The Gist of Juries: Testing a Model of Damage Award Decision Making

Abstract

Despite the importance of damage awards, juries are often at sea about the amounts that should be awarded, with widely differing awards for cases that seem comparable. We tested a new model of damage award decision making by systematically varying the size, context, and meaningfulness of numerical comparisons or anchors. As a result, we were able to elicit large differences in award amounts that replicated for 2 different cases. Although even arbitrary dollar amounts (unrelated to the cases) influenced the size of award judgments, the most consistent effects of numerical anchors were achieved when the amounts were meaningful in the sense that they conveyed the gist of numbers as small or large. Consistent with the model, the ordinal gist of the severity of plaintiff’s damages and defendant’s liability predicted damage awards, controlling for other factors such as motivation for the award-judgment task and perceived economic damages. Contrary to traditional dual-process approaches, numeracy and cognitive style (e.g., need for cognition and cognitive reflection) were not significant predictors of these numerical judgments, but they were associated with lower levels of variability once the gist of the judgments was taken into account. Implications for theory and policy are discussed.

The legal system regularly requires factfinders to translate their qualitative judgments into quantities such as fines, prison terms, and damage awards, but the relationship between qualitative and quantitative judgments in law is not well understood (Hans, Rachlinski, & Owens, 2011). In this article, we test a new model of damage-award decision making in which injuries and negligence are interpreted in a causal narrative, producing qualitative representations of damages as fuzzy categories (e.g., horrible injury) and potential award amounts as fuzzy categories (e.g., large amount of money). These two sets of fuzzy categories are brought into alignment after adjustments are made to accommodate salient numbers, such as attorneys’ ad damnum requests.

Even when jurors agree on injuries qualitatively, they often disagree on how to translate qualitative categories into quantities. At the heart of our model is the juror’s search for meaningful numbers as award judgments, meaningful in the sense that their magnitude is understood as appropriate in the context of that case. To illustrate, a horrible injury deserves a huge award, but “huge” is a relative concept that depends on the historical period, cultural context, and individuals involved in the case. Thus, a million dollars could be a relatively small damage award to a juror in Manhattan, but a relatively large award to a juror in Muncie. In our view, this fuzzy flexibility in interpreting numbers as small or large depending on content (what they apply to) and context is generally a strength of human cognition.

In the following text, we first describe the difficulties that people experience in making numerical judgments, focusing on problems identified in prior research on damage awards. We then summarize prior research supporting fuzzy-trace theory and other relevant theories, such as the story model, which undergird our model of damage award decisions (e.g., Pennington & Hastie, 1992; Reyna, 2012a). Fuzzy traces are mental representations of the essential meaning or gist of information. Finally, we introduce the model and predictions tested in our experiments, notably, that numerical anchors bias award judgments because representations of damages are fuzzy, but meaningful anchors—anchors that put the size of awards in perspective so that they can be understood as small or large—have a greater effect on numerical award judgments than meaningless anchors. We also test hypotheses that individual differences in analytical deliberative thinking (Type 2; Evans & Stanovich, 2013) such as numeracy, cognitive reflection, and need for cognition determine variability in awards (see also Greene’s, 1989, discussion of juror competence).

Challenges in Generating Numerical Judgments of Damages

People face substantial challenges in generating reliable quantitative judgments for a variety of reasons. Numerical judgments often do not map onto linear scales (Cantlon & Brannon, 2006). Many people suffer from low numeracy, and, what is more, numerical judgments are prone to systematic biases (Gilovich, Griffin, & Kahneman, 2002; Reyna, Nelson, Han, & Dieckmann, 2009).

Juries and judges face a particularly challenging task in determining money damages. In tort cases, the aim of money damages is to make a plaintiff “whole.” Winning plaintiffs are entitled to receive damages that compensate them for both the economic and noneconomic consequences of an injury (Abraham, 2012). However, deciding on damages is a profoundly ambiguous task because of limited guidance from the court about how to evaluate damages, because of uncertain projections of injury, and because many losses are intangible ones (Diamond, Rose, Murphy, & Meixner, 2011; Greene & Bornstein, 2000). Some of the most significant losses have no concrete market value or referent (Sherwin, Eisenberg, & Re, 2012). The loss of a spouse, the pain of a severe physical injury, the embarrassment of facial disfigurement, the inability to hug a child, and the lost opportunity to pursue a favorite sport are all intangibles that are difficult if not impossible to value (Mott, Hans, & Simpson, 2000). Although some research finds strong ordinal relationships between injury severity and award amounts (Eisenberg, Rachlinski, & Wells, 2002; Greene & Bornstein, 2003; Robbennolt, 2002), other projects suggest substantial problems in deciding on damage awards (Diamond et al., 2011; Kahneman, Schkade, & Sunstein, 1998; Sunstein, Kahneman, Schkade, & Ritov, 2002).

An additional complication is that dollar awards are in essence an unbounded scale (Hastie, 2011; Kahneman et al., 1998). The minimum is set at zero, but there is, at least theoretically, no maximum amount, which contributes to upside variability in numerical judgments. This variability contrasts with bounded scales, in which both the minimum and maximum levels are identified a priori. In mock jury research on punitive damages, Sunstein, Hastie, Payne, Schkade, and Viscusi (2002) found that mock jurors displayed a great deal of consensus in evaluating the merits of different cases on fixed response scales measuring the acceptability of the defendant’s behavior and desire to punish the defendant, but their dollar awards were very different. As study subjects translated their qualitative judgments about the defendant’s wrongdoing into dollars, people with similar wrongfulness judgments chose different dollar values. Likewise, Wissler, Hart, and Saks (1999) surveyed laypersons, judges, and lawyers, asking for their assessments of civil cases. They found greater variability in dollar award recommendations than in the underlying injury severity judgments.

In sum, fact finders deciding on money damages struggle with limited legal guidance, intangible injuries, uncertain projections, an unbounded scale for dollar awards, and low numeracy. How jurors and judges arrive at specific dollar awards is not well-understood. Individual differences in the use of dollar scales are sometimes considered nuisances to be controlled for, rather than judgments to be explained. For example, Kahneman et al. (1998) observe that the “unpredictability of raw dollar awards is produced primarily by large (and possibly meaningless) individual differences in the use of the dollar scale” (p. 67).

