Elaboration and numerical anchoring: Implications of attitude theories for consumer judgment and decision making

Researchers across many domains have examined the impact of externally presented numerical anchors on perceiver judgments. In the traditional paradigm, “anchored” judgments are typically explained as a result of elaborate thinking (i.e., confirmatory hypothesis testing that selectively activates anchor-consistent information in memory). Consistent with a long tradition in attitude change, we suggest that the same judgments can result from relatively thoughtful or non-thoughtful processes, with more thoughtful processes resulting in judgments that have more lasting impact. We review recent anchoring research consistent with this elaboration-based perspective and discuss implications for past anchoring results and theory in judgment and decision making. © 2009 Society for Consumer Psychology. Published by Elsevier Inc. All rights reserved. People are confronted with hundreds of decisions every day ranging from the mundane (e.g., which beverage will I select from the vending machine?) to the consequential (how much life insurance will I purchase?). What determines these choices? Two prominent domains of research in psychology have made this topic a specialty: (1) research on judgment and decision making (JDM) and (2) research on attitudes and persuasion (A&P). Although each area has much to say about why people make the particular choices that they do, the areas generally do not share specific theories or research paradigms. Indeed, scholars in each area often attend different conferences and typically publish in different journals. Although researchers in these two areas have worked mostly in isolation from each other, much would be gained from a closer relationship and sharing of methods and ideas (see also Gilovich & Griffin, 1998). To illustrate the potential benefits of convergence, we take a well known JDM phenomenon— numerical anchoring—and show how principles and findings from A&P (our own area of training and expertise) might generate research questions (and provide understanding related ⁎ Corresponding author. E-mail address: [email protected] (D.T. Wegener). 1057-7408/$ see front matter © 2009 Society for Consumer Psychology. Publish doi:10.1016/j.jcps.2009.12.003 to those questions) that have not been generated by existing JDM theories. Although our approach in this paper is to examine implications of attitude theory for anchoring and JDM, we are confident that use of JDM theories might similarly generate research questions and insight for A&P that have not been generated using existing A&P theories. A strong common interest across the JDM and A&P areas is in the many “irrationalities” that shape perceptions and judgments. People's perceptions are biased by the presence of pleasant or unpleasant environmental stimuli (Griffitt, 1970), likeability of the source of a persuasive message (Petty, Wegener, & White, 1998), the incidental or even subliminal presentation of words that activate mental concepts (Bargh & Pietrimonico, 1982; Srull & Wyer, 1980), and the presence of standards of comparison (Mussweiler, 2003; Sherif, Taub, & Hovland, 1958), and even the weather (Schwarz & Clore, 1983). Within the JDM area, however, perhaps no judgmental bias is more “prototypic” than the consistent and powerful bias created by numerical anchoring (Tversky & Kahneman, 1974). In typical studies of numerical anchoring, research participants are first asked to consider whether a target judgment is higher or lower than a high or low anchor value. After stating whether the true value of the target is higher or lower than the ed by Elsevier Inc. All rights reserved. 6 D.T. Wegener et al. / Journal of Consumer Psychology 20 (2010) 5–16 anchor value, participants provide their estimates of the true value for the target judgment (Jacowitz & Kahneman, 1995; Strack & Mussweiler, 1997). In some studies, anchor values are simply said to have been randomly generated (Mussweiler & Strack, 1999). In other studies, participants are exposed to numerical values that are clearly random or otherwise irrelevant to the correct target estimates (e.g., social security numbers or phone numbers of participants, Ariely, Loewenstein, & Prelec 2003; Russo & Schoemaker, 1989; a spin of a “wheel of fortune,” Tversky & Kahneman, 1974). Regardless of how anchor values are generated or presented, their effects can be stunning. Numerical anchors influence just about any type of judgment. After considering and rejecting high rather than low numbers as potential values for the judgment at hand, people think that nuclear war is more likely (Plous, 1989), that their own abilities are higher (Cervone & Peake, 1986), that consumer products or gambles are worth more (Ariely et al., 2003; Chapman & Johnson, 1999), and that defendants in civil cases are liable for larger damage awards (Chapman & Bornstein, 1996). As noted earlier, the primary purpose of the current paper is to apply principles and findings from the A&P area to further our understanding of numerical anchoring. Researchers in A&P focus on the construct of attitudes (i.e., people's global evaluations of objects and issues) including how attitudes develop and how they influence behavior (see Eagly & Chaiken, 1993; Petty & Wegener, 1998). An attitudinal perspective on numerical anchoring therefore speaks to