As a case study, we compare two of De Neys’s important proposals to findings from the literature on anchoring effects. Evidence for an anchoring-and-adjustment heuristic supports De Neys’s proposal that systems 1 and 2 are non-exclusive (Epley & Gilovich, Reference Epley and Gilovich2001; Simmons, LeBoeuf, & Nelson, Reference Simmons, LeBoeuf and Nelson2010; Tversky & Kahneman, Reference Tversky and Kahneman1974). The increase in anchoring effects with quantifiable measures of uncertainty (Lee & Morewedge, Reference Lee and Morewedge2022), however, challenges his proposal that uncertainty monitoring drives the involvement of system 2 correction processes in judgment.
As background, an anchoring effect occurs when considering an initial value (i.e., an anchor) biases subsequent estimates of a stimulus (i.e., the target). Inaccurate estimates of the target are more likely to fall between the anchor and the correct answer than beyond the correct answer (Tversky & Kahneman, Reference Tversky and Kahneman1974). When people are asked to estimate the duration of Mars’ orbit, for instance, the number that typically first comes to mind is the duration of Earth's orbit (i.e., 365 days). This anchor influences estimates of Mars’ orbit. People are more likely to underestimate the duration of Mars’ orbit than to overestimate its duration (Epley & Gilovich, Reference Epley and Gilovich2001).
An influential anchoring-and-adjustment heuristic theory of anchoring effects suggests that people make estimates by correcting from intuitive (but wrong) anchors until they reach the first plausible value of the target of their estimate. As the first value in the range of plausible values is usually incorrect, adjustment from anchors tend to be insufficient (Epley & Gilovich, Reference Epley and Gilovich2001; Simmons et al., Reference Simmons, LeBoeuf and Nelson2010; Tversky & Kahneman, Reference Tversky and Kahneman1974). Two forms of evidence from tests of the anchoring-and-adjustment heuristic support De Neys’s proposal that systems 1 and 2 are non-exclusive. First, few participants in anchoring studies give anchors as their final responses (e.g., <2.8% in study 1B, Simmons et al., Reference Simmons, LeBoeuf and Nelson2010). Even participants constrained by cognitive load or intoxication, for instance, would be unlikely to guess the duration of Mars’ orbit to be the same as Earth's orbit. They would guess the duration of Mars’ orbit to be closer to 365 days than thinkers who are unconstrained (Epley & Gilovich, Reference Epley and Gilovich2001, Reference Epley and Gilovich2006). Second, motor movements associated with the rejection and acceptance of answers influence the degree to which people correct from anchors (e.g., head nodding and shaking; Epley & Gilovich, Reference Epley and Gilovich2001). These results suggest system 2 correction occurs even under constraint and system 1 can influence when system 2 correction ends.
Anchoring paradigms are also useful for examining De Neys’s uncertainty monitoring proposal. Anchoring is a bias where the degree of uncertainty within the judge can be quantified. Uncertainty can be expressed as the width of the plausible range of values of the target of judgment (Jacowitz & Kahneman, Reference Jacowitz and Kahneman1995); the distance between the lowest and highest plausible value. This range varies with factors like the expertise of the judge (Smith, Windschitl, & Bruchmann, Reference Smith, Windschitl and Bruchmann2013; Wilson, Houston, Etling, & Brekke, Reference Wilson, Houston, Etling and Brekke1996) and with correlates of uncertainty like psychophysical noise. Because of the increasing psychophysical noise associated with numbers as they increase in magnitude (Feigenson, Dehaene, & Spelke, Reference Feigenson, Dehaene and Spelke2004), the plausible range (uncertainty) of values for estimates of larger numbers is wider than for smaller numbers (Lee & Morewedge, Reference Lee and Morewedge2022; Quattrone et al., Reference Quattrone, Lawrence, Finkel and Andrus1984). People estimate the calories in a small serving of McDonald's French fries be anywhere between 141.48 and 223.46, whereas they estimate the calories in a large serving to be anywhere between 266.98 and 423.36. In other words, they perceive the plausible range of calories in a large serving of French fries to be wider than the plausible range of calories in a small serving of French fries (widths of 156.38 and 81.98 calories, respectively). The same pattern holds for novel unfamiliar stimuli like smaller and larger dot-arrays.
