Special Session Summary Personalization and Decision Support Tools: Effects on Search and Consumer Decision Making

Kristin Diehl, University of South Carolina
[ to cite ]:
Kristin Diehl (2003) ,"Special Session Summary Personalization and Decision Support Tools: Effects on Search and Consumer Decision Making", in NA - Advances in Consumer Research Volume 30, eds. Punam Anand Keller and Dennis W. Rook, Valdosta, GA : Association for Consumer Research, Pages: 166-169.

Advances in Consumer Research Volume 30, 2003     Pages 166-169

SPECIAL SESSION SUMMARY

PERSONALIZATION AND DECISION SUPPORT TOOLS: EFFECTS ON SEARCH AND CONSUMER DECISION MAKING

Kristin Diehl, University of South Carolina

SESSION SUMMARY

With the emergence of the Internet, the number of products available to users at low search costs has increased dramatically. Without sophisticated search and decision support tools, however, "the contents of the Internet might be likened to the contents of the Library of Congress, without call numbers, and dumped out on the floor." (Burk 1999). In his discussion of this ACR session, Eric Johnson pointed out that contrary to computer processors, for which processing capabilities per square inch double every 18 months (Moore’s Law), the capacity of the human brain does not expand. In the face of large amounts of information, human processing capacity remains limited, creating the need for sophisticated decision support tools. All three papers presented in this session investigated the effects that different types of decision support tools can have on consumer decision making. The research presented by Diehl and the paper by Haubl and Dellaert investigated the effects of personalized screening and ordering tools, while the paper by Fasolo and McClelland looked at the effects of information display boards and tools supporting EBA decision rules.

The paper by Diehl demonstrates that, when options have previously been screened by an ordering tool, reducing search costs can lead to lower quality choices. Her research shows that consumers make worse choices because lower search costs cause them to consider inferior options. Similar to what Hauser (1978) has shown, considering inferior options in the first place has much stronger effects on the final decision than selectivity among these considered options. She also shows that trying to be more accurate can exacerbate this effect. When search costs are low, a strong accuracy goal can motivate consumers to consider a wider array of alternatives and further decrease choice quality.

Haubl and Dellaert examine how electronic recommendation agents affect the micro-level decisions consumers make at each stage of a sequential search process. Expanding on normative predictions about sequential search, the authors’ model features several behavioral extensions. In addition to rational cost-benefit considerations, their framework incorporates consumers’ use of perceptual cues as well as local comparison referents. Results from a large-scale experiment find support for the influence of these behavioral factors on the duration and extent of search as well as on decision quality.

Fasolo and McClelland demonstrate how negative interattribute correlations affect information acquisition, and consumer satisfaction for choices made from web-based comparison tables, opinion portals, and decisionBfacilitating sites. They show that when attributes are negatively related consumers search comparison tables more by product, than when attributes are positively related, but find choice more difficult and dissatisfactory. Negative correlations make choice more difficult also on opinion portals, where options are accompanied by summary star ratings, and on decision web sites that facilitate an attribute-wise elimination-by-aspect strategy. When attributes are negatively related, choice is easier and more satisfying when decision sites facilitate a simple compensatory process like MAU. Their research shows the importance of online decision tools robust to the frequent and unpleasant cases where attributes are negatively related.

The discussion lead by Eric Johnson focused on the implications of this research for market competition, consumer welfare and policy advice. Participants agreed that understanding the net benefits from different types of decision tools will be crucial to making informed policy recommendations.

 

TEMPTED BY THE CHEAP AND EASY: THE NEGATIVE EFFECTS OF LOW SEARCH COSTS IN ORDERED ENVIRONMENTS

Kristin Diehl, University of South Carolina

Alba et al. (1997), argue that the most important benefit of online shopping to consumers are electronic screening tools. Screening tools can order all available options according to their predicted fit with the consumer’s utility function. Although these tools only provide a good but not a perfect ordering, they allow consumers to find options that better fit consumers’ preferences with less effort (e.g. Haubl and Trifts 1999). Findings from the research presented in the following, however, suggest that lowering search costs in such ordered environments decreases the benefits of screening tools and can actually lead to lower quality decisions.