Theoretical Background for a New Approach to Damage Award Decision Making

Hans and Reyna (2011) have proposed a new model of juror damage award decision making, based on an established model of cognition called fuzzy-trace theory that has particular relevance for quantitative judgments (Reyna, 2012a; Reyna & Brainerd, 2011). It builds on the well-known story model developed by Pennington and Hastie (1988, 1992), incorporates new research insights on cognition and numeracy, and applies them to the concrete decision task of damage award determinations. The model aims to explain the process by which jurors move from the qualitative to the quantitative, from a judgment about injury to a decision about dollars.

Fuzzy-trace theory is a theory of memory and cognition that draws on insights from research on judgment and decision processes (e.g., Kahneman, 2011; Weber & Johnson, 2009). Numerous experiments have tested and confirmed the central claims of fuzzy-trace theory in other decision-making domains (e.g., Kühberger & Tanner, 2010; Reyna, Chick, Corbin, & Hsia, 2014; Reyna & Lloyd, 2006). According to fuzzy-trace theory, people encode two parallel types of mental representations of information (e.g., Singer & Remillard, 2008). One is a verbatim representation and the other is a gist representation. Verbatim representations are literal, veridical, and detailed. In contrast, gist representations are the essential meaning (or meanings) that a person derives from the information. Verbatim representations preserve the exact surface form of information, such as exact wording and precise numbers, compared to the bottom-line meaning captured in gist representations. Verbatim and gist representations are encoded, retrieved, and stored separately and independently.

Verbatim and gist mental representations are associated with different kinds of thinking, and people engage in both forms of thinking at roughly the same time. However, gist tends to be more dominant (Reyna, 2012a). So, for example, when presented with a probability describing the likelihood of a future event, an individual may encode the exact probability given, but the underlying meaning of the probability (e.g., coded for gist as low or high risk) is more likely to govern risk-related behavior (Peters, 2012; Reyna, 2012b; Zikmund-Fisher, 2013). Gist usually trumps verbatim representations in judgment and decision making and is used as a default whenever the task allows.

One implication of the model is that the gist or meaning of a number is inherently relative. It depends on both the content of the number as well as its context. Regarding content, note that a probability of .20 of imminent rain is low but the same .20 probability of an imminent heart attack is high (Reyna, 2013; Reyna & Lloyd, 2006). As for context, a risk of 12% is perceived as low relative to 20%, but the same 12% is perceived as high relative to 4% (e.g., Windschitl, Martin, & Flugstad, 2002).

Numbers that differentiate categories are especially significant because they span qualitatively different states. The gist of such numbers differs even when they are literally similar. All-or-none comparisons such as a difference between a 1% chance of cancer and a 0% chance are meaningful, as are differences such as above and below 50% or between 95% and 100% (e.g., Reyna, 2008). For example, people are willing to pay a great deal to reduce cancer risk from 1% to 0% because these differ categorically. When numbers cross category boundaries as in these examples, so that an event becomes possible rather than impossible, or becomes certain rather than uncertain, their qualitative gist is apparent. Specific numbers can also acquire meaning because of content (e.g., a company’s yearly profit) or context (e.g., average wages of a person from the 19th century vs. that of someone from the 21st century).

It is important to note that, according to fuzzy-trace theory, gist-based intuitive judgments are not presumed to constitute a primitive or intellectually inferior type of thinking. Traditionally, dual-process theories characterize Type 1 intuitive thinking as impulsive, evolutionarily old, or generally less advanced compared with Type 2 analytical deliberative thinking (e.g., Evans, 2008; Kahneman, 2011; Keren & Schul, 2009; but see Evans & Stanovich, 2013). Type 2 thinking has been assessed using measures of numeracy, need for cognition, and cognitive reflection (e.g., Frederick, 2005; Peters et al., 2006), although the disposition to engage in Type 2 thinking (e.g., need for cognition) is sometimes distinguished from the computational ability to do so (e.g., numeracy; see Stanovich, West, & Toplak, 2012).

In contrast, research on fuzzy-trace theory suggests that gist thinking characterizes advanced cognition. Adults are more likely to engage in gist thinking compared to children, and, in expertise-relevant domains, experts have been found to engage in gist thinking more than nonexperts (e.g., Reyna et al., 2014; Reyna & Lloyd, 2006). For example, in a study with expert cardiologists, general practitioners, and students, gist thinking increased with greater expertise: Cardiologists used fewer dimensions of information than generalists and students to make admission decisions for patients with chest pain, but they were more accurate according to clinical guidelines (Reyna & Lloyd, 2006). In these and other decisions, reliance on the gist of information has been associated with better decision making (e.g., Adam & Reyna, 2005; Fraenkel et al., 2012; Fukukura, Ferguson, & Fujita, 2013; Reyna & Mills, 2014; Wolfe et al., 2015).

However, decisions about damage awards involve placing a number on the gist of the damages, which poses special challenges. Dual-process theorists have argued that higher numeracy should improve decision making involving numbers (e.g., Peters et al., 2006). The literature on numeracy indicates that many people have difficulty appreciating the magnitude of numbers (Reyna et al., 2009). They lack “number sense” (Laski & Siegler, 2007) and, thus, will have difficulty recognizing and producing a reliable gist of low or high dollar amounts (Reyna & Brainerd, 2008). Lower numeracy would then be associated with greater variability in numerical estimates of damages.

Moreover, a gist model emphasizes the distinction between computation, which is how numeracy is usually measured, and meaningful interpretation of numbers (Liberali, Reyna, Furlan, Stein, & Pardo, 2012). Illustrating this distinction, in some studies, people higher in computational numeracy rate numerically inferior monetary gambles as more attractive than superior gambles, missing the gist of the gambles (Reyna et al., 2009). Further, literal numbers are not translated one-for-one into gist representations— they are interpreted in context as low or high. Thus, rote numeracy without understanding the gist of damages is unlikely to enhance award decision making.

How Jurors Decide on Damages: The Gist-Based Model

Figure 1 displays the Hans–Reyna (2011) model of juror damage award decision making. The model posits that jurors make categorical (damages are warranted or not) and ordinal (the damages deserved are low, medium, or high) gist determinations on the basis of the narrative of the case (Singer & Remillard, 2008) and then search for dollar award amounts that fit the gist judgments. For simplicity, the figure shows the stages as linear, as happening one after another, although there is likely to be significant interactivity among the stages.

Figure 1.

The adapted Hans–Reyna (2011) model of juror damage award decision making.