what anchoring effects might be observed, what variables would serve as moderators of anchoring, and the extent to which anchored judgments will be consequential (the topic to which we devote the most attention in the current paper). As explained in more detail later, we will see that numerical anchors, like other variables studied in persuasion settings (e.g., the credibility of the person delivering a message) can serve in “multiple roles.” That is, anchor values sometimes serve as simple cues that influence judgments rather directly, but at other times they serve to bias more effortful thinking about a judgment (see Petty & Wegener, 1998, 1999). Thus, the attitudinal approach incorporates both relatively thoughtful and non-thoughtful processes to account for the effects of numerical anchors. These different processes and the different circumstances in which they occur are important because according to contemporary attitude theory, judgments produced by relatively thoughtful processes have more lasting impact (i.e., they are more consequential) than judgments produced by relatively non-thoughtful processes. A brief history of explanations Anchoring and adjustment The most widely cited explanation of numerical anchoring up through the mid-1990s was operation of an anchor-andadjust heuristic (Tversky & Kahneman, 1974). That is, people use the anchor value as an initial starting point for the judgment and then insufficiently adjust their assessment away from the anchor value toward an answer that appears more plausible (see Jacowitz & Kahneman, 1995; Quattrone, Lawrence, Warren, Souza-Silva, Finkel, & Andrus, 1984). According to Quattrone et al. (1984; discussed in Plous, 1993), this “anchoring-andadjustment” process is explained by people having a range of plausible answers for any given question. Anchors outside that range lead people to adjust their estimates until they reach the nearest boundary of the range of plausible values. People give this boundary value as their estimate of the requested value. The anchor-and-adjust perspective was taken to predict that implausibly extreme anchors lead to the largest possible anchoring effects compared to plausible anchors because one's boundary values are always the most extreme values of the plausibility range (Quattrone et al., 1984; Strack & Mussweiler, 1997). When anchors fall outside the range of plausible values, however, increases in extremity of the anchor should have no additional effect; people will adjust their estimates until they reach the boundary values, which remain constant regardless of how far an anchor lies outside that range. Although anchoring and adjustment formed the basis for initial conceptions of numerical anchoring, more recent work of the past 10–15 years suggests that adjustment per se (i.e., starting with an anchor value and moving toward the range of plausible values) is relatively infrequent in the traditional anchoring paradigm (e.g., Epley & Gilovich, 2001, 2005; see Chapman & Johnson, 2002). Instead, the anchoring literature has shifted to an account of anchoring that rests more heavily on the knowledge that becomes activated as people consider the anchor and formulate their answers. Selective accessibility/anchoring as activation In recent years, most research on numerical anchoring has been based on the assumption that anchoring results from the activation of anchor-consistent knowledge that occurs during judges' testing of hypotheses about potential target estimates (e.g., Chapman & Johnson, 1999, 2002; Mussweiler & Strack, 1999, 2001b). When a person considers a plausible anchor, he or she (implicitly) tests the hypothesis that the anchor is the correct answer to the judgment at hand. In doing this, the person looks for ways in which the real answer to the question is similar to the anchor value. As a result of this “confirmatory search” process (Chapman & Johnson, 1994; Klayman & Ha, 1987), aspects of the target that are similar to the anchor become accessible and disproportionately influence judgments of the target (Chapman & Johnson, 1999; Mussweiler & Strack, 1999; Strack & Mussweiler, 1997). Hence, the “selective accessibility” or “activation” of anchor-consistent knowledge leads to the anchoring effect. According to this view, however, anchors that are too extreme are dealt with somewhat differently. According to Mussweiler and Strack (1999), people encountering implausibly extreme anchors are thought to test a modified hypothesis. Instead of testing the hypothesis that the real answer is equal to the anchor, judges test the hypothesis that the boundary of the range of plausible values is the correct answer to the judgment. That is, “participants may process implausible anchors by first 7 D.T. Wegener et al. / Journal of Consumer Psychology 20 (2010) 5–16 adjusting to the boundary value of a distribution of plausible values and then testing the hypothesis that the [judgment value] is equal to this boundary value” (Mussweiler & Strack, 1999, pp. 158–159). As a result, although the proposed mechanisms are different when comparing the traditional anchoring and adjustment and more recent selective accessibility views, predictions regarding anchor extremity are the same. Implausibly extreme anchors should lead to larger anchoring effects than more plausible (moderate) anchors, but increases in anchor extremity beyond the range of plausible