Anchoring as Adjustment

ANCHORING AS ADJUSTMENT

Anchoring effects have most often been explained in conjunction with the idea of insufficient adjustment away from the anchor. The name anchoring and adjustment implies a particular cognitive process whereby decision makers first focus on the anchor and then make a series of dynamic adjustments toward their final estimate. Because these adjustments are insufficient, the final answer is biased toward the anchor. Tversky and Kahneman (1974) described this type of adjustment process as occurring when subjects answered the United Nations question. This type of account raises a central question: Why are adjustments insufficient (Lopes, 1982)? Most accounts focus either on uncertainty about the true value or a lack of cognitive effort.

The first class of explanations uses the idea that judges are uncertain about the value they want to report. For example, Quattrone et al. (1981) proposed that subjects adjust the anchor until shortly after it enters a range of plausible values for the target item. Thus, when adjusting from a high anchor, decision makers stop at the high end of plausible values, but stop at the low end when adjusting from low anchors. Their study showed a larger anchoring effect for questions judged to have a broad range of plausible values.

A more formal model with a similar underlying idea was developed by Busemeyer and Goldstein (1992; see also Busemeyer & Townsend, 1993) as an anchoring and adjustment account of preference reversals. This dynamic model of matching is not a proposed mechanism of anchoring per se, but rather an anchoring account of preference reversals. This model is similar to that put forward by Quattrone et al. (1981) in that it posits a process in which decision makers test for a match between the anchor and the target and make adjustments if a match is not achieved. Premature matches lead to insufficient adjustment.

A second class of explanations for insufficient adjustment follows from the notion that adjustment is effortful. Consequently, lack of effort or lack of cognitive resources cause adjustment to be terminated too soon, resulting in a final response that is too close to the anchor. One line of evidence supporting such an adjustment process comes from studies that use cognitive load or busyness manipulations. These paradigms examine situations in which judges place too much weight on one type of information, much like an anchor in our description (Gilbert, Miller, & Ross, 1998). These include the overreliance on information about behavior, as opposed to situational constraints (i.e. the correspondence bias; Gilbert, Pelham, & Krull, 1988), and overreliance on one’s own knowledge, as opposed to information available to others (i.e. the “curse of knowledge”; Keysar & Barr, Chapter 8, this volume). These phenomena can be understood as instances of anchoring on one piece of information and consequently underweighting subsequent knowledge by a process of insufficient adjustment.

For example, Kruger (1999) proposed an anchoring account of the “above average” effect – the result that, when asked to compare themselves with their peers, people tend to judge themselves as above average. Kruger argues that people anchor on their own abilities and then make adjustments to account for their peers’ abilities. Whereas focusing on an anchor is an automatic process, adjustment is more effortful, and consequently tends to be insufficient. Kruger found that increasing cognitive load (by asking subjects to rehearse a string of digits) increased the above-average bias, suggesting that limiting cognitive resources makes adjustment more insufficient.

Other studies (e.g., Gilbert et al., 1988) provide similar evidence that increases in cognitive load affect the second stage (adjustment) of the process, and not the first stage (anchoring), suggesting that the second stage is more effortful and less automatic than the first. These studies do not use the traditional anchoring procedure (Tversky & Kahneman, 1974), so their relationship to other anchoring demonstrations is unclear. In addition, Gilbert et al. (1988, p. 738) point out, “The resource metaphor is only one way to describe such effects, and unfortunately, no critical experiment seems capable of distinguishing between resource and other viable interpretations (e.g., structure or skill).” We agree, and take these findings to be evidence that the first stage of the process is different in some ways from a second integration or expression stage, and that a difference in automaticity appears to be a likely candidate. It is less clear, however, whether this evidence requires an adjustment account of anchoring effects.