Throughout the course of the trial, jurors engage in mainly gist-based reasoning about the plaintiff’s injury and the defendant’s culpability. The case facts, the character of the parties, the context and nature of the case, jurors’ attitudes, views, and world knowledge all combine to lead a juror to arrive at a more or less coherent interpretation of events, or gist representation of the case, which would typically support a decision about whether the defendant is liable for the plaintiff’s injury. Jurors essentially follow Pennington and Hastie’s (1988;s (1992) story model, organizing the evidence around causal interpretation of a narrative to arrive at a determination of liability and resulting injury (see also Reyna, 2012a). We envision a similar process in the first stage of our model. For example, numbers that are relevant to the case would be easier to incorporate into a causal narrative than irrelevant numbers (Glöckner & Englich, 2015).

In the second stage, which occurs during the trial or deliberation, the juror will also reach a categorical conclusion about whether damages are warranted. This categorical judgment (yes or no on damages) is followed by a third stage in which jurors make ordinal judgments relating to the injury’s severity. Jurors code the specific injury the plaintiff has suffered in terms of severity as low or high and they consequently determine whether the damage award that is deserved is low or high (ordinal gist; Reyna & Brainerd, 2008). These expectations flow from research showing that phenomena are categorized (e.g., Harnad, 2005) and intuitions about ordinal distinctions between low and high (but not specific numbers) are generated in parallel with categorical gist (e.g., Reyna, 2008).

According to the theory, categorical and ordinal gist judgments are not based on adding up a list of facts, but rather on the interpretation of those facts, often as a coherent narrative, which supports a simple bottom-line intuition of yes or no to damages and, if yes, as low or high. If an injury cannot be ranked as either low or high, it will be relegated to a medium level of severity. In the final stage, the ordinal gist judgment about the injury’s severity (low, medium, or high) will be mapped onto a number that corresponds to the gist of the award judgment. The juror will identify a number that is, to him or her, low, medium, or high, to match the perceived severity of injury.

The model proposes that jurors rely on symbolic numbers from everyday life that already have meaning to them as low or high numbers. As examples, one dollar is low and a million dollars is high. The juror also retrieves numbers from the case, such as medical costs, lost wages, and attorney ad damnum requests, numbers that the attorneys have convincingly portrayed as having a meaningful interpretation in the context of the case. The juror then selects, and often adjusts, or constructs a number that matches the ordinal (low, medium, high) gist of the injury.

Predictions

This experiment tests key predictions of the Hans–Reyna model, as well as alternative dual-process approaches. To do so, we conducted a mock juror damage award decision-making study, varying the meaningfulness and dollar amount (low or high) of an anchor, and manipulating comparisons of the anchor amount as relatively low or high.

Presented numbers have been found to bias numerical judgments because people use them as a starting point and fail to adjust sufficiently, hence the term “anchor” (e.g., Chapman & Johnson, 1999). Following the substantial literature on anchoring, subjects given a high ($250,000) anchor should produce higher award amounts than subjects given a low ($40,000) anchor (see Furnham & Boo, 2011, for a review). Anchoring effects for jury awards are consistent with the Hans–Reyna model because severity of damages is assumed to be mentally represented as a vague or “fuzzy” gist rather than as a precise value, and, thus, numerical judgments are malleable (Reyna, 2008).

Similarly, the perceived size of awards should be affected by comparisons that make them seem relatively small or relatively large (e.g., $40,000 seems larger when it is described as “way above the national median annual income”). As we have discussed, damage awards are inherently uncertain and must be constructed. Research on relative judgment in perception and psychophysics, as well as decision-by-sampling, has shown that even arbitrary anchors and reference points shape numerical judgments (e.g., Pettibone & Wedell, 2000; Ungemach, Stewart, & Reimers, 2011). In other words, numerical reference points provide a context that helps people judge whether award amounts are low or high (as illustrated in our earlier example comparing 12% with either 4% or 20%; Windschitl et al., 2002). Therefore, when the anchor is portrayed as a relatively small amount of money, subjects should adjust upward, but when the anchor is portrayed as a relatively large amount of money, subjects should adjust downward.

Most important, anchors that allow the numbers involved in damage awards to be meaningfully interpreted, for example, in terms of median income, should be more influential than arbitrary numbers, for example, the cost of courtroom renovations. Meaningful numbers provide greater insight into whether an award amount is low or high, reducing uncertainty about estimates. The Hans–Reyna model predicts that meaningful numbers become influential as jurors attempt to match their conception of the severity of the damages to a potential award amount. Meaningful anchors promote extraction of the ordinal gist of the numbers (i.e., what a small or large award would be in the context of the case). Thus, jurors should more readily anchor onto meaningful numbers and use them as a benchmark for damage award amounts.

Hence, the meaningfulness of the anchor should interact with its amount. Meaningful anchors should make it easier to understand the gist of award amounts as low or high, in contrast to an anchor that is meaningless. In other words, there should be a greater difference between award amounts given low versus high anchors in the meaningful as opposed to meaningless anchor conditions. In addition, awards in the meaningful conditions should cluster more closely around the provided anchor amounts, whereas awards in the meaningless conditions should vary (because of uncertainty about the gist of the numbers).

Although the Hans–Reyna model does not make a strong prediction about the interaction between meaning and relative perception for award amounts, meaning and relative perception of the anchor might be expected to interact. For example, if meaningfulness reduces uncertainty about estimates, relative perception should have less of a biasing effect on award judgments in the meaningful compared to meaningless anchor conditions.

Despite large variability in awards, the Hans–Reyna model predicts ordinal coherence, namely, that, award amounts should vary systematically with respect to ordinal judgments of injury severity and liability (e.g., negligence). Given that the model assumes that ordinal gist representations determine award judgments, and that we focus on pain and suffering damages rather than complex economic damages, we do not expect that numeracy will have a large effect on the amount of awards. However, lower numeracy should be associated with greater variability in numerical judgments. To test traditional dual-process predictions that greater computational ability (numeracy), need for cognition, and/or cognitive reflection affect the amount or variability of judgments, we included a range of such measures of Type 2 processing.