Challenging De Neys’s proposal that uncertainty monitoring determines the activation and engagement of system 2 adjustment processes, epistemic uncertainty does not increase the probability or extremity of system 2 correction from anchors (i.e., system 1 intuitions). De Neys’s proposed mechanism for the intervention of system 2, uncertainty monitoring, implies that correction from anchors should be more likely and more extreme when uncertainty about the value of the target is greater: When the plausible range of values for a target stimulus is wider. However, research participants do not exhibit more correction from anchors when plausible ranges of a target are wider. Correction from anchors is as likely and proportionally similar for targets with wider and narrower plausible ranges. When estimating the number of dots in 35-dot and 273-dot arrays, for instance, people first exposed to a low anchor (a 10-dot array) tend to give answers similar in relative distance from the lowest plausible values of the target. When one examines the raw size of anchoring effects (i.e., absolute values), people exhibit larger anchoring effects when the plausible range of stimulus values are wider than narrower (i.e., when uncertainty is greater). This pattern holds whether people are estimating the number of dots in larger or smaller dot-arrays, the prices of larger and smaller servings of donuts, the weight of larger or smaller dog breeds, the prices of higher or lower rated hotels, or the number of calories in larger and smaller servings of McDonald's French fries (Lee & Morewedge, Reference Lee and Morewedge2022). These findings suggest epistemic uncertainty bounds the extent to which anchors influence judgment. It does not determine the extent of system 2 adjustment from anchors.
More generally, our comparison illustrates the value of anchoring paradigms for tackling the exciting questions De Neys raises about the relationship between intuitive and corrective mental processes.
As a case study, we compare two of De Neys’s important proposals to findings from the literature on anchoring effects. Evidence for an anchoring-and-adjustment heuristic supports De Neys’s proposal that systems 1 and 2 are non-exclusive (Epley & Gilovich, Reference Epley and Gilovich2001; Simmons, LeBoeuf, & Nelson, Reference Simmons, LeBoeuf and Nelson2010; Tversky & Kahneman, Reference Tversky and Kahneman1974). The increase in anchoring effects with quantifiable measures of uncertainty (Lee & Morewedge, Reference Lee and Morewedge2022), however, challenges his proposal that uncertainty monitoring drives the involvement of system 2 correction processes in judgment.
As background, an anchoring effect occurs when considering an initial value (i.e., an anchor) biases subsequent estimates of a stimulus (i.e., the target). Inaccurate estimates of the target are more likely to fall between the anchor and the correct answer than beyond the correct answer (Tversky & Kahneman, Reference Tversky and Kahneman1974). When people are asked to estimate the duration of Mars’ orbit, for instance, the number that typically first comes to mind is the duration of Earth's orbit (i.e., 365 days). This anchor influences estimates of Mars’ orbit. People are more likely to underestimate the duration of Mars’ orbit than to overestimate its duration (Epley & Gilovich, Reference Epley and Gilovich2001).
An influential anchoring-and-adjustment heuristic theory of anchoring effects suggests that people make estimates by correcting from intuitive (but wrong) anchors until they reach the first plausible value of the target of their estimate. As the first value in the range of plausible values is usually incorrect, adjustment from anchors tend to be insufficient (Epley & Gilovich, Reference Epley and Gilovich2001; Simmons et al., Reference Simmons, LeBoeuf and Nelson2010; Tversky & Kahneman, Reference Tversky and Kahneman1974). Two forms of evidence from tests of the anchoring-and-adjustment heuristic support De Neys’s proposal that systems 1 and 2 are non-exclusive. First, few participants in anchoring studies give anchors as their final responses (e.g., <2.8% in study 1B, Simmons et al., Reference Simmons, LeBoeuf and Nelson2010). Even participants constrained by cognitive load or intoxication, for instance, would be unlikely to guess the duration of Mars’ orbit to be the same as Earth's orbit. They would guess the duration of Mars’ orbit to be closer to 365 days than thinkers who are unconstrained (Epley & Gilovich, Reference Epley and Gilovich2001, Reference Epley and Gilovich2006). Second, motor movements associated with the rejection and acceptance of answers influence the degree to which people correct from anchors (e.g., head nodding and shaking; Epley & Gilovich, Reference Epley and Gilovich2001). These results suggest system 2 correction occurs even under constraint and system 1 can influence when system 2 correction ends.
