color first % choosing replacement option with the same material 64.9% 42.1% 76.5% 58.3% Study 4 (n = 201): Tasks Nonrandom conditions Random conditions Color first Letter first Color first Letter first % choosing replacement option with the same color 62.5% 34.1% 66.7% 40.0% Study 5 (n = 205): Coffee mugs Control mental representations conditions Fixed mental representations conditions Color first Design first Color first Design first % choosing replacement option with the same color 60.8% 43.2% 37.2% 40.5% Study 2 (n = 158): Backpacks Conditions Logo first Color first % choosing replacement option with the same logo 65.8% 40.5% Study 3a (n = 286): Coffee mugs Choice conditions View conditions Color first Style first Color first Style first % choosing replacement option with the same color 39.1% 14.7% 23.6% 24.7% Study 3b (n = 145): Pens Agency conditions No-agency conditions Material first Color first Material first Color first % choosing replacement option with the same material 64.9% 42.1% 76.5% 58.3% Study 4 (n = 201): Tasks Nonrandom conditions Random conditions Color first Letter first Color first Letter first % choosing replacement option with the same color 62.5% 34.1% 66.7% 40.0% Study 5 (n = 205): Coffee mugs Control mental representations conditions Fixed mental representations conditions Color first Design first Color first Design first % choosing replacement option with the same color 60.8% 43.2% 37.2% 40.5% View Large REFERENCES Alba Joseph W., Hutchinson J. Wesley ( 1987), “Dimensions of Consumer Expertise,” Journal of Consumer Research , 13 4, 411– 54. Google Scholar CrossRef Search ADS Anderson Richard C., Ortony Andrew ( 1975), “On Putting Apples into Bottles—A Problem of Polysemy,” Cognitive Psychology , 7 2, 167– 80. Google Scholar CrossRef Search ADS Barsalou Lawrence W. ( 1982), “Context-Independent and Context-Dependent Information in Concepts,” Memory & Cognition , 1, 82– 93. Google Scholar CrossRef Search ADS Barsalou Lawrence W. ( 1983), “Ad Hoc Categories,” Memory & Cognition , 11 3, 211– 27. Google Scholar CrossRef Search ADS PubMed Bem Daryl J. ( 1967), “Self-Perception: An Alternative Interpretation of Cognitive Dissonance Phenomena,” Psychological Review , 74 3, 183. Google Scholar CrossRef Search ADS PubMed Bettman James R., Park C. Whan ( 1980), “Effects of Prior Knowledge and Experience and Phase of the Choice Process on Consumer Decision Processes: A Protocol Analysis,” Journal of Consumer Research , 7 3, 234– 48. Google Scholar CrossRef Search ADS Billings Robert S., Marcus Stephen A. ( 1983), “Measures of Compensatory and Noncompensatory Models of Decision Behavior: Process Tracing Versus Policy Capturing,” Organizational Behavior and Human Performance , 31 3, 331– 52. Google Scholar CrossRef Search ADS Boland Wendy Attaya, Brucks Merrie, Nielsen Jesper H. ( 2012), “The Attribute Carryover Effect: What the ‘Runner-up’ Option Tells Us about Consumer Choice Processes,” Journal of Consumer Research , 38 5, 872– 85. Google Scholar CrossRef Search ADS Chakravarti Amitav, Janiszewski Chris, Ülkümen Gülden ( 2006), “The Neglect of Prescreening Information,” Journal of Marketing Research , 43 4, 642– 53. Google Scholar CrossRef Search ADS Cohen Joel B., Basu Kunal ( 1987), “Alternative Models of Categorization: Toward a Contingent Processing Framework,” Journal of Consumer Research , 13 4, 455– 72. Google Scholar CrossRef Search ADS Diehl Kristin, Kornish Laura J., Lynch John G. ( 2003), “Smart Agents: When Lower Search Costs for Quality Information Increase Price Sensitivity,” Journal of Consumer Research , 30 1, 56– 71. Google Scholar CrossRef Search ADS Festinger Leon ( 1957), A Theory of Cognitive Dissonance , Stanford, CA: Stanford University Press. Goldstone Robert L., Lippa Yvonne, Shiffrin Richard M. ( 2001), “Altering Object Representations through Category Learning,” Cognition , 78 1, 27– 43. Google Scholar CrossRef