values should not increase the anchoring effect any further because a fixed boundary value will form the basis of the modified hypothesis test (Mussweiler & Strack, 2001a; Strack & Mussweiler, 1997; but see Chapman & Johnson, 1994; Kahneman, 1992). The process of confirmatory hypothesis testing has been characterized as relatively effortful. For example, Mussweiler and Strack (2001b) noted that “larger anchoring effects occur under conditions which promote the extensive generation of anchor-consistent target knowledge” (p. 238). They concluded that anchoring in the standard paradigm “appears to involve a relatively elaborate process of testing the hypothesis that the target quantity may be similar to the comparison standard” (p. 252). As noted earlier, our elaboration-based approach to anchoring incorporates both relatively thoughtful and nonthoughtful processes to account for the effects of standard numerical anchors. Thus, although selective activation of anchor-consistent knowledge might capture one important aspect of numerical anchoring, we believe that our “attitudinal” approach provides a more complete picture. At the very least, distinguishing between thoughtful and non-thoughtful processes allows us to make predictions that do not follow naturally from previous anchoring theories. Elaboration and anchoring: Multiple roles for numerical anchors Our elaboration-based view of anchoring is based directly on theories of attitude change in which persuasion variables can take on different roles at different levels of elaboration (see Petty & Wegener, 1998, 1999). In the elaboration likelihood model (ELM; Petty & Cacioppo, 1986), a given persuasion factor, such as a credible source or a positive mood state, can influence attitudes in relatively non-thoughtful (peripheral or heuristic) ways or in relatively thoughtful (central or systematic) ways. For example, when not thinking carefully about a persuasive message, a message recipient can develop a favorable attitude toward an advertised product because his or her current positive mood becomes associated with the product 1 Attitude theories contemporary with the original anchor-and-adjust view predicted curvilinear effects of message extremity (see Bochner & Insko, 1966; Brock, 1967; Sherif & Hovland, 1961). Wegener, Petty, Detweiler-Bedell, and Jarvis (2001) produced results consistent with this approach by showing smaller anchoring effects with extremely implausible anchors than with more moderate anchors. Similarly, Wegener, Blankenship, Detweiler-Bedell, and Petty (2009) found smaller anchoring effects with lowand high-extremity anchors than with anchors of moderate extremity. (Gorn, 1982; Stayman & Batra, 1991) or because of a simple mood heuristic (i.e., positive mood signals liking; Petty & Cacioppo, 1983; Schwarz & Clore, 1983). However, when thinking more carefully about the ad, the positive mood can bias the thoughts that come to mind and create the same favorability bias in judgments (Petty, Schumann, Richman, & Strathman, 1993; Wegener, Petty, & Klein, 1994; see Chaiken & Maheswaran, 1994, for similar findings with respect to source credibility). In evaluating relatively thoughtful and nonthoughtful effects of mood, assessing the resulting judgments alone could make it seem like the amount of elaboration (thinking) about the message makes no difference. However, by measuring the processes at work (e.g., by assessing cognitive responses, Petty et al., 1993) or by examining the different consequences linked to highversus low-elaboration processes (Petty, Haugtvedt, & Smith, 1995), it becomes clear that there are meaningful differences between highand low-elaboration effects of persuasion variables. Similarly, our elaboration-based approach to anchoring suggests that numerical anchors could take on multiple roles (Blankenship, Wegener, Petty, Detweiler-Bedell, & Macy, 2008; Wegener, Petty, Detweiler-Bedell, & Jarvis, 2001). In some cases, numerical anchoring may result from relatively thoughtful, high-elaboration processes, but in other cases, numerical anchoring may result from relatively non-thoughtful, low-elaboration processes. High-elaboration anchoring Persuasion theories have long characterized people as assessing the merits of an advocacy by comparing the advocacy with their existing knowledge and beliefs (see Hovland, Janis, & Kelley, 1953; McGuire, 1985). That is, people elaborate on a persuasive claim by comparing it with their existing knowledge and by using that knowledge to interpret related information in order to determine what a reasonable perception of the object might be (Petty & Cacioppo, 1986). As people assess the merits of a claim, their reactions can have many dimensions, including viewing the claim as relatively acceptable or unacceptable, correct or incorrect, unbiased or biased (e.g., Petty, Ostrom, & Brock, 1981; Sherif & Hovland, 1961). High levels of elaboration are most likely to occur when the person is both motivated and able to put cognitive effort into assessing the central merits of the claim and corresponding attributes of the issue or target object (Petty & Cacioppo, 1979, 1986). Elaboration can also occur when people generate a persuasive message themselves (Janis & King, 1954) or simply think about the attitude object (Tesser, 1978). Thus, it seems quite reasonable for elaboration to occur when thinking about a target after receiving