In evaluating whether adjustment is insufficient because of lack of effort or resources, another line of relevant research concerns the effect of incentives. If insufficient adjustment were the result of lack of effort or allocation of cognitive resources, we might expect individuals to exhibit less anchoring (that is, more sufficient adjustment) when the stakes are high. Busemeyer and Goldstein’s (1992) adjustment account makes just this prediction. Evidence about the effect of incentives is mixed but mostly negative. More broadly, several judgment phenomena that are attributed to anchoring, most notably preference reversals, do not diminish in the face of incentives (see Slovic and Lichtenstein, 1983, for a review). If anchoring were due to lack of cognitive resources, we might also find that experts show less anchoring because they make greater use of specialized task heuristics that conserve cognitive effort. However, this does not appear to be the case (Northcraft & Neale, 1987). Finally, we might expect that warning people about anchoring would diminish anchoring biases. However, Wilson et al. (1996) as well as Quattrone et al. (1981) found that warning participants to avoid the anchoring effect was unsuccessful.

In addition to asking why adjustment is insufficient, one might also look for more direct evidence as to whether the anchoring effect does, in fact, result from a cognitive process that involves adjustment. Several studies have tested directly for adjustment processes. Following a suggestion by Lopes (1982), one type of study uses process-tracing measures to look for evidence of adjustment. A preference reversal study conducted by Schkade and Johnson (1989) asked subjects to price lotteries or to rate them on a 100-point scale. They hypothesized that the lottery attribute (dollars or probability) most similar to the response scale served as an anchor, causing the evaluation to be overly influenced by that attribute. They used process measures by, for example, asking subjects to indicate a response by pointing or dragging a cursor along a response scale while the computer recorded the amount of movement. The point at which the subject first touched the cursor to the scale could be an indication of an anchor. These starting points on pricing scales were correlated with the amount to win in the lottery; starting points for rating judgments were correlated with probability information. However, although these starting points may indicate the beginning of an adjustment process, they might also be just an indication of subjects’ final answers and an anchoring effect, rather than a specific anchoring process. This latter interpretation is bolstered by the fact that subjects’ adjustments along the response scale did not correspond to an adjustment process. For example, larger adjustments should be associated with a smaller anchoring effect because anchoring results from insufficient adjustment, according to this view. However, Schkade and Johnson’s data did not show this pattern. Chapman and Johnson (unpublished data) conducted a similar experiment in which subjects considered monetary anchors before evaluating lotteries. Using the same process measures, they likewise did not find an association between adjustments along the response scale and size of the anchoring effect.

A second type of study testing whether anchoring involves adjustment uses tasks where no adjustment could occur. For example, Jacowitz and Kahneman (1995) first presented trivia questions (e.g., the height of Mount Everest) to a calibration group. The 15th and 85th percentiles of the estimates from this group were used as anchors for a second group of subjects. These subjects were presented with a high or low anchor and asked whether it was higher or lower than the target value (e.g., the height of Mount Everest); they then provided their own target estimate. Subjects’ estimates showed an anchoring effect, as expected. Of greater interest, the comparison judgments themselves showed an anchoring effect. Although 15% of the calibration subjects had given target estimates lower than the low anchor (or higher than the high anchor), a much larger percentage of the experimental subjects said that the target value was lower (or higher) than the anchor. That is, the comparison question influenced judgments of the target value even before subjects were actually asked to estimate the target. Green, Jacowitz, Kahneman, and McFadden (1998) found a similar result. It may be reasonable to posit that adjustment is involved in estimating the target, but not in judging whether the anchor is higher or lower than the target. Thus, anchoring occurred prior to any possible adjustment, indicating that an adjustment process is certainly not necessary for an anchoring effect.

In sum, although anchoring and adjustment have been close traveling companions since the 1970s, there is only limited evidence that the origins of anchoring lie in an insufficient adjustment process. Although research on increases in cognitive load suggests that an effortful adjustment process may underlie some anchoring-based biases, research using incentives and process tracing methods does not establish an adjustment mechanism. In addition, the work by Jacowitz and Kahneman (1995) indicates that adjustment is not necessary for the anchoring effect. Thus, although the anchoring phenomenon is well established, an adjustment mechanism is more questionable. Because at least some conceptions of adjustment characterize it as occurring in response expression, the limited support for an adjustment mechanism suggests that the third stage in Fig. 6.1 may not be well supported.