Anchoring (Cognitive Bias)

ANCHORING AS ACTIVATION

Several authors have suggested that the origin of anchoring lies in the influence of anchors on the first, retrieval stage (see Fig. 6.1). Jacowitz and Kahneman (1995), Chapman and Johnson (1994, 1999), Strack and Mussweiler (1997) and Mussweiler and Strack (1999, 2000) all suggest that the anchor acts as a suggestion, making information consistent with the anchor more available, either in memory through priming mechanisms or because of a biased external search. Because the anchor is considered as a candidate response that subjects entertain, at least as a transient belief, it influences the target value. This account is consistent with work by Gilbert (1990, 1991) and Gilbert, Tafarodi, and Malone (1993) showing that comprehension includes an initial belief in the assertion presented, followed only later by rejection of false information.

Strack and Mussweiler (1997) examined the idea that anchoring is a special case of semantic priming. They propose that information retrieved in order to compare the anchor to the target is consequently more available for use when estimating the target value. This selective accessibility account (Mussweiler & Strack, 1999, 2000) predicts that the primed information will influence the target judgment only if it is relevant. Strack and Mussweiler (1997) found this pattern. Anchors representing the width of the Brandenburg Gate had only a small influence on judgments of its height, although they had a large influence on judgments of its width. In a further experiment, Strack and Mussweiler (1997) found that extreme anchors resulted in shorter response times for the comparative judgment but longer response time for the absolute judgment (target estimation). This result indicates that comparisons to implausibly extreme anchors do not require relevant target information to be retrieved, yielding a faster comparative judgment. Because this target information has not been primed, however, the absolute judgment takes longer. The implication is that for less extreme anchors, the information primed during the comparative judgment is used in the absolute judgment.

Kahneman and Knetsch (1993) proposed that the target question acts as a memory probe that retrieves the anchor mentioned earlier. The anchor is then treated as a candidate response. This priming is in some ways shallow, however, because the anchor does not always influence the target judgment, even when it primes relevant information. For example, information primed in the course of a comparative judgment about whether one would pay $200 to clean up Canadian lakes is relevant to an absolute judgment about the amount an average Toronto resident would pay, but it should also be relevant to a judgment about the percentage of Toronto residents who would pay $100. Nevertheless, the anchor influences responses to the first question but not the second, presumably because the first shares more surface features with the anchor.

Jacowitz and Kahneman’s (1995) finding that anchoring occurs in response to the comparative judgment (and not just in target estimation) implies that the retrieval of target information primed by the anchor is biased, such that target features similar to the anchor are disproportionately retrieved. This biased retrieval explains why an unexpectedly large number of subjects judge that the target value is lower than the low anchor (or higher than the high anchor). In other words, the comparison between anchor and target results in the anchor appearing too similar to the target. Chapman and Johnson (1999) point to this biased retrieval or asymmetric priming of target features as the key process that produces anchoring. Specifically, they hypothesize that the presence of an anchor increases the availability of features that the anchor and target hold in common while reducing the availability of features of the target that differ from the anchor.

There is evidence from process measures that decision makers concentrate their attention on target features similar to the anchor. Schkade and Johnson (1989) report that subjects spent proportionally more time looking at payoffs in a pricing task (in which they posit a monetary anchor) than in a choice task. In addition, they spent more time looking at probabilities in a 100-point rating task (in which they posit a probability anchor) than in a pricing task. Although there was no explicit anchor used in these studies, Chapman and Johnson (1999) found similar results in a task that did use explicit anchors. Subjects compared apartments described on three attributes. When a provided anchor value was high, they spent more time looking at positive features of the apartment; when the anchor was low, they spent more time looking at negative features.

