ANCHORING LEGAL STANDARDS By YUVAL FELDMAN, AMOS SCHURR AND DORON TEICHMAN* Abstract This Article presents the first empirical study on the way in which irrelevant anchors influence the interpretation of vague legal standards. A large body of psychological research demonstrates that when people make judgments on a continuum, they are often affected by meaningless anchors. Building on this body of work, legal scholars have shown that judicial decisions related to remedies (i.e., damages and penalties) are also influenced by such anchors. This Article extends this insight and hypothesizes that the process of interpreting vague legal norms is subject to an anchoring effect as well. To test this hypothesis, the Article presents a series of stylized experiments that measure and compare participants’ interpretation of a vague norm after they have been exposed to irrelevant anchors. Overall, the results confirm the proposed hypothesis and suggest that the content of substantive legal rules might be altered by anchors. This effect is documented in numerous legal settings and across both expert (i.e., experienced lawyers) and non-expert (i.e., students) populations. Furthermore, the effect is shown to exist when participants express judgments about a hypothetical scenario and when they make decisions with ramifications for others. Based on these findings, the Article revisits several long-standing legal debates and reevaluates their conclusions. * Feldman is a professor at the Faculty of Law, Bar-Ilan University; Schurr is assistant professor at the Faculty of Business and Management, Ben-Gurion University of the Negev; Teichman is the Joseph H. and Belle R. Braun Professor, Faculty of Law, the Hebrew University of Jerusalem and the Justin D'Atri Visiting Professor of Law, Business & Society Columbia Law School. For helpful comments we thank Valerie Hans, Gideon Parchomovsky, Jeff Rachlinski, and participants of workshops at Columbia and Cornell. Financial support for this study was provided by the Israel Science Foundation (grant 1283/11). For valuable research assistance we thank Nathan Johnson and Hadas Berkovitz. Electronic copy available at: http://ssrn.com/abstract=2408193 ANCHORING LEGAL STANDARDS I. INTRODUCTION Imagine that you enjoy listening to Mozart at extraordinarily high volumes (the louder, the better). Your roommate, on the other hand, prefers silence. The norm among roommates is somewhat similar to a nuisance rule; that is, one may listen to music only at a "reasonable volume." Thus, the two of you need to figure out what reasonable volume actually means. Can you do anything to influence your roommate's judgment of what a reasonable volume is in order to better serve your interests? The findings presented in this Article suggest that you can. By making sure that your discussion begins from a high starting point, for example, by conducting it after you have listened to The Magic Flute in its entirety at full volume, you might influence your roommate's judgment as to what constitutes a reasonable volume and cause her to agree to volume levels that she would not otherwise view as reasonable. A large body of psychological research has demonstrated that when people need to make judgments on a continuum, they are often influenced by meaningless anchors.1 In traditional anchoring studies, subjects are first asked to assess whether a certain target is larger or smaller than a given figure (i.e., the anchor) and are then asked to present their absolute estimate of the target. As it turns out, the initial figure, even if it is highly implausible or set randomly, tends to influence the ultimate absolute assessment. For example, when asked to estimate at what age Mahatma Gandhi died, subjects who were first asked whether he died before or after the age of 140 estimated that his age of death was 66.7, whereas subjects who were first asked whether he died before or after the age of 9, estimated that his age of death was 50.1.2 This finding has been replicated numerous times in very distinct contexts and with a broad set of subjects. As one recent review concluded, "[a]nchoring effects are among the most robust and ubiquitous psychological phenomena in judgement and decision making."3 The relevance of anchoring to judicial decision making has not been overlooked by legal scholars. An abundance of studies have forcefully shown that different anchors 1 For a few recent contributions to this body of work see, for example, Karen E. Jacowitz & Daniel Kahneman, Measures of Anchoring in Estimation Tasks, 21 PERSONALITY & SOC. PSYCHOL. BULL. 1161 (1995); Nicholas Epley & Thomas Gilovich, The Anchoring-and-Adjustment Heuristic: Why the Adjustments Are Insufficient, 17 PSYCHOL. SCI. 311 (2006); Daniel Mochon & Shane Frederick Anchoring in Sequential Judgments, 122 ORG. BEHAV. & HUM. DECISION PROCESSES. 69 (2013). For a complete review of the literature see infra part II 1. 2 See Fritz Strack & Thomas Mussweiler, Explaining the Enigmatic Anchoring Effect: Mechanisms of Selective Accessibility, 73 J. PERSONALITY & SOC. PSYCHOL. 437, 442-43, 446 (1997). 3 Thomas Mussweiler, Birte Englich & Fritz Strack, Anchoring Effect, in COGNITIVE ILLUSIONS: A HANDBOOK OF FALLACIES AND BIASES IN THINKING, JUDGEMENT, AND MEMORY 183, 196 (Rüdiger F. Pohl ed., 2004). 2 Electronic copy available at: http://ssrn.com/abstract=2408193 might influence legal outcomes.4 This vast body of work, however, has focused on one dimension of legal decisions: the remedial outcome. In the civil context, it has been shown that anchors alter the amount of damages that is considered to be appropriate. Similarly, in criminal settings, it has been established that anchors can change perceptions of a just punishment. The general result of this body of work is that irrelevant anchors can bias remedies both upward and downward depending on the nature of the anchor. For example, one study conducted with professional judges showed that those judges' sentencing decisions were swayed by an arbitrary anchor that was produced by a toss of a die.5 Despite the tremendous breadth of the legal-anchoring literature, this body of literature has neglected to address a fundamental question: can anchors affect the way in which vague legal standards are interpreted?6 We hypothesize that they can. More specifically, when interpretive decisions lie on a scale of different options, the choice between those options might be driven by anchors. Thus, just as irrelevant anchors might drive the allocation of damages and penalties, they might also affect the substantive content law. Consider, for example, tort law and the application of the negligence standard. When judges or jurors apply this standard in a certain case, they often need to determine how much resources the defendant was required to invest in precautions. This decision lies on a continuum that can begin with absurdly low figures and that can end with ridiculously high amounts. If the decision maker is drawn to a certain anchor in the process of pinpointing the desired standard of care, this anchor might influence the interpretative process. To test this hypothesis, we designed a series of experiments that measured and compared the way in which participants interpreted a vague legal norm. Following the prevailing paradigm of research on judicial decision making, the first four experiments used detailed vignettes that presented a legal dilemma to participants. By including in the text of these vignettes subtle manipulations that created a high or low anchor, we could measure the influence of those anchors on peoples' interpretative decisions. The final experiment used a novel design that enabled the participants to fully experience an activity (i.e., a game involving trivia questions) before they were asked to interpret an unclear norm governing it. Again, the participants in this study were randomly assigned to either a high or a low anchor that was created through nuanced alterations of the instructions. Overall, these studies confirmed our central hypothesis that peoples' 4 For a few recent contributions to this body of work see Jeffrey J. Rachlinski, Andrew J. Wistrich & Chris Guthrie, Arbitrary Adjudication: How Anchoring and Scaling Distort Awards and Sentences (unpublished manuscript) (on file with authors) (2014) (presenting a series of experiments on anchoring); Shari Seidman Diamond et al., Damage Anchors on Real Juries, 8 J. EMP. LEGAL STUD. 5148 (2011) (presenting data on the effects of anchors on jury deliberations). For a complete review of the literature see infra part II 2. 5 See Birte Englich, Thomas Mussweiler & Fritz Strack, Playing Dice With Criminal Sentences: The Influence of Irrelevant Anchors on Experts’ Judicial Decision Making, 32 PERSONALITY SOC. PSYCHOL. BULL. 188, 194-195 (2006). 6 Eyal Zamir & Doron Teichman, Judicial Decision Making, in THE HANDBOOK ON BEHAVIORAL ECONOMICS AND THE LAW (Eyal Zamir & Doron Teichman eds., forthcoming 2014). 3 normative judgments are influenced by anchors significantly. This effect was documented in numerous legal settings ranging from tort law to corporate law, and across both expert (i.e., experienced lawyers) and non-expert (i.e., students) populations. The finding that the operation of legal norms is influenced by anchors has numerous legal implications. Taken at the positive level, this finding sheds new light on the way in which different doctrines operate. For example, our results suggest that legislated safe harbors might influence judicial decisions in ways that are not appreciated by legal scholars. As our analysis shows, even if such safe harbors are nonbinding, their mere existence might pull judges and jurors towards them and alter the way in which they interpret vague standards. On the normative level, our findings raise a wide range of questions related to the desirable way in which the legal system should deal with substantive anchors. These implications are linked to both evidentiary and procedural rules that are geared toward shielding jurors from irrelevant anchors and substantive rules that might influence judicial decision making in a fashion that has not been recognized thus far. For instance, our findings demonstrate that a rule capping comparative negligence at 50% can influence the level of fault attributed to plaintiffs. More specifically, once the law affords special status to the 50% figure, decision makers tend to gravitate towards it. Thus, participants in Experiment 3 almost doubled their fault assessment of the plaintiff's behavior as a result of the anchoring effect generated by the 50% cap. The Article unfolds as follows: after this brief introduction, Part II presents the background to the questions that we explore. It describes the psychological literature on anchoring and decision making and reviews the legal literature that builds on this body of work. Part III outlines the experiments that we conducted and reports their results. Part IV discusses the results, highlights their policy implications, and outlines potential avenues for future research. Finally, Part V offers some brief concluding remarks. II. BACKGROUND In this Section, we review the literature on anchoring. We begin by outlining the psychological literature that first documented the phenomenon. That done, we turn to describe the legal literature that has examined how anchoring affects judicial decision-making. 1. The Psychology of Anchoring An anchoring effect alludes to peoples' tendency to estimate values in relation to certain focal values—anchors—that they are drawn to focus on while making their 4 decisions.7 As a large body of psychological work has demonstrated, anchors may unduly influence peoples' choices. More specifically, anchors might draw decision makers toward them and thus cause decision makers to systematically misestimate target values.8 In the typical anchoring study, subjects are asked to estimate the value of a target quantity, yet before doing so, they are exposed to a certain numeric figure that serves as the anchor. In one of their early papers, Tversky and Kahneman demonstrated how such irrelevant anchors might alter peoples' evaluations.9 The participants in this study were asked to estimate the percentage of African countries in the United Nations. Before giving their estimates, however, the participants observed a spin of a "wheel of fortune" that was rigged to stop at either 10 or 65 and were asked whether the percentage of African countries in the United Nations was higher or lower than the figure that came up on the wheel.10 This initial meaningless spin greatly influenced peoples' decisions. Whereas participants who were exposed to a wheel outcome of 10 estimated the number of countries at 25%, those who were exposed to a wheel outcome of 65 estimated the number at 45%.11 Numerous studies have replicated this result and demonstrated the centrality of anchors to our decisions. Typical psychological studies examine whether and how anchors effect the assessment of factual questions, such as the length of the Mississippi River,12 the height and width of the Brandenburg Gate,13 the number of countries in the United Nations,14 and the year during which Einstein first visited the United States.15 At some level, these findings are unsurprising. When people are asked to estimate values that they are clueless about, they may hold on to any piece of information floating out there.16 More recent findings on anchoring, however, have demonstrated that the phenomenon is not limited to the narrow category of estimating obscure facts. For instance, Dan Ariely and his colleagues conducted a study in which they demonstrated that anchors can influence peoples' willingness to pay for goods, such as a bottle of 7 See, e.g., DANIEL KAHNEMAN, THINKING, FAST AND SLOW, 119-120 (2011). Id. 9 See Amos Tversky & Daniel Kahneman, Judgment Under Uncertainty: Heuristics and Biases, 185 SCI. 1124, 1128 (1974). 10 Id. 11 Id. 12 Daniel M. Oppenheimer et al., Anchors Aweigh: A Demonstration of Cross Modality Anchoring and Magnitude Priming, 106 COGNITION 13, 16-17 (2008). 13 See Fritz Strack & Thomas Mussweiler, Explaining the Enigmatic Anchoring Effect: Mechanisms of Selective Accessibility, 73 J. PERSONALITY & SOC. PSYCHOL. 437, 439-440 (1997). 14 See Timothy D. Wilson et al., A New Look at Anchoring Effects: Basic Anchoring and Its Antecedents, 125.4 J. EXPERIMENTAL PSYCHOL. 387, 390-392 (1996). 15 See Strack & Mussweiler, supra note 13, at 442-443. 16 Kahneman, supra note 7, at 125 (noting that "it is not surprising that people who are asked difficult questions clutch at straws, and the anchor is a plausible straw"). 8 5 wine.17 Their results showed that subjects who were exposed to a high anchor were willing to pay more than subjects who were exposed to a low anchor. 