The benefits of lower search costs depend on the degree of ordering in the environment. If the ordering is good, additional search should yield little benefit in finding better fitting options. Rather, any additional search decreases the average quality of considered options, since lower ranked options are, in expectation, worse alternatives. Following Hauser (1978), I argue that consumers’ selectivity into the consideration set will be more important for the quality of their choices than selectivity among considered options and that considering inferior options will lead to worse choices.

This argument does not take into account that consumers can be selective when deciding between options. However, such electivity may be unlikely in the environment studied here. For example, more search may increase time pressure and limit cognitive resources consumers can devote to any given option, leading to more superficial, heuristic processing (e.g. Payne, Bettman and Johnson 1993). Furthermore, Martin, Seta and Crelia (1990) have shown that ignoring irrelevant information requires significant cognitive resources. If cognitive resources are limited due to increased search, ignoring irrelevant information will be difficult and poor options may not get screened out. Therefore consumers are expected to choose worse options with lower search costs because they consider inferior options and do not have the cognitive resources to be selective. These predictions are tested in two computerized laboratory experiments using principal-agent tasks.

In the first experiment, participants searched for and selected electronic birthday cards to send to a target consumer described on several dimensions (i.e. relationship to the sender, likes and dislikes, age, etc.). Independent judges rated all cards on their appropriateness for the recipient and showed high interjudge agreement (Chronbach alpha > 0.85). The average of these ratings was used to assess choice quality. Everybody saw an ordered list of 50 options. Cards were ordered using a regression model that provided a good but imperfect prediction of the judges’ ratings. Participants received a monetary reward for picking a card that fit the recipient well minus the search costs incurred. The first time a new card was inspected, participants were charged either 20 cents (high search cost condition) or 1 cent (low search cost).

As predicted, lower search costs decreased the quality of consumers’ decisions. Also, lower search costs caused consumers to consider worse options. A mediation analysis showed that the negative effect of lower search costs on decision quality was completely mediated by the quality of options considered.

These results show that more search initiated by lower search costs actually leads to worse decisions in an ordered environment. In principle, any factor that causes consumers to search too much from an ordered environment and to consider inferior options can have these effects. In experiment 2 that causal factor investigated is a consumer’s goal to be accurate.

Research on accuracy goals shows that such goals can stimulate more systematic processing and decrease consumers’ susceptibility to biases (Payne, Bettman and Johnson 1993, Tetlock 2002). This suggests that high accuracy goals could partially undo the negative effects of lower search costs by putting decision makers in a more careful mind set. However, depending on the decision environment, trying to be accurate paradoxically can lead to negative outcomes. Tetlock and Boettger (1989) show that under high accuracy motivation people tend to take a wider array of information into account. This behavior introduces non-diagnostic information into the decision process and leads worse decisions.

This prior research suggests that the effect of accuracy will strongly depend on the favorability of the decision environment. Under high search costs, consumers only consider a few good options. This constitutes a favorable environment where greater accuracy motivation may lead to better decisions. However, since options are already ordered, there is little room for improving decision quality. Under low search costs, however, consumers inspect more inferior options. This represents an unfavorable environment where high accuracy motivation may cause consumers to include a wider range of inferior options in their decision process, leading to lower quality decisions. Therefore a high accuracy motivation should worsen the negative effect of lower search on decision quality.

Experiment 2 tests these predictions using a principal-agent task in a different domain (MP3 players). Participants were rewarded based on the utility of the chosen option for the target consumer minus the search costs incurred. In addition, the five best performers received $10 (high accuracy condition) or $1.

As predicted, when search costs were low, greater accuracy goals led to significantly worse choices, whereas with high search costs greater accuracy goals marginally improved choices. A similar pattern of results emerges for the quality of the consideration set. When search costs were low, high accuracy goals led to marginally lower quality consideration sets. A mediation analysis indicates that the negative effect of search cost and accuracy on decision quality is mediated by consumers considering lower quality options.

These studies demonstrate the importance of consideration set formation in ordered decision environments. Any factor that decreases the quality of consideration sets dilutes the benefits of personalized ordering mechanisms. This also suggests that merely making personalized orderings available may not be enough for consumers to actually benefit from such tools. Marketers need to foster effective use of these tools for consumers to take advantage of this new technology. Factors that encourage too much search (e.g. lack of trust in the marketer) need to be taken into account carefully should consumers truly benefit from ordering tools.