Mussweiler and Strack (1999, Experiment 4) found similar evidence. In their study, some subjects answered an anchoring question (e.g., about the length of the River Elbe) and were instructed to list the features of the target that came to mind. Subjects given a high anchor tended to list thoughts that implied a high target value, whereas subjects given a low anchor tended to list thoughts that implied a low target value. Mussweiler and Strack also found evidence that the presence of an anchor primes target features that are similar to the anchor. In this study (described in Mussweiler & Strack, 2000), subjects answered the anchoring question, “Is the annual mean temperature in Germany higher or lower than 5°C (or 20°C)?” and then participated in a lexical decision task. Subjects given the low anchor were faster at identifying words such as “cold” and “snow,” whereas subjects given the high anchor were faster at identifying words such as “hot” and “sun.” This result shows that the anchor primed consistent information in memory.

This activation account of anchoring suggests methods for reducing the anchoring bias. Chapman and Johnson (1999) asked subjects to make predictions (e.g., the likelihood of a Republican winning the presidential election) after considering a numeric anchor. Subjects who were prompted to think of a reason opposing the implications of the anchor value (e.g., for subjects with a low anchor, reasons why a Republican would win) showed less anchoring than a control group given no prompt. In contrast, subjects prompted to think of a similar reason (e.g., for subjects with a low anchor, reasons why a Republican would not win) showed no more anchoring than did the control group. Mussweiler, Strack, and Pfeiffer (2000) replicated this finding. These results suggest that subjects ordinarily retrieve target features or reasons that are similar to the anchor (as evidenced by no effect of the similar prompt) but not those different from the anchor (unless prompted).

Mussweiler and Strack (1999, Experiment 4) also reduced anchoring in a manner consistent with this activation account. Participants in one group were asked to report the target features that came to mind when answering the comparative question. Each participant in a second group was presented with a list of thoughts generated by one of the subjects in the first group. Members of a control group were asked to list target-irrelevant thoughts. The first (own-thoughts) group showed anchoring that was no different than the third (control) group; however, the second (others’-thoughts) group showed less anchoring. The authors proposed that thoughts generated by another subject are more likely to be viewed as suspect or biased than self-generated thoughts, and consequently subjects are more likely to correct for their influence by generating alternative thoughts. Thus, a subject who is presented with a thought suggesting a high target value may counter by retrieving information supporting a low target value. Such a correction would cause the information activated to be less selective and more balanced.

Manipulations of information activation could be used not only to debias anchoring but also to augment it. If anchoring occurs because the anchor facilitates retrieval of target features similar to the anchor, then anything that enhances this selective facilitation should strengthen the anchoring effect. If a large pool of target features exists, the anchor has more opportunity to enhance the similar features selectively. Chapman and Johnson (1999, Experiment 5) asked half their subjects to list actions they undertook that affected their health, a task hypothesized to increase the pool of health-related information. All subjects then estimated the number of Americans who would die of heart disease or cancer in the next 10 years after considering a numeric anchor. Anchoring was stronger among subjects who had first elaborated on the health topic by listing their health-related actions. Making all target features more accessible through elaboration expanded the pool of features, thereby enhancing the anchor’s ability to facilitate retrieval of similar features.

Thus, evidence from a number of studies points to anchors as a type of memory prompt or prime to activate target information similar to the anchor. Anchoring is reduced if subjects are explicitly prompted to consider the anchor’s irrelevance (Chapman & Johnson, 1999; Mussweiler et al., 2000) or if the information primed by the anchor is irrelevant to the target judgment (Chapman & Johnson, 1994; Kahneman & Knetsch, 1993; Strack & Mussweiler, 1997). Strack and Mussweiler (1997) point out that the primed information has an effect only if the decision maker is uncertain of the target’s value, a prediction consistent with data from Quattrone et al. (1981), Wilson et al. (1996), and Chapman and Johnson (unpublished). Kahneman and Knetsch (1993) describe the priming process as automatic, which may explain why anchoring is largely unaffected by incentives (Wilson et al., 1996). The accounts of anchoring described in this section, although highly similar, are not identical. Further research is needed to specify more precisely the activation processes that produce the anchoring effect. The evidence is mounting, however, that anchoring involves a constructive process of priming or memory retrieval that influences judgments of preference (Payne, Bettman, & Johnson, 1992) and belief.