Method

Subjects

A total of 173 subjects (64.7% women; M_Age = 20, SD = 1.5) participated in the experiment varying anchor amount, relative perception, and meaningfulness of the anchor between subjects, which was conducted as an online survey on Qualtrics. Subjects were recruited through social networking sites, or participated for course credit in psychology or human development courses at Cornell University. It was a racially diverse group: 53.2% of subjects identified themselves as White, 31.2% as Asian, 6.9% as mixed ethnicity, 5.2% as African American and 3.5% selected other. Furthermore, 13.9% of the subjects identified themselves as having Hispanic, Latino, or Spanish origin. All subjects indicated that they were 18 years of age or older and could communicate effectively in English, with 84.4% reporting English as their first language, two criteria for jury service. Most (88.4%) subjects were also American citizens and hence fulfilled another criterion for jury service in the jurisdiction. Five subjects had previously served on a jury.

To compare our results with those of a control sample, we collected data from an additional 93 subjects (73% female; M_age = 21.3, SD = 5.1) who were randomly assigned to read one of the two cases. The procedure for these subjects was identical to that of the previous experiment, except that they did not receive an anchor. These subjects were also recruited through social networking sites, or participated for course credit for psychology or human development courses at Cornell University. As for racial demographics, 54.8% of subjects identified as White, 27% as Asian, 9.7% as African American, 5.3% as mixed, and 3.2% selected other. Furthermore, 10.8% of the subjects identified as having Hispanic, Latino, or Spanish origin. All subjects indicated that they could communicate effectively in English, and 87.1% reported English as their first language (and 96.8% of the sample reported American citizenship). Three subjects had previously served on a jury.

Materials and Procedure

After reading instructions about the study, subjects then read one of two trial summaries (Jeansonne v. Landau or Monroe v. Rumson). Both cases were negligence cases involving a personal injury. Both cases had one plaintiff, a noncorporate defendant and the plaintiff did not bear any responsibility for the accident. The Jeansonne v. Landau case was previously used in Greene, Johns, and Smith (2001). The trial summary consisted of witness testimony from the plaintiff and the defense. Greene et al. (2001) described the case facts as follows:

The driver of a semitruck was forced to negotiate road conditions caused by a construction area. The construction area forced the lanes on a four lane highway to merge into two lanes, with the two remaining lanes diverted around the construction site. In this area of the highway, the defendant lost control of his truck and crashed through the guardrail. The semitruck crossed into the oncoming lanes of traffic and struck the plaintiff’s automobile. The plaintiff suffered injuries when the truck collided with his automobile. The plaintiff suffered a fractured parietal bone, receiving a concussion as well as soft tissue injury to the cervical spine, which caused recurrent back and neck pain. (pp. 229–230)

In Munroe v. Rumson, subjects read the case materials created by McAuliff and Bornstein (2010), which was adapted from Abbinante v. O’Connell (1996). The defendant was described as an inattentive driver who swerved to avoid a rear-end collision and instead accidentally struck an 18-year-old female pedestrian who was walking on the side of the road. The plaintiff suffered vertebrae injuries and spent two nights in intensive care. She suffered back pain as well as physical mobility problems, but fully recovered two years after the injury. The trial summary also referred to statements by the treating physician and physical therapist, and concluded with closing statements from the plaintiff and defense lawyers.

After reading the case facts, subjects read instructions from the judge. They were told that the defendant had been found liable for the plaintiff’s injuries and that economic damages had been paid to the plaintiff. The judge explained that their task was to arrive at a dollar figure to compensate the plaintiff for pain and suffering. The judge also explained that there are no definite mathematical formulas or instructions for determining pain and suffering awards. Jurors were asked to use their best judgment in coming up with an award amount. They were instructed not to consider punitive elements in their awards.

Finally, jurors read statements which they were told were made by the jury foreman at the start of deliberation. The jury foreman reiterated the task at hand, which was to determine a pain and suffering award. The experimental manipulations were incorporated into the jury foreman’s statement.

The experiment was a 2 (anchor size: large anchor, small anchor) × 2 (anchor meaning: anchor is meaningful, anchor is meaningless) × 2 (relative perception: anchor seems relatively small; anchor seems relatively large) × 2 (case: Jeansonne v. Landau, Munroe v. Rumson) between-subjects factorial design. To illustrate how these manipulations were embedded in the jury foreman’s statement, consider the statement in the low anchor, meaningful anchor, small perception condition:

Thanks to everyone for selecting me as your foreman. I just want to reinforce that we are here today to come up with a monetary award amount for the plaintiff’s pain and suffering. The defendant has already been found responsible for negligently causing the plaintiff’s injuries and the plaintiff has already been compensated for all economic damages. This means that all we need to consider is monetary compensation for the plaintiff’s pain and suffering. The judge also wanted me to remind everyone to be careful when leaving the room for any purpose. As you can see, the courtroom is under construction. If everyone is ready, let’s get started talking about how much money the plaintiff should be awarded. To begin the conversation, I was thinking $40,000 would be a good award because that is way above the national median annual income.

The $40,000 anchor is low compared with the alternate anchor of $250,000. It is meaningful in that median annual income adds meaningful context to the number $40,000. It is made to seem large in that it is placed in the context of median income, which is smaller than $40,000. The context of median income should therefore influence subjects’ perception of the “largeness” or “smallness” of the anchor amount.

Perception of $40,000 was made to seem small by indicating that it was below the average income; the $250,000 anchor was alternatively described as below lifetime and above annual income. Each relative perception comparison for meaningful anchors referred to income. In contrast, for meaningless anchors, subjects were told how much courtroom construction costs were ($40,000 or $250,000). Perception was manipulated by telling subjects that this was high or low relative to other construction projects in the state. A control condition, which included no anchor information, was presented to a separate group of subjects, described subsequently.

After subjects finished reading the jury foreman’s statement, subjects gave their award amount for pain and suffering. After being asked for their award amount, subjects also selected a level of pain and suffering award on an 11-point scale (1 = $0; 2 = $1–$12,000; 3 = $12,001–$25,000; 4 = $25,001–$50,000; 5 = $50,001–$100,000; 6 = $100,001–$250,000; 7 = $250,001–$500,000; 8 = $500,0001–$1 million, 9 = $1 million–$2 million; 10 = $2 million–$4 million; 11 = more than $4 million) that was used in Greene et al. (2001). Finally, subjects were asked to state the highest acceptable award, lowest acceptable award; and to describe the reasons for the award amount. The most often mentioned reasons were physical pain (63 mentions), missed experiences (61 mentions), and the permanence of injuries (58 mentions). Subjects’ reasoning was not analyzed in-depth in this paper, but collected to aid in future work.