BIBLIOGRAPHY

Alba, Joe, John G. Lynch Jr., Barton Weitz, Chris Janiszewski, Richard Lutz, Alan Sawyer and Stacy Wood (1997), "Interactive Home Shopping: Consumer, Retailer, and Manufacturer Incentives to Participate in Electronic Marketplaces," Journal of Marketing, 61 (July), 38-53.

Haubl, Gerald and Valerie Trifts (2000), "Consumer Decision Making in Online Shopping Environments: The Effects of Interactive Decision Aids," Marketing Science, 19 (Winter), 4-21.

Hauser, John R. (1978), "Testing the Accuracy, Usefulness and, Significance of Probabilistic Choice Models: An Information Theoretic Approach," Operations Research, 26 (3), 406-421.

Payne, John, James Bettman and Eric Johnson (1993). The Adaptive Decision Maker, New York, NY: Cambridge University.

Tetlock, Philip E. (2002), "Social Functionalist Frameworks for Judgment and Choice: Intuitive Politicians, Theologians and Prosecutors," Psychological Review, 109 (3), 451-471.

Tetlock, Philip E. and Richard Boettger (1989), "Accountability: A Social Magnifier of the Dilution Effect," Journal of Personality and Social Psychology, 57 (3), 388-398.

 

CONSUMER SEARCH AND DECISION MAKING IN PERSONALIZED INFORMATION ENVIRONMENTS

Gerald Haubl, University of Alberta

Benedict G. C. Dellaert, Maastricht University

Electronic information environments (e.g., web sites) are both an important source of product information for consumers and an increasingly common setting in which shoppers make purchase decisions. A defining property of such digital environments is that they allow a high level of personalization, i.e., the adaptation to individual consumer preferences. A common form of personalization is the implementation of electronic search or recommendation agents designed to assist consumers in making their purchase decisions (Diehl, Kornish and Lynch 2003; HSubl and Trifts 2000). While such technologies play an increasingly important role in aiding consumers’ information search and product selection, little is known about the manner in which they affect or transform human behavior.

The objective of this paper is to obtain a deeper understanding of how consumers search for product information and make purchase decisions in information environments that are personalized via electronic recommendation agents. The latter are software tools that generate personalized product recommendations in the form of a list in which alternatives are sorted by their predicted attractiveness to an individual shopper, thus allowing the latter to screen a large set of alternatives in a systematic and efficient manner (see, e.g., Haubl and Murray 2002). This personalization is based upon the agent’s understanding of the consumer’s multiattribute preference in connection with a given product category, which the agent has obtained through prior interaction with the consumer.

The present study examined how consumer information search and purchase decision making are influenced by (1) the presence vs. absence of personalized product recommendations, (2) the accuracy of the personalization technology in determining a consumer’s preferences, (3) the level of consumer search cost, (4) the extent of product differentiation in the market, and (5) the amount of price variability in the market. The effects of these factors are investigated both at the level of overall decision outcomes and with respect to the sequence of micro-level decisions that a consumer makes in the course of his/her pre-purchase search.

A number of specific research questions were addressed in this study. A subset of these questions is as follows: How does the presence or absence of personalized product recommendations affect the extent of consumer information search and decision quality? How can the product selection process in personalized information environments be explained by a combination of theoretical notions from standard search and choice theory? How do consumers choose the next alternative to be viewed from a personalized list? What types of comparisons among products do consumers make in sequential search? Which product, the most recently viewed or the most attractive one viewed so far, tends to serve as the primary comparison referent for the currently inspected alternative? When are product comparisons more based on subjective utility and when are they more attribute-based? What causes consumers to terminate their search process to complete their purchase in a personalized information environment? Do consumers tend to stop searching after encountering a particularly attractive or a particularly unattractive product? Under what types of market conditions are personalized product recommendations of greatest value to consumers?

A rich set of empirical findings pertaining directly to these questions was obtained in a large-scale web-based experiment. A total of 776 members of a consumer panel participated in this study. The primary experimental task was to engage in pre-purchase information search of a set of available products, and ultimately make a simulated purchase, in one of two product categories, compact stereo systems and vacation packages. The total number of available products in each category was 500. All stereo systems and vacation packages were described in terms of price and six quality attributes.