18 In a different study, LeBoeuf and Shafir demonstrated that anchoring influences the way in which people assess physical stimuli, such as length, weight, and sound.19 In one of their experiments, participants first listened to a music clip at a volume level of 35 (the participants could not see the numeric representations of volume throughout this experiment).20 At the second stage of the experiment, the participants viewed the clip again and were asked to adjust the volume to replicate the volume level that they just previously heard.21 While half of the subjects started this process from a level of 1 and were required to adjust the volume upward (i.e., low anchor), the other half started the process from a level of 70 and were required to adjust to volume downward (i.e., high anchor).22 The results showed that even in this non-numeric, purely physical setting, anchoring affected peoples' choices. The participants in the low anchor group chose a volume level that was significantly lower than those in the high anchor group.23 A central aspect of anchoring studies is that they usually build on an uninformative anchor. As noted above, in their original anchoring experiment, Tversky and Kahneman used a wheel of fortune to generate the anchor.24 Later studies used other tools, such as the result of a die toss and subjects' social security number.25 It is this nature of anchors that enables us to interpret the phenomenon as a bias—there is no reason that your valuation of a rare bottle of wine should be influenced by the last two digits of your social security number.26 As for the explanations for the phenomenon, the psychological literature has identified two potential mechanisms that might drive anchoring. The first focuses on the process of adjustment.27 According to this line of thought, the anchor serves as the starting point for the analysis, and people slowly adjust their estimates from the anchor toward their final estimate. However, this adjustment process tends to end prematurely, and as a result, final estimates are biased in the direction of the anchor. A second theory of anchoring focuses on the suggestive process that is triggered by the anchor.28 Dan Ariely, George Loewenstein & Drazen Prelec, “Coherent Arbitrariness”: Stable Demand Curves Without Stable Preferences, 118 Q. J. ECON. 73 (2006). 18 Id. at 75-77. 19 Robyn A. LeBoeuf & Eldar Shafir, The Long and Short of It: Physical Anchoring Effects, 19 J. BEHAV. DECISION MAKING 393 (2006). 20 Id. at 400-401. 21 Id. 22 Id. 23 Id. 24 Tversky & Kahneman, supra note 7, at 1144. 25 See Englich, Mussweiler & Strack, supra note 5, at 194-195 (die toss); Ariely, Loewenstein & Prelec, supra note 17, at 75-77 (last two digits of participants' social security number). 26 See Ariely, Loewenstein & Prelec id. 75-77 (reporting experimental results on pricing). 27 For a more detailed review of the theory and the studies supporting it see Kahneman, supra note 7, at 120-22. For a critical review of the theory see Gretchen B. Chapman & Eric J. Johnson, Incorporating the Irrelevant, in HEURISTICS AND BIASES: THE PSYCHOLOGY OF INTUITIVE JUDGMENT 120, 127-130 (Thomas Gilovich, Dale Griffin & Daniel Kahneman eds., 2002). 28 For a more detailed review of the theory and the studies supporting it see Kahneman, id. at 122-23; Chapman & Johnson id. at 130-33. 17 6 According to this line of thought, anchoring is an automatic process that occurs subconsciously. The anchor focuses our attention on a certain potential answer to the question that we face and causes us to retrieve from our memory information that is consistent with the anchor as a plausible solution. 2. Anchoring and Judicial Decision Making Legal scholars have not overlooked the significant role that anchoring might play in the courtroom. Numerous studies have shown that an array of anchors might influence judicial decisions. As the following review shows, however, these studies have exclusively focused on the outcomes of trials: the amount of damages awarded to plaintiffs in civil suites and the size of the penalty imposed on defendants in criminal proceedings. Most of the legal-anchoring scholarship has focused on damages in civil litigation. This body of work has demonstrated the influence of external anchors on determinations of damages. Numerous studies have shown that plaintiffs are better off when they request high amounts of damages, since the requested amount can serve as an anchor that pushes the award upward.29 Other studies have shown how different types of damages can serve as anchors. Using data obtained from actual cases, Hans and Reyna showed that the amount of compensatory damages might serve as an anchor for jurors in setting the amount of punitive damages.30 Arguably, however, these factors might have some bearing on the appropriate damages.31 It is completely rational to assume that greater compensation demands reflect greater harm especially if those demands are later examined in trial.32 Similarly, one can argue that holding everything else equal, defendants who caused greater harm should pay higher punitive damages as well.33 Anchoring also suggests that numeric figures that decision makers are required to ignore might nonetheless function as a point of reference that influences the amount of damages that plaintiffs are awarded. For example, settlement offers that adjudicators are exposed to might affect the level of damages that are awarded if negotiations fail 29 See, e.g., W. Kip Viscusi, The Challenge of Punitive Damages Mathematics, 30 J. LEGAL STUD. 313, 319-334 (2001); Reid Hastie, David A. Schkade & John W. Payne, Juror Judgments in Civil Cases: Effects of Plaintiff's Requests and Plaintiff's Identity on Punitive Damage Awards, 23 L. & HUM. BEHAV. 445, 465-467 (1999); John Malouff & Nicola S. Schutte, Shaping Juror Attitudes: Effects of Requesting Different Damage Amounts in Personal Injury Trials, 129 J. SOC. PSYCHOL. 491, 493-495 (1989); Mollie W. Marti & Roselle L. Wissler, Be Careful What You Ask For: The Effect of Anchors on Personal-Injury Damages Awards, 6 J. EXPERIMENTAL PSYCHOL. 91, 100-101 (2000). Relatedly, even a statement by the plaintiff regarding an award in a court TV show was shown to function as an anchor. See Chris Guthrie, Jeffrey J. Rachlinski & Andrew J. Wistrich, The “Hidden Judiciary”: An Empirical Examination of Executive Branch Justice, 58 DUKE L.J. 1477, 1501-1506 (2009). 30 See Valerie P. Hans & Valerie F. Reyna, To Dollars from Sense: Qualitative to Quantitative Translation in Jury Damage Awards, 8 J. EMPIRICAL LEGAL STUD. 120, 144 (2011). 31 Zamir & Teichman, supra note 6. 32 Id. 33 Id. 7 and the case moves to trial.34 Again, however, while this information might not be allowed to influence judicial decisions, it could reasonably be considered relevant to determinations of damages. In one stylized experiment, Guthrie, Rachlinski, and Wistrich managed to construct a decision-making environment in which a purely irrelevant factor could influence damages judgments.35 The anchoring manipulation in this study involved a meritless motion to dismiss a tort case involving significant harm merely because it did not meet the minimum threshold of damages that are required to establish jurisdiction. While half of the subjects in this experiment were first asked to rule on this motion, the other half proceeded directly to assess the damages (the level of damages was the only disputed issue in the case).36 The subjects who first decided whether the claim met the low threshold (and they practically unanimously decided it did) awarded the plaintiffs significantly less damages than the subjects who were not exposed to the jurisdictional question.37 Finally, researchers have also examined the ability of legislated award caps to function as anchors. Numerous legislatures have attempted to rein in juries by setting maximal levels for punitive damages.38 Presumably, the rationale behind this tool is that the cap will cut off the upper end of the damages distribution and thus lower the amount of damages that are awarded. Behavioral researchers have shown, however, that this rationale overlooks the ability of caps to function as an anchor. Empirical findings suggest that caps can inadvertently bring about a rise in damages awards since they shift the entire distribution upward toward them.39 Aside from the damages context, researchers have also studied the influence of anchoring on criminal sentences. Englich and Mussweiler reported that much like the amount of damages requested by plaintiffs in civil litigation, prosecutors' sentencing requests can function as an anchor.40 Yet one should acknowledge that this figure is arguably relevant to the task of determining the appropriate sentence since prosecutors often hold private information regarding the details of cases. In a later study, it was demonstrated that even a clearly irrelevant anchor might also influence penal 34 See Andrew J. Wistrich, Chris Guthrie & Jeffrey J. Rachlinski, Can Judges Ignore Inadmissible Information? The Difficulty of Deliberately Disregarding, 153 U. PENN. L. REV. 1251, 1286-1291 (2005). 35 Chris Guthrie, Jeffrey J. Rachlinski & Andrew J. Wistrich, Inside the Judicial Mind, 86 CORNELL L. REV. 777, 787-794 (2001). 36 Id. at 790-791. 37 Id. at 791-792. 38 See Edith Greene, David Coon & Brian Bornstein, The Effects of Limiting Punitive Damage Awards, 25 L. & HUM. BEHAV. 217, 231-233 (2001); Jane C. Arancibia, Statutory Caps on Damage Awards in Medical Malpractice Cases, 13 OKLA. CITY U. L. REV. 135, 139-140 (1988). 39 See, e.g., Jennifer K. Robbennolt & Christina A. Studebaker, Anchoring in the Courtroom: The Effects of Caps on Punitive Damages, 23 L. & HUM. BEHAV. 353, 367 (1999) (presenting the results of experiments conducted with lay people); Rachlinski, Wistrich & Guthrie, supra note 4 (presenting the results of experiments conducted with professional judges). 40 Birte Englich & Thomas Mussweiler, Sentencing Under Uncertainty: Anchoring Effects in the Courtroom, 31 J. APPLIED SOC. PSYCHOL. 1535, 1538-1546 (2001). 8 decisions.41 Sentencing requests in this study were determined by the throw of a pair of dice that were rigged to roll either 3 or 9.42 The results showed that subjects were significantly influenced by the random anchor. Whereas the subjects who rolled the high anchor (i.e., 9) chose to impose an average sentence of 7.81 months of probation, those who rolled the low anchor (i.e., 3) imposed an average sentence of only 5.28 months of probation.43 Despite the magnitude of the legal literature on anchoring, the current focus of this body of literature is exclusively on the remedial outcomes of trials: damages awards in the civil context and penalties in the criminal context.44 It is quite natural that the literature began to evolve from these topics, since the original anchoring literature focused on numeric estimates and since remedies by their very nature are often placed on a numeric scale. Nonetheless, judicial decisions related to the substantive content of law may also be influenced by anchors. For example, when a juror needs to decide what level of noise constitutes a nuisance, she might be influenced by different anchors that she is exposed to during litigation (e.g., examples of noise levels, customs related to noise levels). Thus, much like the level of damages and punishment, law itself might be subject to an anchoring effect. Shifting the analysis toward legal standards, however, requires empirical verification, as there are forces that might limit the influence of anchors in this domain. The way that legal standards are interpreted has been studied by legal scholars approaching the issue from economic,45 institutional,46 sociological,47 and 41 Englich, Mussweiler & Strack, supra note 5. Id. at 194. 43 Id. 44 One paper that touches on the relationship between anchoring and substantive legal rules is Gretchen B. Chapman & Brian H. Bornstein, The More You Ask for the More You Get: Anchoring in Personal Injury Verdicts, 10 APPLIED COGNITIVE PSYCHOL. 519 (1996). While the focus of this paper is on the effects of anchors on the judgment size, it also reports on the way in which anchors influence causality judgments. In the first experiment reported in the paper the authors document an intriguing correlation between the amount of damages asked for and judgments of causality. Id. at 523-28. As plaintiffs ask for more money, this causes fact finders to view the causation question more favorably. The authors do not offer a clear explanation for this finding, which arguably might result from the fact that the low anchor manipulation included a ridiculously low demand for damages, $100. Arguably, asking for such modest compensation might have caused people to view the claim itself as meritless. The second experiment involved an anchor that was directly related to the strength of the evidence relating to causality. Id. at 528-33. Nonetheless, it is unsurprising that evidence that was relevant to causality judgment influenced them. As the authors acknowledge, "the effect of causal evidence is rational and not an anchoring bias." Id. at 530. 45 See Richard Craswell & John E. Calfee, Deterrence and Uncertain Legal Standards, 2 J. L. ECON. & ORG. 279, 279-280 (1986) (discussing the likelihood of over- and under-compliance); Louis Kaplow, Rules Versus Standards: An Economic Analysis, 42 DUKE. L.J. 279, 310-317 (1992) (discussing the information costs associated with rules and standards). 46 Gunther Teubner & Oren Perez, The Institutionalisation of Inconsistency: From Fluid Concepts to Random Walk, in PARADOXES AND INCONSISTENCIES IN THE LAW, 119 (Gunther Teubner & Oren Perez eds., 2006) (arguing that ambiguity allows for groups with conflicting ideological views to agree to disagree). 47 Lauren B. Edelman, Legal Ambiguity and Symbolic Structures: Organizational Mediation of Civil Rights Law, 97 AM. J. SOC. 1531, 1554-1569 (1992) (demonstrating empirically how organizations use ambiguity to create their own policies). 