Subjects responded to a number of survey questions. Subjects were asked about whether they agreed that the defendant was liable. The placement of this question was counterbalanced such that subjects either received it prior to making an award judgment, or after all the case-related survey questions.

The subjects rated the severity of the plaintiff’s injuries and the severity of the plaintiff’s pain and suffering on a scale, ranging from 1 (low) to 7 (high). They also rated their perceptions of the plaintiff and of the defendant on a scale, ranging from 1 (extremely negative) to 7 (extremely positive). In addition, subjects rated the extent to which the defendant was negligent and the extent to which the defendant contributed to the plaintiff’s injuries on a scale, ranging from 1 (not at all) to 11 (extremely). Subjects were further asked to estimate how difficult it was to decide on an award, how motivated they were while reading the case, and how much cognitive effort they exerted. These estimates were made on a scale, from 1 to 7 (1 = very little, 7 = great amount). They were asked if their award decision was affected by their desire to punish the defendant, on a scale from 1 to 7 (1 = not at all, 7 = a great deal), if they took economic damages (medical bills, lost wages, etc.) into account, and if so, the extent to which that influenced their decision, on a scale from 1 to 7 (1 = not at all, 7 = a great deal).

Subjects were given tasks which measure numeracy and thinking styles. Specifically, subjects completed the 15-item expanded numeracy scale, which is composed of a minor variation of the three-item general numeracy scale created by Schwartz, Woloshin, Black, and Welch (1997), with an additional eight items proposed by Lipkus, Samsa, and Rimer (2001), and four items added by Peters et al. (2007) that tests familiarity with simple arithmetical operations (e.g., multiplication), basic probability and related ratio concepts (e.g., fractions, decimals, proportions, percentages, and probability), and ability to keep track of class-inclusion relations (Liberali et al., 2012). They also completed the 8-item subjective numeracy scale (Fagerlin, Zikmund-Fisher, Ubel, Jankovic, Derry, & Smith, 2007), which assesses individuals’ beliefs about their mathematical skills, and the cognitive reflection test (Frederick, 2005), which tests the ability to inhibit intuitive responses in favor of deliberative responses to mathematical problems. Finally, subjects completed the 18-item need for cognition scale (Cacioppo, Petty, & Kao, 1984), which measures how much individuals enjoy engaging in effortful thinking.

Results

As has been found in previous research on jury awards, our raw award, lowest acceptable award, and highest acceptable award amounts were positively skewed. In line with the approach recommended by other investigators (see Eisenberg & Wells, 2006; Robbennolt & Studebaker, 1999), a log transformation was used on the three dollar award variables (although for the sake of completeness, means for transformed and untransformed variables are shown in Table 1). The instructions indicated that a prior jury already had established the defendant’s liability for damages. Therefore, in line with Wissler, Hart, and Saks (1999), eight respondents who gave an award of zero were dropped from the analyses, leaving 165 subjects.

Table 1.

Mean Award Amounts (Dollars) and Standard Errors for the Experimental Conditions and Control Group

Manipulation			Award variables
Meaning	Relative perception	Anchor size	Award	Lowest award	Highest award	Log award	Log low award	Log high award	Scaled award
Meaningful	Low (Give more than anchor)	Low	84,276.19	49,885.76	210,171.43	10.64	9.31	11.32	3.29
			36,609.89	23,232.53	100,192.27	0.24	0.59	0.25	0.33
		High	241,710.53	141,842.16	417,368.47	11.80	10.81	11.93	5.00
			38,801.59	25,111.46	74,773.88	0.37	0.67	0.71	0.44
	High (Give less than anchor)	Low	49,000.00	15,466.76	256,285.71	10.01	8.15	10.72	2.43
			17,457.02	4,727.50	131,156.62	0.26	0.66	0.36	0.31
		High	147,952.38	85,957.19	255,571.43	11.39	10.15	11.98	4.14
			345,13.32	20,803.14	48,627.03	0.27	0.64	0.28	0.30
Meaningless	Low (Give more than anchor)	Low	155,952.38	46,993.00	24,014,714.29	10.37	7.79	11.36	3.14
			63,211.15	24,386.15	23,799,406.54	0.42	0.84	0.58	0.50
		High	485,995.45	133,900.05	925,209.52	11.51	10.01	12.07	4.55
			152,189.88	45,598.83	307,179.62	0.49	0.65	0.51	0.53
	High (Give less than anchor)	Low	70,190.48	45,976.19	252,976.19	10.24	9.43	11.23	3.00
			19,266.87	15,416.07	73,218.85	0.34	0.38	0.40	0.40
		High	64,790.53	34,181.05	117,976.84	9.83	9.17	10.45	3.16
			17,509.11	8,890.85	30,271.10	0.54	0.51	0.53	0.40
Control group			318,373.33	180,157.81	947,072.22	10.67	9.56	11.59	3.61
			101,362.75	70,023.00	384,748.52	0.23	0.30	0.23	0.25

Note. Means are in bold, standard errors are in italics. The relative perception manipulation is meant to drive award amounts downwards/upwards by making the anchor seem like a large/small amount of money in comparison to the context surrounding the number. Therefore, Give more than anchor and Give less than anchor indicate the direction that the perception manipulation should drive award amounts.