The task was one of sequential search with recall (Weitzman 1979) in a web-based shopping environment. Subjects sampled products one at a time, in whichever order they chose. Every time a new alternative was inspected, participants viewed all attributes of that product at once. Subjects were able to flag any product as their current favorite, which resulted in the capability to recall that alternative at any time, i.e., to purchase it instantly. Products were listed in the information environment either in random order (i.e., no personalization) or in descending order of expected subjective utility to the subject (i.e., personalized recommendation). In the latter case, this order either reflected the subject’s multiattribute preference very closely (high accuracy), or it was contaminated with substantial random error (low accuracy). Search cost was manipulated by either allowing direct click-throughs from the list of available products to an alternative’s detailed description (low cost) or by requiring subjects to type a short product ID number in order to request this description (high cost). The extent of product differentiation in the market was manipulated by varying the ranges of the attribute levels of the available products and, similarly, the amount of price variability in the market was manipulated by varying the range of prices. All of these experimental factors were manipulated independently between subjects.

The results of this study are presented in terms of effects on (1) overall decision outcomes and (2) micro-level outcomes. At the overall level, the results show that, consistent with recent findings by Haubl and Trifts (2000), personalized recommendations improved consumers’ decision quality significantly and reduced the number of products inspected. However, this form of personalization had no effect on the average deliberation time per alternative. Interestingly, the joint impact of product differentiation and price variability in the market was moderated by the presence or absence of personalized recommendations. High degrees of product differentiation and price variability led to higher decision quality when the shopping interface was personalized, but to lower decision quality in the absence of such personalization. At the micro level, we find that differences among products in terms of both their subjective utility and their attribute profiles are important factors in consumers’ decisions to flag an alternative as the current favorite and/or to purchase a given alternative. Overall, differences in subjective utility tend to be more important determinants of consumers’ micro-level decisions than attribute profile differences. Finally, while price considerations appear not to influence consumers’ decisions to flag an alternative as their current favorite, they do have a strong effect on whether or not a given product is selected for purchase.

BIBLIOGRAPHY

Diehl, Kristin, Laura J. Kornish and John G. Lynch Jr., "Smart Agents: When Lower Search Costs for Quality Information Increase Price Sensitivity", Journal of Consumer Research, forthcoming, June 2003.

Haubl, Gerald and Valerie Trifts (2000), "Consumer Decision Making in Online Shopping Environments: The Effects of Interactive Decision Aids," Marketing Science, 19 (Winter), 4-21.

Haubl, Gerald, and Kyle B. Murray (2002), "Preference Construction and Persistence in Digital Marketplaces: The Role of Electronic Recommendation Agents," Journal of Consumer Psychology, forthcoming.

Weitzman, Martin L. (1979), "Optimal Search for the Best Alternative," Econometrica, 47, 3 (May), 641-654.

 

INTERATTRIBUTE CORRELATIONS AFFECT INFORMATION PROCESSING AND DECISION MAKING ON ONLINE COMPARISON TABLES, OPINION PORTALS AND DECISION-FACILITATING WEBSITES

Barbara Fasolo, University of Colorado at Boulder

Gary McClelland, University of Colorado at Boulder

According to the Pew Internet and American Life Project Survey (Feb.-Mar. 2001), nearly 9 millions Americans go online on an average day to "research a product or service before buying it". An increasing amount of decision facilitating websites (e.g., www.activebuyersguide.com, or www.netscape.com/decisionguides) and opinion portals (e.g. www.epinions.com) have appeared on the World Wide Web to assist users formulate judgments and to make choices for a large number of consumer decisions. The goal of decision sites is to display the information necessary for users to make a good decision. To prevent the risk of information overload, decision sites guide users through an interactive process that allows consumers to reduce the number of options to a manageable size (winnowing), and to compare a few surviving options side-by-side (comparison). Our research investigates how people make decisions online both on side-by-side comparison tables and with winnowing features.

Most comparison tables are simple and static attributes-by-options matrices containing information about a handful of options, on dozens of attributes. We observe decision processes on web-based comparison tables by means of WebIDB (http://samiam.colorado.edu/~mcclella/webIDB/), an original WWW-based version of the traditional Mouselab paradigm developed in the 90’s by Johnson, Payne, and Bettman (e.g., Johnson, Payne, Schkade and Bettman, 1991; Payne, Bettman and Johnson, 1993). On WebIDBs we studied how people process information on web-based product comparison tables. When these tables display options that require trade-offs among conflicting attributes, we found that users find the decision difficult, attempt to cope with the trade-offs despite their difficulty, but are unable to deal with them in the fully optimal ways prescribed by optimal rules of decision making (Fasolo, McClelland and Lange, 2001). Moreover, with conflicting attributes, users are dissatisfied with the choice made.