42 9 philosophical48 perspectives. This body of research suggests that when it comes to legal standards, there are numerous moral, social, motivational, and institutional forces that influence the way that people evaluate the connection between a given behavior and the application of the relevant standard. Many of these factors might limit the ability of anchors to shape peoples' choices. Interestingly, in documenting how anchors influence the way that people remember physical stimuli, such as sound, LeBoeuf and Shafir seem to draw the conclusion that these anchors also influence normative judgments regarding those stimuli.49 More specifically, they speculate that "[a] well-intentioned child who turns down exceptionally loud music to meet a parent’s demand that it be played at a ‘reasonable’ volume may fail to adjust sufficiently from a high anchor."50 Note, however, that this assertion does not stem from their empirical findings. Subjects in LeBoeuf and Shafir's study faced a purely descriptive task—they were required to identify a volume level that they had previously heard.51 This task does not entail a normative judgment as to what is a reasonable volume is.52 In the following section, we present a series of experiments in which we aim to fill this gap in the literature and show that anchors may influence the way that people interpret vague legal terms. III. EXPERIMENTAL RESULTS As the preceding review suggests, the effect of anchors on the interpretation of vague legal standards has yet to be studied. In this section, we present a series of four experiments in which we examined the extent to which irrelevant anchors might alter the way in which people resolve legal uncertainty. All studies used a between-subject design in which the participants were randomly assigned to one of the different anchoring manipulations. This experimental design enabled us to isolate the effect of anchors and to measure their influence on the way in which subjects interpret vague norms. 1. Experiment 1: Setting the Standard of Care in Torts 48 ANDREI MARMOR, INTERPRETATION AND LEGAL THEORY 28 (1992) (discussing the jurisprudential aspects of what legal interpretation is). 49 LeBoeuf & Shafir, supra note 19, at 400-401. 50 Id. at 404. Relying on these findings Daniel Kahneman has also argued that this mechanism can explain why people might drive unreasonably fast when exiting the highway into the city. See Kahneman, supra note 7, at 121. 51 LeBoeuf & Shafir, id. at 400. 52 The closest psychological study we could find that touches on the topic of anchoring normative judgments is Barry Markovsky, Anchoring Justice, 51 SOC. PSYCHOL. Q. 213 (1988). While this study examines how anchors influence fairness judgments, much like the legal literature, it focuses on questions relating to the final monetary treatment people are afforded (i.e., how much money should be paid as a reward or as a salary). 01 Our first experiment focused on one of the paradigmatic legal standards—the negligence standard in tort law. Stated generally, the negligence standard alludes to the duty of an actor to "exercise reasonable care when the actor’s conduct creates a risk of physical harm."53 (a) Experiment 1A: Examples as Anchors One hundred and seven law students at an Israeli university participated in the experimental survey in return for two cash prizes that were worth approximately $25. Their age ranged between 18 and 41 years, with a mean of 23.66 years. The study used a between-subjects design and included three conditions: High Anchor, Low Anchor, and Control. The participants were randomly assigned to the conditions. The participants in all conditions were asked to evaluate a tort case in which the water company “Spring Water” was sued by Simon, one of the company’s clients, on grounds of negligence. The case described significant damage that was caused to Simon’s bakery, due to a breakdown in one of the sewer lines to which he is connected. All participants learned that according to the law, a citizen may sue a water company owing to its failure to perform reasonable repairs within the sewage system. The practical implication of this rule is that it is upon the court to balance between the chance that a malfunction could occur in the sewer line and the extent of the damage that would be caused if repairs were not performed, on one hand, and the costs of the repairs, on the other. The vignette then went on to describe the undisputed factual background of the case, which was based on the findings of a court-appointed expert. This background suggested that the sewer line was showing some wear and tear, yet it was not clear that it required immediate attention. The background also included the different potential contingencies with respect to the amount of damage that a rupture to the sewer line could cause. Following these facts, the vignette specified the following ten potential measures that the water company could have taken to reduce the probability and magnitude of harm (see Table 1). All measures were presented with their relative probability, their operational/actual consequences, and the costs associated with them. Finally, the factual background also included the actual level of care that the company took, which was the third level on this scale. 53 RESTATEMENT (THIRD) OF TORTS: LIABILITY FOR PHYSICAL & EMOTIONAL HARM § 7A (2009). 00 TABLE 1: DIFFERENT LEVELS OF POTENTIAL CARE AND CORRESPONDING RISK PRESENTED TO PARTICIPANTS IN EXPERIMENT 1 Safety Level 0 Cost Details Comment 511,111 N.I.S. Installing electronic monitoring equipment. This monitoring equipment would not prevent a rupture, but it would reduce its damage. Electronic monitoring equipment is more reliable and more expensive than nonelectronic monitoring equipment. Making small repairs in places where the probability of a leak is high (5% of the pipeline) Making small repairs in places where the probability of a leak is moderate to high (15% of the pipeline) Making small repairs in places where the probability of a leak is moderate (30% of the pipeline) Making small repairs in places where the probability of a leak is low (50% of the pipeline) Replacing the pipeline in places where the probability of a leak is high (5% of the pipeline) Replacing the pipeline in places where the probability of a leak is moderate to high (15% of the pipeline) Replacing the pipeline in places where the probability of a leak is moderate (30% of the pipeline) Replacing the pipeline in places where the probability of a leak is low (50% of the pipeline) Replacing the entire pipeline Presented in court 2 811,111 N.I.S. 3 0,211,111 N.I.S. 4 0,811,111 N.I.S. 5 2,511,111 N.I.S. 6 3,211,111 N.I.S. 7 4,411,111 N.I.S. 8 7,111,111 N.I.S. 9 01,511,111 N.I.S. 01 07,111,111 N.I.S. Presented in court Chosen level of care Presented in court Presented in court Presented in court Presented in court Presented in court Presented in court Presented in court After the participants read the same vignette, all they were asked to apply the negligence rule to the facts of the case. More specifically, they were instructed to render their opinion regarding the legally required level of care in the described scenario. This judgment, in turn, served as the dependent variable of this experiment.54 54 “Dependent variable” refers to the variables that are explained by the model. 02 According to the psychological literature, for an anchor to be effective, the decision maker needs to view it as a potential response to the target question.55 Thus, our anchoring strategy was to present the subjects with two examples of how the negligence rule functions before we asked them to apply it to the facts of the case. The examples were essentially the same—but used different anchors. Specifically, the participants in the High Anchor condition were exposed to an example based on the highest possible level of care (i.e., replacement of the entire sewer pipe at a very high cost). They were told that if they set the negligence standard at that level, then the legal outcome would be a ruling of negligence since the company took less care than required. The participants in the Low Anchor condition, on the other hand, were exposed to an example based on the lowest possible level of care (i.e., installation of electronic monitoring equipment at a very low cost). They were told that if they set the negligence standard at that level, then the legal outcome would be a ruling of nonnegligence since the company took more care than required. Finally, the participants in the Control condition were not presented with any example regarding the way in which the negligence rule operates. The experimental design that was used in this study has three useful features. First, the selected anchor is irrelevant. The wording of the manipulation provided no indication that the example held any normative power regarding the desirable outcome. Second, by choosing the two extreme and arguably illogical options, we further assured that the participants understood that the example did not carry any substantive meaning. Finally, despite its simplicity, the design carries into the courtroom directly— presenting an example of how a rule could function is a strategy that litigants could easily use. Figure 1 summarizes the main findings of experiment 1A. It shows that the anchors had a significant effect on the participants' judgment regarding the required level of care.56 While the mean required level of care for the Control condition was 4.79 (stdev = 1.37),57 the required care levels in the Low Anchor and High Anchor conditions were respectively set above and below the control condition (M = 4.08, stdev = 0.92, M = 5.636, stdev = 0.243). A one-way ANOVA analysis revealed that the conditions differed significantly (F(2,104) = 11.35, p < 0.001), indicating that the different anchors had a significant effect on the participants’ judgments.58 A post-hoc 55 Timothy D. Wilson et al., A New Look at Anchoring Effects: Basic Anchoring and its Antecedents, 4 J. EXPERIMENTAL PSYCHOL. 387, 399 (1996). 56 A significant effect in a statistical measurement relates to the odds that a certain result was created by chance. In the context of this paper, every time a difference or a result is presented as “significant,” it means that there is less than a 5% likelihood that this difference was coincidental. The 5% level of significance is a common threshold used in statistical analysis in the social sciences. A 6% to 10% level of significance is commonly considered marginally significant. 57 The term “standard deviation” (abbreviated in the text as “stdev”) refers to the level of dispersion of the data from the mean. 58 An “ANOVA” is a very common statistical technique, which aims to identify the sources of variance among participants. In our design, the purpose of the statistical analysis is to examine whether the experimental groups are different from each other. The procedure allows us to tell how much of the difference between participants could be attributed to the assignment to the different subgroups while 03 analysis (Scheffe) revealed that the judgments for the High Anchor and Low Anchor conditions differed significantly (p < 0.001). It also revealed a significant difference in the judgments between the High Anchor and the Control condition (p=0.05) and a marginally significant difference in the judgments between the Low Anchor and the Control condition (p < 0.1). FIGURE 1: THE MEAN LEVEL OF CARE THAT RESPONDENTS VIEWED AS REASONABLE IN THE DIFFERENT CONDITIONS 10 Level of Negligence 8 5.64 6 4.79 4.08 4 2 0 Low Anchor No Anchor (control) High Anchor (b) Experiment 1B: Extreme Options as an Anchor The participants in our second study included seventy-nine Master in Legal Studies students at an Israeli university. Their age ranged between 24 and 68 years, with a mean of 37.54 years. The subjects faced a similar fact pattern to that used in Experiment 1A, yet we used a distinct anchoring procedure. The subjects in the Low Anchor condition were presented with the level of care taken by the defendant and four additional levels of care that could have been taken: two cheaper and less effective measures that were presented by the defendant during trial and two costlier and more effective measures that were presented by the plaintiff during trial. In the High Anchor condition, the subjects were presented with the level of care that was taken by the defendant and nine additional levels of care that could have been taken: two cheaper and less effective measures that were presented by the defendant during trial (as in the low condition), and seven costlier and more effective measures that were presented by the plaintiff during trial. Our hypothesis was that the highest preventative measure that controlling for possible variations. In plain words, the statistical analyses tell us how much of the differences in the participants’ responses can be explained by the type of anchor to which they were exposed. 04 was presented by the plaintiff would bias determinations of the required level of care upward.59 Figure 2 summarizes the main findings of experiment 1B. It shows the same pattern that we observed in Experiment 1A: when the participants in the Low Anchor condition were presented with only five possible measures, the mean level of care was 3.22 (stdev = 1.17). In contrast, when the participants in the High Anchor condition were presented with ten possible measures, the mean level of care was 4.74 (stdev = 2.14). A comparison of the two experimental conditions revealed a significant difference in the selected mean level of care (t(77) = 3.95, p < 0.001), indicating that the anchors had a significant impact on the participants’ judgments. FIGURE 2: THE MEAN LEVEL OF CARE THAT RESPONDENTS VIEWED AS REASONABLE IN THE DIFFERENT CONDITIONS 10 Preventive Measure 8 6 4.74 4 3.22 2 0 Low Anchor High Anchor 2. Experiment 2: Experts In the first two experiments, we demonstrated that anchors might influence determinations of the standard of care in tort cases. Our next study aimed to broaden this finding along two dimensions. First, the substance matter of the experiment focused 59 A related behavioral phenomenon that is tied to the treatment employed in this experiment is the compromise effect. The compromise effect alludes to people's tendency to prefer options that are presented as an intermediate, rather than an extreme, choice. For an early depiction of the phenomenon see Amos Tversky & Itamar Simonson, Context-Dependent Preferences, 39 MGMT. SCI. 1179, 1183 (1993); for a demonstration of its legal implications see Mark Kelman, Yuval Rottenstreich & Amos Tversky, Context-Dependence in Legal Decision Making, 25 J. LEGAL STUD. 287, 290-295 (1996). Theoretically, the influence of the two mechanisms is somewhat related. While the anchoring effect relates solely to the effect of the added options, the compromise effect relates to the way in which these added options appear vis-à-vis other options. Notwithstanding this nuance, the core insight of both phenomena is the same: irrelevant background stimuli cause people to change their judgments. 