Anchoring paradigms are also useful for examining De Neys’s uncertainty monitoring proposal. Anchoring is a bias where the degree of uncertainty within the judge can be quantified. Uncertainty can be expressed as the width of the plausible range of values of the target of judgment (Jacowitz & Kahneman, Reference Jacowitz and Kahneman1995); the distance between the lowest and highest plausible value. This range varies with factors like the expertise of the judge (Smith, Windschitl, & Bruchmann, Reference Smith, Windschitl and Bruchmann2013; Wilson, Houston, Etling, & Brekke, Reference Wilson, Houston, Etling and Brekke1996) and with correlates of uncertainty like psychophysical noise. Because of the increasing psychophysical noise associated with numbers as they increase in magnitude (Feigenson, Dehaene, & Spelke, Reference Feigenson, Dehaene and Spelke2004), the plausible range (uncertainty) of values for estimates of larger numbers is wider than for smaller numbers (Lee & Morewedge, Reference Lee and Morewedge2022; Quattrone et al., Reference Quattrone, Lawrence, Finkel and Andrus1984). People estimate the calories in a small serving of McDonald's French fries be anywhere between 141.48 and 223.46, whereas they estimate the calories in a large serving to be anywhere between 266.98 and 423.36. In other words, they perceive the plausible range of calories in a large serving of French fries to be wider than the plausible range of calories in a small serving of French fries (widths of 156.38 and 81.98 calories, respectively). The same pattern holds for novel unfamiliar stimuli like smaller and larger dot-arrays.
Challenging De Neys’s proposal that uncertainty monitoring determines the activation and engagement of system 2 adjustment processes, epistemic uncertainty does not increase the probability or extremity of system 2 correction from anchors (i.e., system 1 intuitions). De Neys’s proposed mechanism for the intervention of system 2, uncertainty monitoring, implies that correction from anchors should be more likely and more extreme when uncertainty about the value of the target is greater: When the plausible range of values for a target stimulus is wider. However, research participants do not exhibit more correction from anchors when plausible ranges of a target are wider. Correction from anchors is as likely and proportionally similar for targets with wider and narrower plausible ranges. When estimating the number of dots in 35-dot and 273-dot arrays, for instance, people first exposed to a low anchor (a 10-dot array) tend to give answers similar in relative distance from the lowest plausible values of the target. When one examines the raw size of anchoring effects (i.e., absolute values), people exhibit larger anchoring effects when the plausible range of stimulus values are wider than narrower (i.e., when uncertainty is greater). This pattern holds whether people are estimating the number of dots in larger or smaller dot-arrays, the prices of larger and smaller servings of donuts, the weight of larger or smaller dog breeds, the prices of higher or lower rated hotels, or the number of calories in larger and smaller servings of McDonald's French fries (Lee & Morewedge, Reference Lee and Morewedge2022). These findings suggest epistemic uncertainty bounds the extent to which anchors influence judgment. It does not determine the extent of system 2 adjustment from anchors.
More generally, our comparison illustrates the value of anchoring paradigms for tackling the exciting questions De Neys raises about the relationship between intuitive and corrective mental processes.
Acknowledgments
We thank Nicholas Epley, Dilip Soman, and Ryan Webb for helpful feedback.
Competing interest
None.