Search ADS PubMed Harnad Stevan ( 1987), “Category Induction and Representation,” in Categorical Perception: The Groundwork of Cognition , ed. Harnad Steven, New York: Cambridge University Press. Henderson Pamela W., Peterson Robert A. ( 1992), “Mental Accounting and Categorization,” Organizational Behavior and Human Decision Processes , 51 1, 92– 117. Google Scholar CrossRef Search ADS Huber Joel, McCann John ( 1982), “The Impact of Inferential Beliefs on Product Evaluations.” Journal of Marketing Research , 19 3, 324– 33. Google Scholar CrossRef Search ADS Huh Young Eun, Vosgerau Joachim, Morewedge Carey K. ( 2016), “More Similar but Less Satisfying: Comparing Preferences for and the Efficacy of Within- and Cross-Category Substitutes for Food,” Psychological Science , 27 6, 894– 903. Google Scholar CrossRef Search ADS PubMed Isaac Mathew S., Schindler Robert M. ( 2014), “The Top-Ten Effect: Consumers’ Subjective Categorization of Ranked Lists,” Journal of Consumer Research , 40 6, 1181– 202. Google Scholar CrossRef Search ADS Jain Shailendra P., Desai Kalpesh K., Mao Huifang ( 2007), “The Influence of Chronic and Situational Self-Construal on Categorization,” Journal of Consumer Research , 34 1, 66– 76. Google Scholar CrossRef Search ADS Kahn Barbara, Moore William L., Glazer Rashi ( 1987), “Experiments in Constrained Choice,” Journal of Consumer Research , 14 1, 96– 113. Google Scholar CrossRef Search ADS Lakoff George ( 1987), Women, Fire and Dangerous Things: What Categories Tell Us about the Mind , Chicago: University of Chicago Press. Google Scholar CrossRef Search ADS Levav Jonathan, Reinholtz Nicholas, Lin Claire ( 2012), “The Effect of Ordering Decisions by Choice-Set Size on Consumer Search,” Journal of Consumer Research , 39 3, 585– 99. Google Scholar CrossRef Search ADS Levin Daniel T., Beale James M. ( 2000), “Categorical Perception Occurs in Newly Learned Faces, Other-Race Faces, and Inverted Faces,” Perception & Psychophysics , 62 2, 386– 401. Google Scholar CrossRef Search ADS PubMed Livingston Kenneth R., Andrews Janet K., Harnad Stevan ( 1998), “Categorical Perception Effects Induced by Category Learning,” Journal of Experimental Psychology: Learning, Memory, and Cognition , 24 3, 732– 53. Google Scholar CrossRef Search ADS Loken Barbara, Ward James ( 1990), “Alternative Approaches to Understanding the Determinants of Typicality,” Journal of Consumer Research , 17 2, 111– 26. Google Scholar CrossRef Search ADS Lynch John G., Chakravarti Dipankar, Mitra Anusree ( 1991), “Contrast Effects in Consumer Judgments: Changes in Mental Representations or in the Anchoring of Rating Scales?” Journal of Consumer Research , 18 3, 284– 97. Google Scholar CrossRef Search ADS Markman Ellen M. ( 1987), “How Children Constrain the Possible Meanings of Words,” in Concepts and Conceptual Development: Ecological and Intellectual Factors in Categorization , ed. Neisser Ulrich, Cambridge, MA: Cambridge University Press, 255– 87. Markman Ellen M. ( 1989), Categorization and Naming in Children: Problems of Induction , Cambridge, MA: MIT Press. Moreau C. Page, Markman Arthur B., Lehmann Donald R. ( 2001), “‘ What Is It?’ Categorization Flexibility and Consumers’ Responses to Really New Products,” Journal of Consumer Research , 27 4, 489– 98. Google Scholar CrossRef Search ADS Olshavsky Richard W. ( 1979), “Task Complexity and Contingent Processing in Decision Making: A Replication and Extension,” Organizational Behavior and Human Performance , 24 3, 300– 16. Google Scholar CrossRef Search ADS Parker Jeffrey R., Lehmann Donald R. ( 2014), “How and When Grouping Low-Calorie Options Reduces the Benefits of Providing Dish-Specific Calorie Information,” Journal of Consumer Research , 41 1, 213– 35. Google Scholar CrossRef Search ADS Reinholtz Nicholas, Bartels Daniel M., Parker Jeffrey R. ( 2015), “On the Mental Accounting of Restricted-Use Funds: How Gift Cards Change