05 on corporate law rather than tort law. By broadening our examination into other legal areas, we can assure that the results of the first two experiments were not an artifact of the unique context of tort law. Second, the subjects of the experiment were practicing lawyers in high-end law firms who are experts in the relevant area of law. This distinct subject pool enabled us to examine whether expertise can help to eliminate the effect of anchors. One might conjecture that experts will not be influenced by irrelevant anchors when they make professional decisions. Arguably, knowledge and experience will attenuate the impact of the anchors.60 Empirical findings, however, suggest that experts are not immune from the effects of anchors.61 For example, Northcraft and Neale conducted a controlled experiment in which experts (i.e., real estate brokers) and nonexperts (i.e., students) were asked to evaluate the value of property.62 The subjects were randomly assigned to either a high or a low asking price, which served as the anchor in the experiment.63 The results showed that both the experts and the non-experts were significantly influenced by the list price: a higher list price elicited higher valuations and vice versa.64 What experts do differ on in this regard is their awareness of the bias, however. While unprofessional subjects concede that they are influenced by anchors, experts presume that they are immune to its effect.65 Similarly, empirical findings show that legal experts (i.e., judges) are also influenced by anchors. Guthrie, Rachlinski, and Wistrich documented the effect of irrelevant anchors on compensation rulings made by professional judges in a controlled experimental setting.66 When discussing their body of work on the matter, they concluded that anchors "triggered intuitive, automatic processing that the judges were unable to override."67 While the studies conducted by Guthrie, Rachlinski, and Wistrich clearly show that anchors can influence legal experts in their area of expertise, they leave open the 60 For a study that shows the moderating effect of knowledge on anchoring effects in certain types of judgment tasks (but not in others) see Birte Englich, When Knowledge Matters—Differential Effects of Available Knowledge in Standard and Basic Anchoring Tasks, 38 EUR. J. SOC. PSYCHOL. 896, 901 (2008). For similar results in a judicial context see Thomas Mussweiler & Fritz Strack, Numeric Judgments under Uncertainty: The Role of Knowledge in Anchoring, 36 J. EXPERIMENTAL SOC. PSYCHOL. 495, 513-516 (2000). 61 See, e.g., Edward J. Joyce & Gary C. Biddle, Anchoring and Adjustment in Probabilistic Inference in Auditing, 19 J. ACCT. RES. 120 (1981) (presenting findings on the effect of anchors on accountants); Gregory B. Northcraf & Margaret A. Neale, Experts, Amateurs, and Real Estate: An Anchoring-andAdjustment Perspective on Property Pricing Decisions, 39 ORG. BEHAV. & HUM. DECISION PROCESSES 84 (1987) (presenting findings on the effect of anchors on real estate brokers). For further research on anchoring and expertise see Dan Orr & Chris Guthrie, Anchoring, Information, Expertise, and Negotiation: New Insights from Meta-Analysis, 21 OHIO ST. J. ON DISP. RESOL. 597 (2005). 62 Northcraf & Neale, id. at 87-94. 63 Id. 64 Id. 65 Id. at 95. 66 See Guthrie, Rachlinski & Wistrich, supra note 35 at 787-794; Guthrie, Rachlinski & Wistrich, supra note 29, at 1501-1506. 67 See Chris Guthrie, Jeffrey J. Rachlinski & Andrew J. Wistrich, Blinking on the Bench: How Judges Decide Cases, 93 CORNELL L. REV. 1, 21 (2007). 06 question of whether legal experts are relatively better at dealing with irrelevant anchors compared with non-experts. This research gap was recently filled by Englich, Mussweiler, and Strack, who conducted an anchoring experiment on criminal sentencing.68 The subjects in this experiment included both judges who specialize in criminal law and judges who specialize in civil law.69 The results of the study showed that judges in both areas were equally influenced by an irrelevant anchor that they were exposed to.70 To test whether anchors influence legal experts' interpretation of a vague standard, we conducted our experiment with practicing lawyers in six leading commercial law firms in Israel. The survey in this experiment was uploaded to an online platform and was distributed to the lawyers through the mailing lists of their firms. This process elicited replies from a total of 56 lawyers. The mean age of participants was 36 years, their average experience as lawyers was 5.89 years, and their average experience in corporate law was 3.51 years. The experimental materials asked the subjects to put themselves in the shoes of the in-house counsel of Long Talk Ltd., a publicly traded firm involved in the import of telecommunication equipment. According to the vignette, after several difficult years resulting from a slowdown in the field, the past year was very good for the company. During the year, Long-Talk signed numerous contracts and became profitable. Likewise, the future of the company appears good. The company has a significant backlog of orders, and it expects to continue to increase its profitability. In the past year, the price of Long Talk shares rose by 135%, although a slight decline occurred in the value of the company during the past two months. The subjects were further informed that Long Talk has a controlling shareholder who owns 55% of its stock. The legal dilemma that the subjects were asked to evaluate concerned the potential distribution of a dividend from Long Talk to its shareholders. More specifically, the materials noted that because of the difficult years that Long Talk went through and a number of technical changes in accounting rules, the company did not to have any retained earnings to distribute. According to section 302 of the Company Code, in order for a company to be able to distribute a dividend, it first needs to satisfy the "earnings test," which is an accounting test. In order to overcome this obstacle, the controlling shareholder wishes to bring Long Talk to request approval for a dividend distribution from the court. Section 303(a) of the Company Code establishes the legal framework that applies in such situations and stipulates that "the court may, at the request of a company, approve for it [the company] to execute a distribution that does not satisfy the earnings test, provided that it [the court] is convinced that it [the company] satisfies the solvency test." The solvency test to which the section refers is a 68 Englich, Mussweiler & Strack, supra note 5. Id. at 190-194. 70 Id. at 193. That said, legal scholars argue that judges are expected to be relatively less influenced by anchors. See Hans & Reyna, supra note 30, at 144 (noting that "anchors presented during trial should have a stronger impact on jury assessments than on judge assessments"). 69 07 legal test that examines the influence of the dividend distribution on the solvency of a company. According to the Code, it is permissible to distribute a dividend provided "that there does not exist a reasonable concern that the distribution will prevent the company from fulfilling its existing and expected financial obligations at the time they are due." At this point, the vignette specified that the board deliberated on the matter, based on the report of an independent expert who examined the possible effects of dividend payments on the financial stability of the company. The materials delineated the expert's findings regarding the financial situation of the company and the main risks that it faced and provided an evaluation of the risk created by each level of dividend distribution. For example, a dividend of 5 to 25 million NIS was described as "a very small dividend that is not expected to change substantially the financial situation of the company" and was found (by the expert) to create a "very low" risk. On the other hand, a dividend of 100 to 200 million was described as "a dividend that is expected to change the financial situation of the company. It generates a number of different risks, and the materialization of one of them (the probability of which is moderate) would be sufficient in order for the dividend to affect the ability of the company to repay some of its debts." A dividend of this amount was found (again, by the expert) to create a "moderate" risk. Based on this legal and factual background, the participants were asked to give their best estimate of the maximal amount of money that a court would approve to distribute as a dividend to the shareholders of Long Talk. This figure, in turn, was the dependent variable in this experiment. The anchor that was introduced in this study was the amount of money suggested by the controlling shareholder to the board. Whereas half of the subjects were informed that the controlling shareholder suggested that the board distribute a dividend of 5 million NIS, the other half were informed that she suggested that the board distribute a dividend of 80 million NIS. Since this amount reflects the private interests of the controlling shareholder, it should not influence the interpretation of the solvency test that focuses on the interests of the debtors. Moreover, in order to stress to the subjects that the controlling shareholder was furthering her self-interests and not the long-term interests of the firm, they were informed that the amount of money that she asked to be distributed "corresponded precisely with an upcoming interest payment on the loan with which she purchased control of the company." Figure 3 summarizes the main findings. When the respondents were presented with a low anchor, the mean dividend was 26,370,370 NIS. When the respondents were presented with a high anchor, the mean dividend was 46,558,931 NIS. A formal comparison of the conditions revealed that the difference was significant (t(54) = 2.89, p< 0.001). This finding indicates that anchors also affect the judgments of professionals in their area of expertise. 08 FIGURE 3: THE MEAN DIVIDEND DISTRIBUTION THAT RESPONDENTS VIEWED AS REASONABLE IN THE LOW-ANCHOR AND HIGH-ANCHOR CONDITIONS ₪ 50,000,000 NIS 46,558,931 ₪ 45,000,000 ₪ 40,000,000 ₪ 35,000,000 ₪ 30,000,000 NIS 26,370,370 ₪ 25,000,000 ₪ 20,000,000 ₪ 15,000,000 ₪ 10,000,000 ₪ 5,000,000 ₪Low Anchor High Anchor Experiment 3: Comparative Negligence Experiments 1 and 2 showed that the way in which an interpretative dilemma is presented can anchor peoples' judgments. In our third experiment, we explored whether law itself can function as an anchor. At times, a legal doctrine might elevate the salience of specific interpretative options. This elevated salience, in turn, might generate an anchoring effect and draw decision makers toward the answer highlighted by the law. As one illustrative example, consider the doctrine of comparative negligence in tort law. Generally, adjudicators operating under such a legal regime are bestowed with the task of appropriating the costs of the accident between the plaintiff and the defendant according to their relative fault.71 While such a regime might be both fair and efficient,72 it leaves significant discretion for decision makers when they are called to decide how precisely to distribute liability. As a practical matter, the law offers limited guidance as to whether responsibility should be divided at a 20:80 ratio or a 30:70 ratio, even though such a decision might have substantial implications.73 71 See, e.g., VICTOR E. SCHWARTZ & EVELYN F. ROWE, COMPARATIVE NEGLIGENCE 31 (4th ed., 2002) (defining comparative negligence as "any system of law that by some method and in some situations apportions costs of an accident, at least in part, on the basis of the relative fault of the responsible parties"). 72 See, e.g., Gary T. Schwartz, Contributory and Comparative Negligence: A Reappraisal, 87 YALE L.J. 697 (1978) (fairness-centered analysis); Oren Bar-Gill & Omri Ben-Shahar, The Uneasy Case for Comparative Negligence, 5 AM. L. & ECON. REV. 433 (2003) (efficiency-centered analysis). 73 SCHWARTZ & ROWE, supra note 71, at 356 (noting that "[i]t is not easy to articulate precisely why a plaintiff was a particular percentage at fault in a particular accident"). 09 Given the legal ambiguity associated with the comparative negligence rule, judicial decision makers applying this rule might be influenced by an array of anchors. These anchors might be introduced by the litigants, who will try to structure the discussion around a number that serves their interests. Alternatively, and more important for present purposes, the law itself might introduce an anchor into the comparative negligence litigation. While many jurisdictions apply a pure comparative negligence regime that allows for any allocation of responsibility between the plaintiff and the defendant, others use some version of a 50% rule.74 In these latter jurisdictions, the doctrine is in play only if the plaintiff's share of responsibility is less than 50%.75 If, on the other hand, the plaintiff's share of fault is more than 50%, then his claim will be dismissed, and he will receive no compensation (even though he is not fully to blame).76 This legal structure presents an opportunity to test whether the salience of the 50% threshold under the 50% rule can alter people's judgments regarding comparative fault. Ninety law students at an American university participated in an experimental survey in return for a chance to win a $30 gift certificate. The survey used a betweensubjects design and included two conditions, to which the participants were randomly assigned: Anchor and No-Anchor. The participants in both conditions read a short vignette describing a hypothetical tort case brought by a pedestrian who was hit by a driver. The case was designed to reflect a relatively low fault on the part of the plaintiff. More specifically, the subjects were informed that prior to the accident, the plaintiff had left a bar where he had a couple of drinks and that he was emailing on his smartphone while walking and thus was not paying very close attention to where he was going. With respect to the defendant, on the other hand, the subjects were informed that he drove his car in reverse against the direction of traffic on a one-way street and was distracted by a sign on the side of the road. The subjects in the No-Anchor group were informed that "the jurisdiction has a comparative negligence rule that covers all tort cases including auto accidents— according to this rule the damages awarded to plaintiffs can be deducted so as to reflect their relative responsibility for the accident." Following this legal background, the subjects were asked "in your opinion, what percent of the responsibility best reflects the plaintiff's (i.e., pedestrian's) share of fault in this case?" The subjects in the Anchor group, on the other hand, were informed that "the jurisdiction has a fifty percent comparative negligence rule that governs all tort cases including auto accidents— according to this rule the damages awarded to plaintiffs can be deducted so as to reflect their relative responsibility for the accident, but if the accident is mostly the plaintiffs fault (more than 50%), then his claim is completely barred, and he receives no compensation." Following this legal background, the subjects were informed that the defendant's lawyer claims that the plaintiff is more than 50% responsible for the accident and were asked to decide whether they would accept this claim. After 74 Id. at 78-88. Id. 76 Id. 75 21 rendering their decision regarding this preliminary claim, the subjects were asked the exact same question as subjects in the No-Anchor condition: "in your opinion, what percent of the responsibility best reflects the plaintiff's (i.e., pedestrian's) share of fault in this case?" Our hypothesis was that the preliminary question would function as an anchor that would pull participants' assessment of the plaintiff's fault upward. The results showed that, as expected, the subjects attributed relatively low fault to the plaintiff. The mean level of responsibility determined by all subjects was 19.33%. Furthermore, the subjects in the Anchor group were almost unanimous in their rejection of the preliminary claim that attributed more than 50% of the fault to the plaintiff, and only two of them indicated that they would accept it. This result suggests that the anchor that we used was in fact irrelevant and that it thus should not have influenced peoples' decisions. While all subjects generally shared the view that the plaintiffs fault was relatively low, the precise level of fault ascribed to him was strongly influenced by the presence of the anchor. When the respondents were presented with no anchor, the mean level of fault assigned to the plaintiff was 15.20%. When the respondents were presented with a high anchor, the mean level of fault assigned to the plaintiff was 25.54%. A formal comparison of the conditions revealed that the difference was significant (t(90)=3.21, p<0.01).77 In sum, a meritless legal claim that anchored the legal analysis on a salient high number managed to almost double the level of fault that the respondents attributed to the plaintiff. 