First, we ran a 2 (anchor size: large anchor, small anchor) × 2 (anchor meaning: anchor is meaningful, anchor is meaningless) × 2 (relative perception: anchor seems relatively small; anchor seems relatively large) ×2 (Case: Jeansonne v. Landau, Munroe v. Rumson) between-subjects analysis of variance (ANOVA), with the logarithm of the award amount as our dependent measure. The analysis yielded a main effect of anchor size, with larger anchors yielding larger award amounts than small anchors (M_{Large Anchor}= 11.12, SE = .18; M_{Small Anchor} = 10.3, SE = .20), F(1, 149) = 9.38, p .003, ${\mathrm{\eta }}_{\mathrm{p}}^2=.059$. There was also a main effect of relative perception such that when the context surrounding the anchors made them seem like a small amount of money, subjects gave more (M = 11.04, SE = .19) and when anchors were made to seem like a lot of money, subjects gave less (M = 10.37, SE = .19), F(1, 149) = 6.25, p = .014, ${\mathrm{\eta }}_{\mathrm{p}}^2=.04$ as well as a marginal effect of anchor meaning, such that awards were larger for meaningful anchors compared to meaningless anchors (M_Meaningful = 10.96, SE = .19; M_Meaningless = 10.46, SE = .19), F(1, 149) = 3.57, p = .061, ${\mathrm{\eta }}_{\mathrm{p}}^2=.023$. Furthermore, anchor size interacted with anchor meaning, F(1, 149) = 4.08, p = .045, ${\mathrm{\eta }}_{\mathrm{p}}^2=.027$. As predicted, anchoring effects were larger for meaningful anchors (Δ = 1.36, 95% CI [−.2.11, −0.62]) compared with meaningless anchors (Δ = −.28, 95% CI [−1.04, 0.48]). There was a marginal three-way Anchor Size × Anchor Meaning × Relative Perception interaction in which anchoring effects were found in the meaningful anchor conditions regardless of the relative perception condition, but anchoring effects were only found in the meaningless anchor condition when the anchor amount was put in a context which made the anchor seem like a small amount of money, F(1, 149) = 3.36, p = .069, ${\mathrm{\eta }}_{\mathrm{p}}^2=.022$ (see Table 1). Awards were higher for the Jeansonne case compared with the Monroe case, F(1, 149) = 10.23, p = .002, ${\mathrm{\eta }}_{\mathrm{p}}^2=.064$, but case did not significantly interact with any other factors in the analysis.

When this analysis was run on only citizens of the United States (reducing the sample by 20 to 145), the overall pattern of effects remained the same. However the Anchor Size × Meaning interaction was reduced to marginal significance, F(1, 129) = 3.75, p = .055, ${\mathrm{\eta }}_{\mathrm{p}}^2=.028$, while the Anchor Size × Meaning × Relative Perception interaction crossed into traditional significance, F(1, 129) = 4.31, p = .04, ${\mathrm{\eta }}_{\mathrm{p}}^2=.032$.

To determine whether the above effects were consistent across award variables, we ran a repeated-measures ANOVA using the larger sample with award type as a within-subjects variable which consisted of the logarithm of the lowest acceptable award amount, the regular award amount, and the highest acceptable award amount. Again, the between-subjects variables consisted of a 2 (anchor size: large anchor, small anchor) × 2 (anchor meaning: anchor is meaningful, anchor is meaningless) × 2 (relative perception: anchor seems relatively small; anchor seems relatively large) × 2 (case: Jeansonne v. Landau, Munroe v. Rumson) factorial design. The repeated measures ANOVA included 164 subjects because one subject failed to enter a lowest or highest award amount so was excluded from this analysis. We report Greenhouse-Geisser corrected p values for within-subject analyses to account for a violation of sphericity (W = .34), χ² = 159.8, p <.01).

There was a main effect of anchor size, F(1, 148) = 8.69, p = .004, ${\mathrm{\eta }}_{\mathrm{p}}^2=.055$, such that large anchors elicited significantly larger awards. The Anchor Size × Anchor Meaning interaction missed significance, F(1, 148) = 3.27, p = .073, ${\mathrm{\eta }}_{\mathrm{p}}^2=.022$, such that anchoring effects were significant for meaningful anchors (Δ = 1.47, 95% CI [0.62, 2.32]), but nonsignificant for meaningless anchors (Δ = 0.35, 95% CI [−0.52, 1.23]). Finally, the Anchor Size × Anchor Meaning × Relative Perception interaction was significant, F(1, 148) = 4.14, p = .044, ${\mathrm{\eta }}_{\mathrm{p}}^2=.027$), demonstrating the same effect as the previous analysis (significant anchoring effects in all conditions except for when the anchor was meaningless and made to seem like a large amount of money).

As for within-subject effects, the analysis confirmed that subjects’ award amounts were smallest when asked to give their lowest amount (M = 9.33, SE = .23), largest when asked to give largest amounts (M = 11.37, SE = .17), and their regular award amounts fell in the middle (M = 10.71, SE = .14), F(1.2, 178.01) = 85.15, p <.001, ${\mathrm{\eta }}_{\mathrm{p}}^2=.365$. Award type did not, however, interact with the three-way Anchor Size × Anchor Meaning × Relative Perception interaction, and therefore, the effect demonstrated by this interaction was reliable across award variables.

As before, the pattern of results was unchanged when excluding noncitizens. The Anchor Size × Anchor Meaning interaction was significant rather than marginal when excluding noncitizens, F(1, 128) = 4.88, p = .029, ${\mathrm{\eta }}_{\mathrm{p}}^2=.037$, and the three-way interaction remained significant, F(1, 128) = 4.84, p = .03, ${\mathrm{\eta }}_{\mathrm{p}}^2=.036$.

Finally, to test our model with a bounded scale as well as with the unbounded dollar damage award judgment, we repeated the initial ANOVA with the scaled award as the dependent variable. This analysis yielded a main effect of anchor size, F(1, 149) = 18.68, p < .001, ${\mathrm{\eta }}_{\mathrm{p}}^2=.111$, relative perception, F(1, 149) = 6.94, p < .009, ${\mathrm{\eta }}_{\mathrm{p}}^2=.444$, and case, F(1, 149) = 13. 06, p <.001, ${\mathrm{\eta }}_{\mathrm{p}}^2=.081$, which were all in the same direction as previous results. The Anchor Size × Anchor Meaning interaction just missed statistical significance, F(1, 149) = 3.89, p = .05, ${\mathrm{\eta }}_{\mathrm{p}}^2=.025$). The Anchor Size × Anchor Meaning × Relative Perception interaction was not found to be statistically significant in the scaled award analysis, F(1, 149) = 1.8, p = .18 ${\mathrm{\eta }}_{\mathrm{p}}^2=.012$, although the overall pattern of results resembled that of previous analyses (see Table 1). Excluding noncitizens, the interaction between anchor size and anchor meaning was significant, F(1, 129) = 4.00, p = .048, ${\mathrm{\eta }}_{\mathrm{p}}^2=.013$, whereas the three-way Anchor Size × Anchor Meaning × Relative Perception interaction remained nonsignificant, F(1, 129) = 2.78, p = .098, ${\mathrm{\eta }}_{\mathrm{p}}^2=.021$.

Control Comparisons

As in previous analyses, we performed a log transformation on the award data due to skewing. We also excluded subjects who gave a zero award amount. Three subjects were excluded, leaving 90 subjects.