What People Purchase,” Journal of Consumer Research , 42 4, 596– 614. Rogers Timothy T., Patterson Karalyn ( 2007), “Object Categorization: Reversals and Explanations of the Basic-Level Advantage,” Journal of Experimental Psychology: General , 136 3, 451– 69. Google Scholar CrossRef Search ADS PubMed Rosch Eleanor, Mervis Carolyn, Gray Wayne, Johnson David, Boyes-Braem Penny ( 1976), “Basic Objects in Natural Categories,” Cognitive Psychology , 8 3, 382– 439. Google Scholar CrossRef Search ADS Roth Emilie M., Shoben Edward J. ( 1983), “The Effect of Context on the Structure of Categories,” Cognitive Psychology , 15 3, 346– 78. Google Scholar CrossRef Search ADS Sato Manami, Schafer Amy J., Bergen Benjamin K. ( 2013), “One Word at a Time: Mental Representations of Object Shape Change Incrementally During Sentence Processing,” Language and Cognition , 5 4, 345– 73. Google Scholar CrossRef Search ADS Schmitt Bernd, Zhang Shi ( 1998), “Language Structure and Categorization: A Study of Classifiers in Consumer Cognition, Judgment and Choice,” Journal of Consumer Research , 25 2, 108– 22. Google Scholar CrossRef Search ADS Shafir Eldar, Simonson Itamar, Tversky Amos ( 1993), “Reason-based Choice,” Cognition , 49 ( 1–2), 11– 36. Google Scholar CrossRef Search ADS PubMed Smith Eliot R. ( 1998), “Mental Representation and Memory,” in Handbook of Social Psychology , 4th ed., ed. Gilbert Daniel T., Fiske Susan T., Lindzey Gardner, New York: McGraw-Hill. Sujan Mita, Dekleva Christine ( 1987), “Product Categorization and Inference Making: Some Implications for Comparative Advertising,” Journal of Consumer Research , 14 3, 372– 78. Google Scholar CrossRef Search ADS Tversky Amos ( 1972), “Elimination by Aspects: A Theory of Choice,” Psychological Review , 79 4, 281– 99. Google Scholar CrossRef Search ADS Tversky Amos, Sattath Shmuel ( 1979), “Preference Trees,” Psychological Review , 86 6, 542– 73. Google Scholar CrossRef Search ADS Weiss Liad, Johar Gita V. ( 2013), “Egocentric Categorization and Product Judgment: Seeing Your Traits in What You Own (and Their Opposite in What You Don’t),” Journal of Consumer Research , 40 1, 185– 201. Google Scholar CrossRef Search ADS Wright Peter, Barbour Fredrick ( 1977), “Phased Decision Strategies: Sequels to an Initial Screening,” in Multiple Criteria Decision Making , ed. Starr Martin K., Zeleny Milan, Amsterdam: North Holland, 1977. Yates J. Frank, Jagacinski Carolyn M., Faber Mark D. ( 1978), “Evaluation of Partially Described Multiattribute Options,” Organizational Behavior and Human Performance , 21 2, 240– 51. Google Scholar CrossRef Search ADS Yorkston Eric, De Mello Gustavo E. ( 2005), “Linguistic Gender Marking and Categorization,” Journal of Consumer Research , 32 2, 224– 34. Google Scholar CrossRef Search ADS © The Author 2017. Published by Oxford University Press on behalf of Journal of Consumer Research, Inc. All rights reserved. For permissions, please e-mail: email@example.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)
Journal of Consumer Research – Oxford University Press
Published: Apr 1, 2018
It’s your single place to instantly
discover and read the research
that matters to you.
Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.
All for just $49/month
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.
Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.
All the latest content is available, no embargo periods.
“Hi guys, I cannot tell you how much I love this resource. Incredible. I really believe you've hit the nail on the head with this site in regards to solving the research-purchase issue.”Daniel C.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud
“I must say, @deepdyve is a fabulous solution to the independent researcher's problem of #access to #information.”@deepthiw
“My last article couldn't be possible without the platform @deepdyve that makes journal papers cheaper.”@JoseServera