77 In the text we report the results for the full sample of subjects. The significant difference between the two groups persists if we drop from the sample the two subjects who accepted the preliminary claim and assigned relatively high fault levels to the plaintiff. More specifically in this case, the mean fault assessment of the anchor group drops to 24.09%, and the difference between the two conditions remains significant (t(88) = 2.851; p<0.01). 20 FIGURE 4: THE MEAN PERCENTAGE OF FAULT THAT RESPONDENTS ASSIGNED TO THE PLAINTIFF WITH AND WITHOUT AN ANCHOR 30.00% 25.54% Plaintiff's Fault 25.00% 20.00% 15.20% 15.00% 10.00% 5.00% 0.00% Without Anchor With Anchor Experiment 4: Setting the Standard with Actual Experience By using vignettes and randomly assigning participants to different conditions, the previous studies demonstrated the effect of anchors on the way in which people interpret ambiguous legal norms. While most studies on the effect of cognitive biases on judicial decision making routinely use this methodology, one should acknowledge two potential drawbacks associated with it. First, even with the richest and most detailed vignettes, subjects still have limited information on the case. Arguably, when the facts are somewhat scarce, the force of the anchor might be greater. Second, the decisions that are made by subjects in such studies bear no consequences, both for subjects and for third parties. People may thus be quick to rely on meaningless anchors when their choices do not influence anyone but may be reluctant to do so when their choices have ramifications. In order to overcome these limitations, our final study used a methodology in which participants actually experienced the regulated activity before they were asked to set a norm that would govern other people engaged in that activity. Additionally, the final experiment attempted to explore whether people can be debiased from the effects of an anchor. More specifically, we wanted to test whether the anchoring effect can be mitigated by holding people accountable for their decision. If the anchoring effect is driven by a quick and intuitive mode of thinking, then it may be attenuated by forcing people to slow down their decision-making process and to shift to a more systematic, reason-based mode of thinking.78 A way to achieve this goal is 78 For a more complete account of the existence of two modes of decisions (system one that is associated with quick and intuitive choices and system two that reflects a slower, more deliberative thought process) and their connection to cognitive biases see Kahneman, supra note 7, at 19-30. 22 by introducing accountability into the decision-making environment.79 Once people know that they will have to explain their decisions, they tend to engage "in an effortful and self-critical search for reasons to justify their actions,"80 which, in turn, will allow them to "gain greater awareness of their cognitive processes by regularly monitoring the cues that are allowed to influence judgment and choice."81 Hence, much like the way that accountability moderates the strength of the effect of other cognitive biases, anchoring might be reduced by requiring people to explain their decision ex ante.82 While accountability may play a significant role in judicial decision making, little research has examined the interaction between the two.83 The single experiment that tested the connection between accountability and anchoring in a legal context (i.e., damages) found that accountability did not diminish the effect of an irrelevant anchor.84 Given the centrality of the issue and the limited ability to draw broad conclusions from a single study with a null finding, we decided to add two conditions to this experiment in which we controlled for accountability. A total of 100 undergraduate students, 50 men and 50 women, at an Israeli university participated in the experiment. The participants were recruited through ads inviting them to participate in a “fun experiment.” The basic task involved a computerized trivia game that included easy and difficult questions. On average, the chance of answering a difficult question correctly was 30%, whereas that of answering an easy question was 60%.85 The participants were told that the purpose of the game was to reevaluate a pool of questions that were used in experiments in the past and that would be used in future experiments. We used a between-subject design that included four conditions in a 2-by-2 experimental design: High-Anchor and Low-Anchor groups, which were further divided into Accountability and No-Accountability subgroups. All the participants were told that they would be asked to answer 20 trivia questions and that for every correct answer, whether the question was easy or difficult, they would receive 1 NIS. In order to create an interpretative dilemma for the participants, the instructions informed them that the law governing experiments conducted with students prohibits us from mandating the number of easy and difficult questions that they must answer in the 79 For a review of the connection between accountability and biases see Jennifer S. Lerner & Philip E. Tetlock, Accounting for the Effects of Accountability, 125 PSYCHOL. BULL. 255 (1999). 80 Id. at 263. 81 Id. 82 Id. at 262-263. 83 Gregory Mitchell, Why Law and Economics' Perfect Rationality should not be Traded for Behavioral Law and Economics' Equal Incompetence, 91 GEO. L.J. 67, 110 (2002) (noting that accountability is "often neglected in experimental studies and in legal decision theorists' analyses of legal decisionmaking"). 84 See Guthrie, Rachlinski & Wistrich, supra note 29, at 1501-1506. 85 We used a pool of 60 pre-tested multiple choice general knowledge questions in topics such as sports, music, geography and science that were used in previous experiments over the past five years. See Yuval Feldman, Amos Schurr & Doron Teichman, Reference Points and Contractual Choices: An Experimental Examination, 10 J. EMPIRICAL L. STUD. 512 (2013). 23 experiment. Therefore, before each question, they were asked to choose whether they wanted the question to be easy or difficult. The instructions then stated that “participants are asked to answer a reasonable amount of difficult questions, so that we have a good mix of both types of questions.” Since the probability of answering a difficult question correctly was significantly lower, interpreting the “reasonable amount” obligation required the subjects to strike a balance between their self-interests and the benefit of others—much like with ambiguous legal norms. In order to make sure that all choices that were made by the subjects were informed, they experienced a trial run of ten questions—five of each type. The participants then received feedback regarding their success rate in each type of question and moved on to the “real game.” After the subjects completed the trivia game, we could turn to collect from them the dependent variable—their judgment regarding what a reasonable amount of difficult questions is. In order to do so, we told the subjects at this point that "[i]n the real experiment that we will hold in the future, which will be very similar to this experiment, we will subtract money from the payment made to subjects who answer an unreasonable amount of difficult questions (participants will of course be notified of this). The law that regulates experiments with students states that in order to do so we must obtain from other students a fair and neutral estimate of what is a 'reasonable amount.' In your opinion, how many difficult questions should we expect students to answer in the future study?" This question enabled us to document participants' judgment of what a "reasonable amount" actually is in a relatively robust way, as they were led to believe that their judgment would bind their peers.86 The anchoring manipulation was conducted by using the same framework that was used in Experiment 1A and was the same in all groups. The participants in the High-Anchor groups (with and without accountability) were given an example based on a high standard as to how their interpretation would operate and were told that "for example, if a reasonable amount will be deemed to mean 20 hard questions (all of the questions), then we will subtract from the payment to subjects who will answer less than 20 hard questions (e.g., 19 hard questions)." On the other hand, participants in the Low-Anchor groups (with and without accountability) were given an example based on a low standard as to how their interpretation would operate and were told that "for example, if a reasonable amount will be deemed to mean 1 hard question (a single question), then we will subtract from the payment to subjects who will answer less than one hard questions (e.g., 0 hard questions)." Much like the manipulation used in Experiment 1A, this manipulation has two main advantages. First, the anchor used has no normative value and was presented as a benign illustration of how a norm functions. Second, given their extreme nature, both the high anchor and the low anchor reflect 86 Note that the actual interpretative choice made by subjects as to the amount of hard questions they answered did not necessarily reflect their best judgment as to how a party ought to behave in this setting. In order to ascertain this value we employed the procedure described in the text, which separated between behavior and judgment. 24 what can arguably be considered erroneous interpretations of the term "reasonable amount" and therefore should have no bearing on the actual interpretative task. The accountability manipulation focused on requiring people to justify their decision. More specifically, the subjects in the No-Accountability conditions were simply asked to render their opinion as to what constitutes a reasonable amount of difficult questions, whereas the subjects in the Accountability conditions were required to provide a written explanation for their decision in a text box that was located right under the box in which they specified what they thought constituted a reasonable amount of hard questions. In this regard, we followed the design of previous psychologists conducting accountability studies.87 To sum up, Experiment 4 used a 2-by-2 experimental design. The participants in the High-Anchor-No-Accountability and Low-Anchor-No-Accountability groups were exposed to high or low anchors and rendered their decision without going through any debiasing treatment. The participants in the High-Anchor-Accountability and LowAnchor-Accountability groups were exposed to high or low anchors and were required to explain in writing the decision that they rendered. Figure 5 illustrates the participants’ responses in the different conditions. The left two bars present the results of the Low and High anchor groups in the NoAccountability conditions. As shown, when the respondents were presented with a low anchor (1 question), they set the mean threshold to 6.08 (stdev = 2.99) difficult questions, whereas when the respondents were presented with a high anchor (20 questions), they set the mean threshold to 11.04 (stdev = 4.62) difficult questions. The right two bars present the results of the Low and High anchor groups in the Accountability conditions. As shown, when the respondents were presented with a low anchor (1 question), they set the mean threshold to 7.04 (stdev = 3.96), whereas when the respondents were presented with a high anchor (20 questions), they set the mean threshold to 9.24 (stdev = 4.16). To test the effect of anchoring on each group, we entered the participants’ estimates in all groups into an ANOVA model with participants’ estimates as the dependent variable condition (Low Anchor vs. High Anchor) and accountability type (No Accountability vs. Accountability) as two between-subject variables. The analysis revealed a significant effect of condition (F(1,96) = 20.23, p < 0.001), indicating, as in the previous experiments, that the anchor had a significant effect on the participants’ estimates. In addition, the analysis yielded a marginally significant interaction between condition and accountability type (F(1,96) = 3.01, p <0.1). This finding suggests that the requirement to explain one’s decisions attenuated the effect of anchors on decisions. To better understand the origin of this effect, we next examined the effect of anchoring on the No Accountability and Accountability groups. 87 See Lerner & Tetlock, supra note 79 at 255-256 (reviewing the ways in which psychologists have operationalized accountability). 