These results raised the question as to whether interattribute correlations would also affect online decision making at the winnowing stage. The WWW makes available to users two kinds of winnowing decision support, varying in ease of use, transparency and optimality of the choice process. The first support is through websites that induce users to follow easy but sub-optimal information processing strategies (e.g., www.point.com, using the Elimination-by-Aspect strategy). This process is "non-compensatory" because it does not entail compensations among conflicting attributes. The second support is through websites that help users be consistent and thorough in their information search and integration, following a difficult but optimal information processing strategy (e.g., www.activebuyersguide.com, using the weighted additive strategy). This process is "compensatory," because it entails compensations among conflicting attributes, and at times even faces users with explicit trade-offs among hypothetical options. In our first test of the effect of correlation on online winnowing, we asked users to make different choices using both a compensatory (i.e., the complex and optimal) decision site, and a non-compensatory (i.e., the easy but potentially sub-optimal) decision site (Fasolo, McClelland and Lange, in press). We found users’ choice behavior and perceptions to depend on the kind of website used as well as on the structure of the attribute information (conflicting or not). When attributes were conflicting, a choice required more "clicks," was rated as more difficult to reach, and left users feeling less confident and less satisfied with the final outcome.

This talk describes three studies that provide further evidence for the effect of negative interattribute correlations on online consumer decision making. The first study tests whether the users’ tendency to become more compensatory (option based and systematic) in the presence of conflicting attributes is related to individual differences in judgment and decision making. Individual differences in search strategy are found to be robust and consistent, but only mildy related to a specific person correlate. Openness to experience is positively related to increased option-based search, and performance in the Wason Selection Task is related to more extensive search.

The second study tests whether interattribute correlation affects use of, and choices off, opinion portals. Opinion portals, like epinions.com, allow users to compare products along subjective ratings offered second-hand by other users, rather than along objective attributes. Trusting other users’ ratings is fast and efficient, but risky. The first obvious risk is that the users contributing the rating have different subjective preferences than the decision makers’. The most sophisticated websites try to overcome this problem relying on "collaborative filtering" techniques able to filter ratings from like-minded users. However, even if used by "like minded users," we find that negative interattribute correlations interfere with the efficiency of opinion portals. Products with negative interattribute correlations are given more variable overall ratings than products with positive interattribute correlations by users sharing the same preference profile. Yet, it is with negative correlations that users report a higher tendency to rely on ratings of experts or of like-minded customers, because less confident in their own judgments.

The third study tests the conditions under which it is advisable to assist decision making with a compensatory or non-compensatory aid. Factors considered are the structure of the attributes (conflicting or not), the choosers’ preferences over the attributes (equal or unequal weight on the attributes), and the aid’s ability to show or hide trade-offs from the decision makers’ sight. We show that negative interatttribute correlations make choice more difficult also on decision web sites that facilitate an attribute-wise elimination-by-aspect strategy. When attributes are negatively related, choice is easier and more satisfying when decision sites facilitate a simple compensatory process like MAU.

Overall, these studies show the importance of making available to online consumers decision tools that are robust to the frequent and unpleasant cases where attributes are negatively related.

BIBLIOGRAPHY

Fasolo, B., McClelland, G.H., & Lange, K. A (2002). Interattribute correlations influence whether decision processing strategies are attribute-based or option-based. To appear in Journal of Consumer Psychology.

Fasolo, B., McClelland, G.H. & Lange, K. A.(in press). The effect of site design and interattribute correlations on interactive web-based decisions. In C.P. Haugtvedt, K. Machleit, and R. Yalch (eds.). Online Consumer Psychology: Understanding and Influencing Behavior in the Virtual World, Lawrence Erlbaum Associates, Inc.

Johnson, Eric, John Payne, David Schkade and James Bettman (1991). Monitoring information processing and decisions: the Mouselab System, Unpublished Manuscript. Center for Decision Studies, Fuqua School of Business, Duke University.

Payne, John, James Bettman and Eric Johnson (1993). The Adaptive Decision Maker, New York, NY: Cambridge University

----------------------------------------