We performed a repeated-measures ANOVA with award type (logarithm of the lowest award, logarithm of the regular award, logarithm of the highest award) and case (Jeansonne v. Landau, Munroe v. Rumson) as between-subjects variables and found a significant effect of award type in which subjects’ lowest awards, regular awards, and highest awards all differed significantly (M_Lowest = 9.58, SE = .3, 95% CI [8.99, 10.18]; M_Regular = 10.68, SE = .23, 95% CI [10.23, 11.14]; M_Highest = 11.62, SE = .23, 95% CI [11.18, 12.07]), F(1.27, 111.3) = 81.38, p < .001, ${\mathrm{\eta }}_{\mathrm{p}}^2=.048$. Unlike the previous sample, there was no significant effect of case, F(1, 88) = .27, p = .61.

Next, we performed analyses comparing the mean of the regular award for the control sample to the means for the high and low anchor conditions in both meaningless and meaningful and high and low relative perception conditions. Awards in the experimental groups differed significantly from the control group’s when the anchors were meaningful, large, and perceived as small, t(107) = 2.15, p = .0344, d = .42 (see Table 1).

Only 3 subjects in this sample were noncitizens; removal of these subjects made no difference in significant effects. For example, the significant effect of Award Type remained when excluding noncitizens, F(2, 107.48) = 79.51, p < .001. Again, awards in the experiment differed significantly from the control group’s when the anchors were meaningful, large, and perceived as small; when noncitizens were excluded, t(100) = 2.26, p = .026, d = .45.

Variability in Award Amounts for Meaningful and Meaningless Anchors

The Hans–Reyna model predicts that awards will be more variable for meaningless anchors than for meaningful anchors. To determine whether award amounts were less variable in the meaningful condition compared to the meaningless condition, we followed the procedure described by Levene (1960), which was subsequently used in jury award research by McAuliff and Bornstein (2010), Robbennolt and Studebaker (1999), and Saks, Hollinger, Wissler, Evans, and Hart (1997). The method calculates the absolute distance between each award and the mean award in each experimental condition. For this analysis, we calculated absolute mean deviation (AMD) for the logarithm of the regular award given (for each condition), and performed a t test comparing the effect of anchor meaning on AMD. Figure 2 graphically displays the frequency of different dollar awards for meaningful and meaningless anchors and for the no anchor control study. As predicted, the results showed greater variability in awards for meaningless anchors (M = 1.71, SD = 1.20) than meaningful anchors (M = 1.09, SD = 0.88), t(163) = 3.82, p < .001, d = .60; this result remained significant when excluding noncitizens, t(143) = 4.07, p < .001, d = .68.

Figure 2.

Percentage of subjects giving award amounts in the meaningful and meaningless anchor conditions (split by high and low anchor) and control conditions.

Furthermore, we compared AMD for meaningful anchors with the control condition to whether the presence of meaningful anchors reduced variability as compared with subjects who received no anchors. Results showed that AMD was less for meaningful anchors as compared to the control condition (M = 1.78, SD = 1.21), t(170) = 4.27, p < .001, d = .65; when noncitizens were excluded, t(156) = 4.44, p <.001, d = .71. Finally, we compared AMD for meaningless anchors and the control group and found no significant differences between conditions, t(171) = 0.45, p = .65; similarly, when noncitizens were excluded, t(159) = 0.57, p = .57. Visual inspection of Figure 2 shows how the dollar values were distributed in the different conditions. The dollar award amounts in the meaningful anchor conditions showed more clustering, compared to the flatter and more variable distributions for the meaningless anchor and no-anchor control conditions.

Individual Differences in Cognitive Ability and Thinking Style

Next, we wanted to determine how subjects’ perceptions of the case, as well as individual differences in numeracy and cognitive ability, influenced award sizes. Two hundred and 38 subjects in the experimental and control groups completed individual difference measures. First, we performed bivariate correlations between all variables (see Table 2). Spearman’s rho was used because underlying scales are assumed to be ordinal.

Table 2.

Spearman Correlations Among Award Amounts or Deviations, Individual Differences, and Case Perceptions

Variable	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19
1. Award	—
2. Lowest award	.89**	—
3. Highest award	.95**	.80**	—
4. Scaled award	.96**	.87**	.92**	—
5. Absolute mean deviation	−.01	−.06	−.00	.02	—
6. Injury severity	.17*	.12*	.13*	.19**	.04	—
7. Pain and suffering	.25**	.24**	.23**	.29**	.07	.54**	—
8. Objective numeracy	−.07	−.10	−.03	−.08	−.11	.00	−.04	—
9. Subjective numeracy	−.02	−.09	.03	−.04	−.10	−.06	−.04	.32**	—
10. Cognitive reflection	−.03	−.08	.00	−.07	−.08	−.14*	−.25**	.32**	.28**	—
11. Need for cognition	.00	−.06	.04	−.04	−.13	.09	.06	.23**	.35**	.22**	—
12. Task difficulty	−.12	−.19**	−.07	−.14*	.00	.04	.06	.16*	.07	.06	.09	—
13. Task motivation	.03	.03	.05	−.00	.04	.25**	.28**	.00	.11	−.03	.17	−.05	—
14. Task cognitive effort	.05	.03	.01	.02	.01	.18**	.22**	−.05	.13	−.13	.14	.08	.57**	—
15. Plaintiff perception	.18**	.16*	.18**	.19**	.04	.21**	.33**	.06	.10	−.03	.02	.05	.21**	.16*	—
16. Defendant perception	−.07	−.12	−.08	−.10	−.00	.08	.00	.06	.10	.05	.10	.01	−.02	.01	−.09	—
17. Defendant negligence	.22**	.25**	.22**	.23**	.10	.10	.24**	−.01	−.02	−.08	.05	.05	.10	.02	.22**	−.37**	—
18. Defendant cause injuries	.13*	.12	.13	.12	.02	.35**	.34**	.09	.04	−.02	.11	.04	.19**	.14*	.30**	−.15*	.44**	—
19. Punishment as a factor	−.04	−.06	−.07	−.02	−.00	.09	.05	−.27**	−.15*	−.14*	−.06	−.09	.14*	.03	.01	.00	−.07	−.09	—
20. Economic damages as a factor	.16*	.18**	.14*	.14*	−.08	−.02	.00	−.14*	−.12	−.13	.00	.00	−0.10	−.04	.03	−.05	−.05	−.04	.27**

Note. Correlations include both experimental and control subjects. Injury severity and pain and suffering were rated on a scale, ranging from 1 (low) to 7 (high), task difficulty, motivation, and cognitive effort were rated from 1 (a small amount) to 7 (a great deal), perception of the plaintiff and defendant were rated from 1 (extremely negative) to 7 (extremely positive), the defendant’s negligence and degree to which the defendant caused the injuries were rated from 1 (not at all) to 11 (extremely), and the extent to which the desire to punish the defendant and to which economic damages were taken into account as a factor were rated from 1 (not at all) to 7 (a great deal).