25 We begin by analyzing the results of the No Accountability groups. A comparison of the two thresholds in the Low and High anchor conditions reveals a significant difference (t(48) = 4.50, p<0.001). Further analysis shows that this effect is not the outcome of alternative theoretical explanations. First, the effect is not qualified by a difference in actual choices made by the participants in the first stage of the experiment. The participants in the Low Anchor condition chose to solve 7.96 difficult questions, on average, roughly the same number as the participants in the High Anchor condition (M = 8.16, t(48) < 1, p = n.s.). Second, the effect is not qualified by the rate of success in solving trivia questions. To test for the possibility that differences in ability moderate the anchoring effect, we entered the participants’ estimates to an ANCOVA model with participants’ estimates as the dependent variable, condition as a between-subjects variable, and the number of correctly solved questions in each stage of the experiment as covariates.88 The analysis revealed a significant effect of condition (F(1,44) = 17.18, p < 0.001) and a non-significant effect of the covariates (F(1,44) < 1 for all four covariates). This finding indicates that respondents’ level of expertise in solving trivia questions did not attenuate the anchoring effect. We turn next to examine the results of the Accountability groups. A comparison of the two thresholds reveals a marginally significant difference (t(48) = 1.92, p<0.1). This finding indicates that although the request to explain one’s decisions affected the participants’ estimates, they were also influenced by the anchor. Similarly to the results for the No-Accountability groups, further analysis shows that this effect is not the outcome of alternative theoretical explanations. First, the effect is not qualified by a difference in actual choices made by participants in the first stage of the experiment. The participants in the Low-Anchor-Accountability condition chose to solve 7.76 difficult questions, on average, roughly the same number as the participants in the HighAnchor-Accountability condition (M = 6.96, t(48) < 1, p = n.s). Second, the effect is not qualified by the rate of success in solving trivia questions. To test for the possibility that differences in ability moderate the anchoring effect, we entered the participants’ estimates to an ANCOVA model with participants’ estimates as the dependent variable, condition as a between-subject variable, and the number of correctly solved questions in each stage of the experiment as covariates. The analysis revealed a marginally significant effect of condition (F(1,44) = 2.93, p < 0.1) and a non-significant effect of the covariates (F(1,44) < 1 for all four covariates). This finding indicates that the respondents’ level of expertise in solving trivia questions did not moderate the anchoring effect. Taken together, the results of Experiment 4 indicate that the subjects in all the groups were influenced by the anchor that they were exposed to but that this influence was diminished when accountability was introduced. FIGURE 5: THE MEAN AMOUNT OF HARD QUESTIONS VIEWED AS REASONABLE 88 An “ANCOVA” is a procedure similar to an ANOVA (see supra note 58), yet it allows to control and account for the influence of a supplementary independent variables (i.e., covariates). 26 Mean Threshold of Difficult Questions 12 11.04 10 9.24 8 7.04 6.08 Low Anchor 6 High Anchor 4 2 0 No Accountability IV. Accountability DISCUSSION AND IMPLICATIONS After presenting our experimental findings, we now analyze them. We begin by reviewing the contribution of our study to the understanding of judicial decision making in general. Then, we explore some of the legal implications that might stem from our results. Finally, we address the limitations of the study and offer several avenues in which these limitations can be addressed in future research. 1. General Discussion In a series of experiments, we showed that anchoring might influence the way in which people interpret vague legal norms. This result was demonstrated in numerous legal settings and was evident within both expert and non-expert populations. Furthermore, the effect was shown to exist when participants expressed their judgments on a hypothetical scenario and when they made decisions with ramifications for others. These results extend the legal-anchoring literature toward a wide range of questions that have yet to be analyzed through this prism. Whereas the existing literature has demonstrated the impact of anchors on the final outcomes of adjudication (i.e., damages and sanctions), our findings suggest that anchors also influence substantive legal rules. When participants were called to determine what "reasonable behavior" means or to compare levels of fault—as jurors and judges so often are—their judgments were altered by meaningless anchors. While the experiments focused on a limited subset of legal situations, their results are potentially applicable to a much broader set of scenarios. Legal uncertainty is prolific and extends from the intricate doctrines of torts, contracts, and property to core constitutional frameworks. One can speculate (with the necessary caution) that the 27 effect documented in this Article might also exist in other settings that involve vague legal norms. Our studies have focused on irrelevant anchors. In this regard, the experiments reported here follow the traditional anchoring framework. This framework was used to isolate the anchor as the sole cause of decision change. Outside the lab, however, people might be swayed by anchors that bear some relevance to their decision. In such contexts, the anchoring effect might manifest itself by people giving the anchor too much weight (rather than by giving it any weight). Furthermore, given the relevance of these anchors to the decision at hand, relevant anchors might have a greater effect on peoples' decisions than purely arbitrary anchors.89 Similarly, the interpretation of vague legal norms outside the lab might be influenced by anchors that are not purely arbitrary. According to this line of thought, a point of reference that should influence, yet not determine, the interpretation might be afforded excessive weight by the adjudicator. We further elaborate on such potential anchors in the following subsection. Finally, the findings of our last experiment suggest that accountability might diminish the effect of anchoring. When participants were forced to give a written explanation for their choice, the gap between the High Anchor group and the Low Anchor group was reduced significantly. Apparently, requiring people to justify their decisions triggers a somewhat slower and more reasoned thought process. This finding has potential broad implications for the behavioral analysis of judicial decision making, as accountability is present in varying degrees in most judicial decisions. While accountability is clearly present in some cases (e.g., high-stake cases that are decided by judges), in other cases, it might play a smaller role (e.g., low-stake cases that are decided by a jury). A nuanced understanding of the topic requires much more research. As noted above, in the single previous study that examined the connection between accountability and anchoring, Guthrie, Rachlinski, and Wistrich were unable to identify an interaction between the two.90 While we cannot fully explain this finding, two characteristics of their design might elucidate the difference between their result and ours. First, Guthrie, Rachlinski, and Wistrich's experiment involved a series of cognitive tasks and vignettes, and the participants were asked to explain their answer only with respect to the anchoring vignette.91 It could be that the workload created by the series of tasks caused the accountability treatment to have a diminished effect. Our subjects, on the other hand, only undertook one task and thus might have been more susceptible to a treatment that was aimed at slowing down their decision-making process. Second, Guthrie, Rachlinski, and Wistrich justified the need for their subjects to write an explanation by telling them that a review of their decision is "very likely."92 89 Hans & Reyna, supra note 30, at 144 (noting that "there is a strong expectation that a meaningful anchor will be more persuasive than an arbitrary one"). But see Chapman & Bornstein, supra note 44, at 532-533 (presenting data suggesting that the relevance of the anchor does not influence its power). 90 See Guthrie, Rachlinski & Wistrich, supra note 29, at 1501-1506. 91 Id. at 1491-1520. 92 Id. at 1503. 28 It could be that in the hypothetical vignette setting that was used in that study, this statement was not taken with sufficient seriousness. 2. Implications In this subsection, we briefly highlight some of the policy implications of our results. We explore the lessons that parties can draw from them and analyze several concrete examples of legal doctrines that might lend themselves to an anchoring effect. Despite its normative tone, the claims made within this subsection are somewhat tentative given the novelty of our findings. In this regard, the main goal of the presented framework can be viewed as a roadmap for future research rather than crisp and clear recommendations. In order to organize our discussion of substantive legal anchors, we divide these anchors into three categories that are based on the process of their creation: anchors that are created during litigation, anchors that are created by the parties prior to litigation emerging, and anchors that are created by external forces. (a) Anchors Produced Ex Post During Litigation Since anchors influence the outcome of litigation, litigating parties are expected to try to create anchors that will help sway the adjudicator in their direction. Such anchors might come in the shape of additional evidence or even mere arguments presented by lawyers. If these anchors become a central point of reference in litigation, they might shift the legal decision maker toward a desirable direction from the perspective of the party presenting them. To the extent that litigating parties do in fact try to anchor legal discussions in such a manner, it might be justified to regulate their behavior and prevent them from doing so. Earlier findings regarding the possibility of altering judgments of punitive damages have served to justify limitations on the ability of lawyers to anchor legal discussion.93 Similarly, one could conceptualize limitations on the ability to present evidence and arguments that are likely to generate a significant anchoring effect. For example, courts routinely prevent the introduction of evidence related to existing, yet not widespread, practices that parties want to present in order to establish the required level of care.94 This well-established rule might reflect a concern that the introduction of such practices could be used by lawyers in order to bias the jury toward them, even if it is clear that such practices do not reflect a reasonable level of care. One should recognize the limits of this type of anchoring, however, and not overstate the risks associated with it. The adversarial process affords parties opportunities to try to overcome each other's litigation tactics. Professional lawyers 93 See, e.g., Hans & Reyna, supra note 30, at 144-145; Marti & Wissler, supra note 29, at 91. See Kenneth S. Abraham, Custom, Noncustomary Practice, and Negligence, 109 COLUM. L. REV. 1784, 1804 (2009) (noting that "[u]nless the practice of taking the precaution is sufficiently widespread to qualify as a custom, however, evidence of the incidence of the practice-regardless of how many actors follow the practice-is not admissible") (emphasis omitted). 94 29 might be able to identify attempts to anchor a legal debate and to derive ways to circumvent them. For example, in the context of damages, researchers have shown that low rebuttal damages figures that are introduced by the defense can counteract the influence of high figures that are presented by the plaintiff.95 In this regard, our findings might be viewed as an explanatory tool geared toward understanding behavior during litigation and not as a justification for regulation. (b) Anchors Produced by the Parties Ex Ante Thus far, the discussion has focused on the behavior of judicial decision makers and the litigating parties. Yet anchoring might also affect behavior in earlier stages of the legal process. If a party knows that the choice that she makes ex ante will serve as an anchor for later litigation, she might alter her behavior in anticipation of this effect. As the example in the opening paragraph of this Article suggests, a party to a nuisance dispute might be able to influence the outcome of the case through the initial level of nuisance that she creates. If the creator of the nuisance manages to anchor the discussion on a high initial nuisance level, the eventual outcome of litigation might be shifted in that direction. Yet this point might go well beyond tort law. Consider, for instance, the body of contract law that addresses fixing unfair contracts.96 This body of law allows courts to regulate the content of contracts in some situations. If, for example, a court finds a certain term in the contract to be unconscionable, it might strike it down.97 Furthermore, in such situations, the court is often required to write a new term into the contract in order to fill the gap created by its refusal to enforce the existing unfair term. In the context of unconscionability, for instance, the U.C.C. provides that the court may "limit the application of any unconscionable clause as to avoid any unconscionable result."98 When a court is confronted with the task of writing a substitute term into a contract, it often faces numerous options that lie on a continuum. To draw on an example presented by Ben-Shahar, if a law professor inserts into her implied contract with the class an unconscionable term—a 200-page reading assignment for the next day—then assuming that this adhesive provision will be struck down, it is still unclear how many pages the students are required to read for the next day.99 One can devise numerous numbers, which could all seem reasonable and fair, leaving wide discretion for the adjudicator of the case.100 95 See Marti & Wissler, supra note 29, at 95. See also Neil Vidmar, Experimental Simulations and Tort Reform: Avoidance, Error, and Overreaching in Sunstein et al.'s Punitive Damages, 53 EMORY L.J. 1359, 1389-1390 (2004) (criticizing anchoring studies that do not account for the adversarial nature of trials). 96 See Omri Ben-Shahar, Fixing Unfair Contracts, 63 STAN. L. REV. 869 (2011). 97 See U.C.C. § 2-302 (2002). 98 Id. 99 See Ben-Shahar, supra note 96 at 870. 