a numerical anchor (cf., Mussweiler & Strack, 2001b). Some high-elaboration processes could include syllogistic or probabilistic reasoning (e.g., Petty & Wegener, 1998; Wegener & Carlston, 2005; cf., Kruglanski & Thompson, 1999). Thus, conceptions of elaboration in contemporary theories of attitude change seem very similar to the process of hypothesis testing that forms the core of current anchoring theories (e.g., Chapman & Johnson, 1994, 1999; Mussweiler & Strack, 1999). However, attitude theorists would be more likely 8 D.T. Wegener et al. / Journal of Consumer Psychology 20 (2010) 5–16 to refer to these effects as biased processing effects of anchors on judgments (see Petty & Wegener, 1999). That is, when people are motivated and able to engage in effortful thinking, a numerical anchor can bias thoughts about a target just as a positive mood or a credible source can bias the thoughts about a product or advocacy. Low-elaboration anchoring Although the notion of elaboration (in persuasion theories) and the notions of hypothesis testing and selective accessibility (in anchoring theories) share some similarities, an elaborationbased approach to anchoring differs from prominent anchoring theories in important ways. Chief among these differences is that our elaboration-based approach suggests that thinking is not always elaborative (cf., Mussweiler & Strack, 2001b). When people lack motivation or ability, changes in attitudes can result from the relatively effortless use of heuristics or other cues to determine judgments (Chaiken, Liberman, & Eagly, 1989; Petty & Cacioppo, 1986). For example, people might accept a claim, regardless of how the claim is supported, if the person making the claim is an expert or is attractive, if the message recipient is in a good mood, or if people simply count the number of arguments presented in a message rather than evaluate the merits of the arguments (Petty & Cacioppo, 1984; Petty, Cacioppo, & Goldman, 1981; Petty, Cacioppo, & Schumann, 1983; Petty et al., 1993; see Petty & Wegener, 1998, for discussion of these potential cue effects). Likewise, numerical anchors should be capable of serving as judgment cues when motivation or ability to elaborate is lacking. Low-effort anchoring could result from a number of possible non-thoughtful processes. Numerical anchors might prime the number (e.g., Jacowitz & Kahneman, 1995; Wilson et al., 1996; see also Wong & Kwong, 2000) or a general sense of the “magnitude” of the target judgment being relatively large or small (Oppenheimer, LeBoeuf, & Brewer, 2008). When the comparative (higher/lower) judgment and the estimate of the true value refer to the same target (as they do in the standard anchoring paradigm), participants lacking ability or motivation to elaboratively might be inclined to treat the anchor as a “hint” to a reasonable judgment (Schwarz, 1994) without recognizing and making use of the fact that the anchor was generated randomly and is irrelevant (as the anchor is typically described in the standard paradigm). We could also imagine that there might be relatively non-thoughtful versions of selective accessibility (in which people use few rather than many anchor-consistent reactions activated by cursory confirmatory hypothesis tests), though no such simplified versions of selective accessibility have been described in selective accessibility articles. The incorporation of relatively non-thoughtful processes by which anchoring can occur differentiates our elaboration-based approach from prominent anchoring theories that propose relatively elaborative versions of confirmatory search/selective accessibility as the sole process responsible for anchoring in the standard paradigm. Those same prominent anchoring theories suggest that alternative processes, such as serial adjustment or numeric priming, only influence judgments outside the standard anchoring paradigm (see Epley, 2004, p. 246; Mussweiler & Strack, 2001b, pp. 241, 252). Consequences of anchored estimates According to the ELM, attitudes and other judgments resulting from differing amounts of target-relevant elaboration should have different consequences (Petty & Cacioppo, 1986; Petty et al., 1995). When a variable such as mood brings about attitude change through a relatively thoughtful mechanism (e.g., biased processing), the new attitude should persist longer over time, resist future attempts at change to a greater extent, and provide stronger guides to future thoughts and behaviors than when the same variable produces the same extent of change, but through a less thoughtful mechanism (e.g., a mood heuristic; see Petty et al., 1995; Petty & Wegener, 1998). These differential consequences could occur for a number of reasons. Higher levels of elaboration could be associated with cognitive changes surrounding the attitude or judgment, such as increased accessibility (Fazio, 2001), integration with relevant knowledge (Petty et al., 1995), or increased structural consistency of judgment-related knowledge (e.g., Chaiken, Pomerantz, & Giner-Sorolla, 1995). Higher levels of elaboration could also create metacognitions, such as higher levels of confidence that the attitude or judgment is correct or that it reflects what the person really believes (e.g., Barden & Petty, 2008; Chaiken et al., 1989; Petrocelli, Tormala, & Rucker, 2007), making a person more motivated to defend or act upon the attitude.