100 A concrete example of a contract term that could lend itself to this type of analysis is a non-compete provision in a labor contract. Generally, such provisions prevent employees from accepting job offers from rival firms during a specified period of time of months or years. While jurisdictions differ as to how 31 If one views the legal doctrine as an exogenous constraint on contract suppliers, then the resulting conclusion is that suppliers of contracts will attempt to adopt provisions that lie just within the given constraint. However, our findings suggest that the legal constraint might be altered by the term that is chosen by the supplier. Since the initial term might serve as an anchor for the discussion surrounding the alternative term that the court will use in its place, suppliers might be better off by adopting extreme initial terms. While such terms are bound to raise the probability of court intervention, they could also pull the eventual outcome of litigation toward a more favorable direction from the perspective of suppliers. Interestingly, this point can perhaps be extended to other legal domains that are usually not analyzed through the lens of cognitive psychology, such as constitutional law. In numerous jurisdictions, much of the judicial review of legislation is based on a vague proportionality test.101 Without getting into the nuances of the doctrine, the gist of it is a balancing test between the protected rights of individuals, on one hand, and the interests served by limiting those rights, on the other.102 From our perspective, in some instances, the questions that are subject to a proportionality test are situated on a numeric continuum. For example, a court functioning within a proportionality framework might be called upon to decide how long a suspect may be detained by the police before he is arraigned.103 In such situations, the legislature enacting the initial statute enjoys a first-mover advantage and can anchor the constitutional discussion on its initial balance. In other words, just like a seller of a house might adopt an "ask-high" pricing strategy in order to sell his house for a greater price, a legislator might adopt an "ask-high" infringement strategy in order to reach an eventual, better outcome with the courts reviewing its acts. To be sure, the constitutional example does not mirror the contractual example analyzed above. While courts in the contractual setting are often required to write a new term into a contract in the shadow of the anchor (i.e., the term that they struck down), in the constitutional setting, they are usually only required to strike down statues that violate a constitutional norm. To the extent that the act of striking down a statute is done in isolation, it is unclear whether anchoring has an influence. That said, this claim should be qualified. First, judicial review under a proportionality test resembles to some degree a negotiation process in which the legislature makes consecutive "offers," and the court decides whether to accept them or not. Viewed from this perspective, the starting point of the negotiation could serve as an anchor. Second, in some cases, courts they treat such provisions, many jurisdictions examine the provisions under some type of a vague reasonableness standard and engage in contract modification if a provision appears too harsh. For a recent comprehensive review of the legal status and a discussion of the policy considerations associated with regulating such provisions see ORLY LOBEL, TALENT WANTS TO BE FREE 49-75 (2013). 101 See, e.g., AHARON BARAK, PROPORTIONALITY: CONSTITUNIONAL RIGHTS AND THEIR LIMITATIONS 175-210 (2012) (reviewing the role of the proportionality doctrine in different legal systems). 102 Id. at 129-169 (describing the way in which the proportionality doctrine functions). 103 See HCJ 6055/95 Zemach v. Minister of Defense, P.D. 53(5) 241 (Israeli Supreme Court examines the constitutionality of a law allowing the arrest of soldiers for 96 days without being arraigned under a proportionality test). 30 might render their opinion as to what constitutes a proportional statute. For instance, after the Israeli Supreme Court decided to strike down a statute allowing for the detention of soldiers for 96 hours without being arraigned, the Court went on to note that shortening that period to 48 hours would be sufficient to withstand a future constitutional challenge.104 Finally, our findings related to the ability of parties to anchor litigation (both ex ante and ex post) shed additional light on the ongoing legal debate regarding the desirable balance between rules and standards.105 Rules and standards can be depicted as two extremes in a one-dimensional space representing the degree of specificity of legal norms.106 A standard is an open-ended legal norm that allows the adjudicator to make fact-specific determinations.107 A rule, conversely, imposes stricter limits on the discretion of the adjudicator and its ability to interpret the rule in more than one way.108 Influential legal scholars have argued for and against the use of standards and rules. For example, Sunstein has argued that because legislatures cannot anticipate all circumstances, judges using a process of analogical casuistry are essential to ensuring that the proper result is achieved in each case.109 By contrast, Scalia has vigorously argued against conferring discretion on judges because it prevents the law from being predictable.110 More recently, scholars have used the tools of behavioral economics and cognitive psychology to further explore the topic. Guttel and Harel, for example, have argued that the distinct types of uncertainty generated by rules and standards should guide policy makers when they choose between the two.111 In several related articles, we have shown that motivated reasoning and the self-serving bias might guide people when they use vague standards.112 As a result, they tend to interpret those standards in a way that is aligned with their private interests and that might diverge from an objective interpretation of the standard. Our current findings add another layer to the behavioral analysis of rules and standards. In all of the studies reported in this Article, anchoring was shown to have a 104 Id. at 284. For some contributions to this vast body of work see, for example, Isaac Ehrlich & Richard A. Posner, An Economic Analysis of Legal Rulemaking, 3 J. LEGAL STUD. 257 (1974); Kaplow, supra note 45; Antonin Scalia, The Rule of Law as a Law of Rules, 56 U. CHI. L. REV. 1175 (1989); Pierre Schlag, Rules and Standards, 33 UCLA L. REV. 379, 381-383 (1985); Cass R. Sunstein, Problems with Rules, 83 CAL. L. REV. 953 (1995). 106 On the definitions of rules and standards see Schlag, id. at 381-383. 107 Id. 108 Id. 109 Sunstein, supra note 105 at 958. 110 Scalia, supra note 105 at 1179. 111 See Ehud Guttel & Alon Harel, Uncertainty Revisited: Legal Prediction and Legal Postdiction, 107 MICH. L. REV. 467, 479-486 (2008). 112 See, e.g., Yuval Feldman & Alon Harel, Social Norms, Self-Interest and Ambiguity of Legal Norms: An Experimental Analysis of the Rule Vs. Standard Dilemma, 4 REV. L. & ECON. 81, 105-106 (2008); Yuval Feldman & Doron Teichman, Are all Legal Probabilities Created Equal, 84 N.Y.U. L. REV. 980, 1014-1017 (2009); Feldman, Schcurr & Teichman, supra note 85, at 532-533. 105 32 relatively large effect on the way in which people interpret vague legal standards. Thus, our results raise additional concern with respect to the use of standards and indicate that the application of standards might be biased. If anchoring functions erratically, then the content of standards might be inherently unpredictable, as they are swayed in random directions by irrelevant anchors. If, on the other hand, various institutional and commercial actors are better positioned to strategically generate anchors that serve their interest, then the interpretation of standards might systematically benefit those actors in a subtle way. Either way, our findings point toward a relative advantage of rules over standards. (c) Anchors Produced by External Forces Finally, external forces such as the law itself might generate anchors that influence peoples' judgments. As Experiment 3 has shown, once the law elevates certain focal values and affords them special normative weight, those values might quickly turn into anchors. As a result, those values might have special influence on peoples' decisions. An illustration of this point can be found in laws that use safe harbors in order to mitigate the uncertainty associated with vague standards.113 Generally speaking, safe harbors function as a type of clear rule that operates in the backdrop of a broader standard.114 Adherence to the dictates of the safe harbor implies compliance with the standard per se, yet deviations from it might still be ruled to be fulfilling the demands of the standard.115 While the existing literature on safe harbors has acknowledged the possibility that the standard might converge with the safe harbor over time,116 it has failed to account for the more subtle possibility of an anchoring effect. To the extent that the safe harbor functions as an anchor, one will continue to observe two separate legal entities: the clear rule-like safe harbor and the vague standard. The content of this standard, however, is expected to shift over time toward the safe harbor. A concrete example of the way in which current legal scholarship on safe harbors has neglected to realize the role of anchors can be found in the context of the fair use doctrine.117 According to this doctrine, people are entitled to use copyrighted work if that use meets certain conditions.118 For instance, an artist who wishes to incorporate a sample of a few seconds taken from a copyrighted musical recording into her own original work might be able to do so if this use is deemed to be fair. By allowing 113 See, e.g., Peter P. Swire, Safe Harbors and a Proposal to Improve the Community Reinvestment Act, 79 VA. L. REV. 349, 371-378 (1993) (presenting a general model of the desired use of safe harbors); Andrew Stumpff Morrison, Case Law, Systematic Law, and a Very Modest Suggestion, STATUTE L. REV. (forthcoming) (reviewing the use of safe harbors and highlighting their advantages). 114 Swire, id. at 371-373. On the distinction between rules and standards see supra notes 105-112 and accompanying text. 115 Id. 116 Id. at 376-377. 117 17 U.S.C. § 107 (2012) (setting out the rules of the fair use doctrine). 118 For a detailed description of the doctrine and the conditions under which it operates see, for example, GRAEME B. DINWOODIE, METHODS AND PERSPECTIVES IN INTELLECTUAL PROPERTY 213-244 (2013). 33 such derivative work, copyright law can diminish the deadweight loss associated with granting monopoly power to copyright holders.119 Given the legal ambiguity associated with the fair use standard and its potential chilling effect, legal scholars such as Parchomovsky and Goldman have proposed that safe harbors be created to define certain types of activity as fair.120 According to the proposal, these safe harbors would reflect minimalist uses of copyrighted works, while other more elaborate uses would continue to be governed by the vague fair use standard.121 For instance, in the context of audio-visual works, they proposed a safe harbor that would "allow users to reproduce the lesser of ten percent or thirty seconds of any protected work."122 Parchomovsky and Goldman were careful to acknowledge the limitations of their proposal,123 yet their discussion takes a purely rational choice perspective. In the context of judicial decision making, they examined the possibility that their proposed safe harbors will become the governing norm over time and rejected this possibility given the minimalistic nature of their proposed safe harbors.124 Similarly, in the context of end-user decision making, Parchomovsky and Goldman focused on cases in which the planned use is slightly greater than that allowed by the safe harbor.125 While they conceded that their proposed safe harbors might exhort a "gravitational force" that would draw users toward them in such cases,126 they presumed that this force would simply cause users to comply with the safe harbor. For instance, they highlighted the case of a visual artist deciding to incorporate only a thirty-second rather than a thirtyfive-second clip into her new work of art given the certainty afforded by the shorter option.127 This analysis, though, ignores the more nuanced effect that safe harbors might have on the fair use doctrine. Our findings suggest that even if courts would not equate all cases with the safe harbors (as Parchomovsky and Goldman assumed), the safe harbors will still alter the fair use analysis that would be conducted in their shadow. Given the ability of safe harbors to function as an anchor, their gravitational force is expected to pull courts and end-users toward them when they determine what fair use means. This, in turn, would cause the fair use line to be drawn more restrictively (yet above the safe harbor). In other words, just as damages caps shifted the entire 119 See Gideon Parchomovsky & Kevin A. Goldman, Fair Use Harbors, 93 VA. L. REV. 1483, 14911497 (2007) (describing the way in which the fair use doctrine strikes a balance between the competing interests associated with copyright law). 120 Id. at 1510-1511. 121 Id. 122 Id. at 1514. 123 Id. at 1524-1532 (reviewing different objections to the thesis). 124 Id. at 1524-1528. 125 Id. at 1526. 126 Id. 127 Id. 34 distribution of awards upward,128 safe harbors might shift the entire distribution of permissible activities downward. To be sure, we do not aim to endorse a normative claim that safe harbors are inherently bad (nor do we claim that the proposal put forward by Parchomovsky and Goldman is necessarily undesirable). An analysis of each safe harbor requires a detailed account of the complex forces surrounding it. Our claim is a modest one that focuses only on the positive analysis of safe harbors: any evaluation of a proposed safe harbor is incomplete without taking into account the potential anchoring effect that it might create. Thus far our discussion has focused on legislated anchors, yet the statutes are not the only source for external anchors can influence adjudication. At times, a nonlegislated point-of-reference might hold special weight when courts are called to interpret a vague legal norm. Customs might serve as an illustrative example in this regard. According to a custom-centered framework, when courts are asked to examine the level of care that a defendant took in a tort case, they can rely on the customary precautions that are taken in the relevant context.129 For example, if the question at hand is whether a tugboat should carry a radio on board, the court might then examine the accepted safety measures in the industry.130 According to this framework, if a tortfeasor complied with the governing custom, such compliance is sufficient in order to serve as a defense in the face of a tort claim. The legal status of customs has generated a vigorous academic debate. At one side of this debate stand scholars who advocate for the use of compliance with customs as a defense in tort law in cases stemming from consensual arrangements (e.g., product liability cases).131 Such scholars have argued that existing customs are likely to reflect efficient practices given the market forces that drive their creation.132 At the other side of the debate lie those who view customs through a more skeptical set of lens.133 According to these scholars, while customs might certainly reflect an efficient standard in some cases, they might also reflect inefficient stagnation in others.134 Legal doctrine, for the most part, is aligned with the latter argument.135 That is, while a custom can be introduced into evidence, compliance with it is not conclusive 128 See supra notes 38-39 and accompanying text. For a recent review of the role of customs in tort law see generally Abraham, supra note 94 at 17881804. 130 Alluding to the famous case of The T.J. Hooper, 60 F.2d 737, 740 (2d Cir. 1932). 131 See, e.g., Richard A. Epstein, The Path to The T. J. Hooper: The Theory and History of Custom in the Law of Torts, 21 J. LEGAL STUD. 1 (1992). 132 Id. at 16-17. 133 See, e.g., Steven Hetcher, Creating Safe Social Norms in a Dangerous World, 73 S. CAL. L. REV. 1 (1999). 134 Id. at 28-64. 135 See RESTATEMENT (THIRD) OF TORTS: LIABILITY FOR PHYSICAL HARM § 13 (2009) (noting that “[e]vidence that the actor has complied with custom in adopting certain precautions may bear on whether there were further precautions available to the actor, whether these precautions were feasible, and whether the actor knew or should have known them. […] The actor’s proof of compliance with custom 129 35 evidence of non-negligent behavior. As one often-cited case stated, "[w]hat usually is done may be evidence of what ought to be done; but what ought to be done is fixed by a standard of reasonable prudence, whether it usually is complied with or not"136 Our findings suggest that the introduction of a custom into evidence might carry greater weight than has previously been recognized. Once the custom enters into the legal discussion, it becomes a "benchmark against which defendants' conduct is evaluated."137 As a result the custom might serve as an anchor that pulls decision makers toward it. To that extent, the reality of the legal doctrine governing customs might be closer (though, obviously not identical) to the model suggesting that customs should be viewed as a complete defense. While courts might (and do) reject the custom and create a more demanding legal standard of care, when doing so, they are cognitively constrained by the anchoring effect generated by the custom. A closing note on the scope of our claims in this subsection is called for. The discussion in this subsection until this point was purely descriptive. It highlighted the potential effect of anchors produced by external forces on judicial decision making, without taking a normative stand on the issue. In some cases, however, these anchors might bring about undesirable consequences. If the anchor is not supposed to influence decisions, then factfinders should be insulated from its effect in order to prevent biased decisions. This goal can be achieved in jury trials by regulating the information that reaches the jurors. Consider, for example, the 50% rule in the context of comparative negligence examined in Experiment 3.138 As the results of this experiment demonstrated, in lowfault cases, the 50% rule might cause jurors to elevate the level of fault that they attribute to the plaintiff. Arguably, the anchor generated by this rule is purely irrelevant—no proponent of this rule claims that its goal is to alter the evaluation of fault.139 Thus, our findings suggest that jurors should not be informed of the rule when they make decisions regarding comparative fault. By excluding the anchor from the decision-making environment policy makers can promote unbiased choices that will not be influenced by irrelevant anchors.140 does not, however, conclusively show that the actor was free of negligence in not adopting further precautions.”). See also KEETON ET AL., PROSSER AND KEETON ON THE LAW OF TORTS, 193-196 (1984) (reviewing the case law). 136 See Tex. & Pac. R.R. Co. v. Behymer, 189 U.S. 468, 470 (1903). 137 See Gideon Parchomovsky & Alex Stein, Torts and Innovation, 107 MICH. L. REV. 285, 286 (2008). 138 For a description of the rule see supra notes 74-76 and accompanying text. 139 See Eli K. Best & John J. Donohue, III, Jury Nullification in Modified Comparative Negligence Regimes, 79 U. CHI. L. REV. 945, 971-973 (2012) (examining the justifications for the doctrine). 140 Interestingly, many law review pages and judicial decisions have been dedicated to the question whether jurors should be aware of the 50% rule. Their focus, however, has been exclusively on highfault cases and the concern that informing the jury of the rule might cause the jury to lower its fault assessment to a level below the 50% threshold in order to supply the plaintiff with some compensation. For a review of the literature and case law on the matter see SCHWARTZ & ROWE, supra note 71 at 375380. For an empirical examination of this hypothesis see Best & Donohue, id. at 957-967. 36 On the other hand, one should also acknowledge that the anchoring effect created by external forces is not necessarily undesirable. Often such anchors carry normative weight, and pulling decision makers toward them might reflect a conscious policy choice. In other words, legislatures might want to use the anchoring capabilities of safe harbors and other types of guidelines to shift courts toward them in a subtle way that maintains judicial discretion. Viewed from this perspective, the anchors that are reviewed in this subsection can be regarded as a type of nudge,141 which influences choices but sustains the liberty of decision makers to deviate from them. 3. Limitations and Future Research In this final subsection, we acknowledge potential criticisms to this research. We outline the limitations of our results as they relate to the methodology that we used and the variables that we studied. Additionally, we suggest potential paths for future research that could help address these limitations and further our understanding of the role that anchors play in the operation of substantive legal rules. As is always the case with experimental studies, one should be cautious as to their external validity. Our studies were neatly designed to isolate the effect of anchors. While doing so, however, we might have distanced the study setting from the complex reality of litigation. This complexity, in turn, might diminish the role of anchors if, for example, numerous anchors are in play or enhance their role if, for example, anchors carry normative weight. Undoubtedly, predicting the precise way in which anchors will influence the content of vague legal norms in every conceivable context is not a goal that can be achieved in one article. The realization that behavior in the "real world" might differ from that shown in experiments suggests that more work should be done in this area. Just as the literature on anchoring and damages slowly evolved from theoretical conjecture to a wide body of empirical literature encompassing both experimental and observational studies,142 the literature on anchoring and substantive legal rules will also need to slowly evolve before clear conclusions can be drawn. This is the natural path through which the science of decision making progresses. However, the need for additional studies in this area should not be read as a call to ignore anchoring in the domain of substantive legal rules until a critical number of studies is amassed. The wide body of anchoring literature—both in general day-to-day contexts and in legal contexts—suggests that the assumption that the legal interpretative process is immune from an anchoring effect is not realistic. Thus, policy makers and legal researchers should be attuned to the phenomenon and consider it in their analysis with the necessary caveats and caution. The term “nudge” alludes to policy tools that push people towards certain choices in non-mandatory ways, allowing them to opt out if they so wish. See generally RICHARD H. THALER & CASS R. SUNSTEIN, NUDGE: IMPROVING DECISIONS ABOUT WEALTH, HEALTH, AND HAPPINESS (2008). 142 See supra notes 29-39 and accompanying text. 141 37 Having recognized this general limitation, we now highlight a few specific points regarding the design of the experiments reported in this Article. While none of these points undermine the general conclusion stemming from our results, they do provide opportunities for concrete future projects. The first point relates to the characteristics of the dependent variables in our experiments. The dependent variable in Experiments 1 and 1A was the legally required level of care. As described in greater detail above, subjects in these experiments shifted their judgments regarding the required level of care upward or downward as a result of the introduction of an anchor into their decision making environment. While this design reflects the way in which a vague legal standard operates, it does not capture the fact that judicial decision makers are required to make an additional dichotomous decision—was the defendant negligent or not? Arguably, this additional step might diminish the effect of the anchor, as it requires further deliberation and somewhat distances decision makers from the anchor. In addition, the deep normative tone associated with many of these decisions—stating publicly that the defendant behaved negligently or unfairly—might cause decision makers to engage in a more thoughtful cognitive process rather than rely on quick-and-easy heuristics.143 The second point relates to the characteristics of the independent variables in our experiments. While we attempted to use anchors that were truly irrelevant to the decision that our subjects were asked to make, constructing such anchors in a legal context is a tricky task if one wishes to sustain the ecological validity of the experiment. In the context of the anchor used in Experiment 2 (i.e., the dividend distribution request proposed by the controlling shareholder), it could be that this request was perceived by subjects as relevant to their decision even though the materials made it clear that it was driven by an irrelevant factor. Similarly, in the context of the anchor used in Experiment 3 (i.e., the claim that the defendant was more than 50% responsible for the accident), it could be that this claim focused subjects' attention on the behavior of the plaintiff and truly convinced them that his behavior is more blameworthy. While we certainly recognize these concerns, it does not seem probable that they can explain away our results. First, it is unlikely that these issues can explain the magnitude of the gaps that we observed in our experiments. The claim that the plaintiff was mostly responsible for the accident in Experiment 3 was articulated to the subjects as follows: "The defendant claims that the plaintiff was more than 50% responsible for the accident and therefore his claim should be dismissed." It is difficult to see how this 143 This conjecture is supported by research on the Lower-Bid-Bias (i.e., the finding that bidding committees tend to evaluate low priced bids in a favorable manner) see Omer Dekel & Amos Schurr, The Lower-Bid Bias in Public Procurement, available at http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2270773 (2013). A recent study indicates that members of bidding committees manage to overcome the bias when making dichotomous decisions such as qualifying bids. See Amos Schurr, Omer Dekel & Simone Moran, Qualifying Faults: The Roles of Professionalism and Accountability, (unpublished manuscript) (on file with authors) (2014). 38 benign statement that does not include any substantive argument could have almost doubled the level of fault attributed to the plaintiff. Second, this concern is addressed directly in Experiments 1, 1A, and 4, in which the anchor had no substantive meaning and could arguably be considered purely irrelevant. A final point that we would like to highlight relates to the extent of the implications of our findings. The focus of this Article—much like the focus of the entire body of legal literature on anchoring—was on judicial decision making. As our discussion suggests, however, the legal implications of anchoring may go well beyond this narrow domain. For example, fully accounting for different aspects of contract design might require that anchoring be incorporated into the analysis. A significant body of legal literature has examined default contract terms through the prism of behavioral economics.144 Interestingly, this body of work has mostly focused on the status quo bias and the reluctance of people to opt out of default terms even in the face of high stakes and low transaction costs. 145 A more nuanced question that this literature has yet to address is whether default terms can also function as anchors and can thus influence the decisions of those who actively choose to opt out of them.146 We leave such questions for future studies that will be able to examine them more carefully. V. CONCLUSION This paper explored how irrelevant anchors might influence the way in which courts interpret vague legal standards. Ultimately, combining the wealth of behavioral research reviewed in this paper with our own findings, we conclude that the interpretive process is prone to an anchoring effect. The meaning of vague legal terms, such as reasonableness, good faith, fairness, and fault, can be altered by the introduction of irrelevant anchors into the decision-making environment. Undoubtedly, incorporating the anchoring effect into a full model of judicial decision making is a challenging task. The precise size of the effect is illusive, and the way that it interacts with other forces may be difficult to decipher. Nonetheless, the alternative, namely, wrongfully assuming that the process of interpretation is immune to anchoring effects, seems difficult to defend. This Article, thus, does not aim to present a definitive picture of the relationship between anchors and vague legal standards. Rather, it aims to start a long and grueling 144 For early contributions to this body of work see, for example, Russell Korobkin, The Status Quo Bias and Contract Default Rules, 83 CORN. L. REV. 608, 637-647 (1998); Russell Korobkin, Inertia and Preference in Contract Negotiation: The Psychological Power of Default Rules and Form Terms, 51 VAND. L. REV. 1583 (1998). For a recent review see Cass R. Sunstein, Deciding by Default, U. PENN L. REV. 1, 11-24 (2013). 145 See, e.g., Sunstein id. at 5-6 (focusing on the stickiness of default rules). For a concrete example see Doron Teichman, Old Habits Are Hard to Change: A Case Study of Israeli Real-Estate Contracts, 44 LAW & SOC’Y REV. 299, 319-322 (2010) (presenting a case study of sticky contract terms in the real estate market). 146 Some scholars have, however, alluded to this possibility. See, e.g., Marcel Kahan & Michael Klausner, Path Dependence in Corporate Contracting: Increasing Returns, Herd Behavior and Cognitive Biases, 74 WASH. U. L. Q. 347, 363 (1996) (recognizing that standard contract terms may have an anchoring effect). 39 process geared toward elucidating the role anchors that might play in the courtroom and beyond. 41