^*p <.05.

^**p <.01.

The absence of any statistically significant relationships between award decisions and the numeracy measures is notable. Table 2 shows that scales measuring both objective numeracy and subjective numeracy were uncorrelated with the dollar value and scaled award decisions. The need for cognition scores were likewise unrelated to case judgments. Subjects higher in objective numeracy reported that the task was more difficult and were less likely to attribute their award judgments to economic damages or to a desire to punish the defendant. Subjects’ cognitive reflection test scores, measuring the subjects’ ability to inhibit their intuitive responses in favor of deliberative responses, were negatively correlated with their case judgments about the plaintiff’s injury and pain and suffering. Subjects who are able to inhibit intuitive responses rated the plaintiff’s injury and pain and suffering as less severe, although there was no relationship between cognitive reflection scores and award decisions. Similar results were obtained when correlations were computed for the control group by itself (see the Appendix).

Given that many of the survey items asked related questions (as seen in the correlations), we performed a factor analysis with varimax rotation to identify orthogonal dimensions using all 15 potential predictors of awards. The analysis produced five factors with eigenvalues above 1 (see Table 3), which explained a total of 57.98% of the variance. Factor 1 is plaintiff injuries/suffering, Factor 2 is numeracy/cognitive style, Factor 3 is motivation/effort, Factor 4 is defendant perception/negligence, and Factor 5 is economic damages/punishment. We then computed factor scores for each subject and performed a regression using each subjects’ score for each of the five factors to predict regular log award amounts (controlling for whether they were in the control group or experimental groups). The analysis revealed that subjects’ ratings of the pain and suffering of the plaintiff (higher scores indicating greater severity; Factor 1) and subjects’ perception of the defendant and his degree of negligence (higher scores indicating more negligence and a worse perception; Factor 4) significantly predicted higher award amounts (see Table 4). We also performed a regression using each subject’s scores for each of the five factors to predict AMD (again, controlling for whether they were in the control group or experimental groups; see Table 5). Only Factor 2 (numeracy/cognitive style) significantly predicted AMD (i.e., variability in award amounts).

Table 3.

Factor Loadings for Principal Components Analysis With Varimax Rotation

Item	Component
	1	2	3	4	5
How severe were the plaintiff’s injuries?	.81	− .07	.12	−.19	.03
How much pain and suffering did the plaintiff go through?	.80	−.14	.16	.04	.01
To what extent did the defendant cause the plaintiff’s injuries?	.59	.12	.07	.42	−.15
What is your perception of the plaintiff?	.48	.16	.15	.31	.07
Objective numeracy	.03	.69	−.10	.02	− .23
Subjective numeracy	−.06	.66	.22	−.05	−.12
Cognitive reflection task	−.24	.65	−.10	− .01	−.12
Need for cognition	.10	.62	.26	−.11	.13
How difficult was it to arrive at an award amount?	.22	.36	−.18	−.00	.12
How motivated were you while reading?	.18	.05	.85	.08	.02
How much cognitive effort did you exert?	.17	.03	.81	− .01	.00
What is your perception of the defendant?	.14	.10	− .02	−.80	− .08
To what extent was the defendant negligent?	.24	−.09	.01	.76	−.11
To what extent were economic damages a factor?	.03	− .02	−.17	.10	.82
To what extent was punishment a factor?	− .02	−.18	.22	−.15	.71

Note. Factor loadings >.4 are in boldface. Factor 1 is plaintiff injuries/suffering, Factor 2 is numeracy/cognitive style, Factor 3 is motivation/effort, Factor 4 is defendant perception/negligence, Factor 5 is economic damages/punishment.

Table 4.

Factors Predicting Regular Log Award for Experimental and Control Groups

Factor	B	SE	β	t	p
Constant	10.690	.212		50.541	.001
Factor 1 – plaintiff injuries/suffering	.438	.127	.222	3.443	.001
Factor 2 – numeracy/cognitive style	−.036	.130	−.018	−.274	.784
Factor 3 – motivation/effort	.045	.127	.023	.354	.723
Factor 4 – defendant perception/negligence	.330	.127	.167	2.599	.010
Factor 5 – economic damages/punishment	.062	.127	.031	.488	.626
Group (0 – control; 1 experimental)	.024	.270	.006	.089	.929
Adj. R2		.054
F		3.19				.005

Note. Regressions using log of lowest and highest awards yielded the same pattern of results.

Table 5.

Factors Predicting Absolute Mean Deviation for Experimental and Control Groups

Factor	B	SE	β	t	p
Constant	1.762	.124		14.237	.001
Factor 1 – plaintiff injuries/suffering	−.005	.074	−.004	−.062	.951
Factor 2 – numeracy/cognitive style	−.228	.076	−.197	−2.996	.003
Factor 3 – motivation/effort	.088	.074	.076	1.186	.237
Factor 4 – defendant perception/negligence	.082	.074	.071	1.104	.271
Factor 5 – economic damages/punishment	−.092	.074	−.079	−1.236	.218
Group (0 – control; 1 experimental)	−.311	.158	−.130	−1.965	.051
Adj. R2		.057
F		3.32				.004

Finally, to explore the extent to which the anchoring results for the experimental groups were linked to subjects’ ratings of plaintiffs’ pain and suffering or defendants’ negligence, we repeated the ANOVA involving the three award amounts (log of lowest, regular, and highest) which was performed previously, but controlled for subjects’ scores for plaintiff injury/suffering (Factor 1) and defendant perception/negligence factor (Factor 4). The Anchor Size × Anchor Meaning × Relative Perception interaction previously found remained statistically significant even when controlling for these factors, F(1, 126) = 8.05, p = .005, ${\mathrm{\eta }}_{\mathrm{p}}^2=.06$.

Discussion

In Determining Damages, Greene and Bornstein (2003) wrote, “Our understanding of how people translate an individual’s misfortune into a monetary value is rudimentary and elusive. … Understanding the psychological processes underlying these complex decisions is of significant import” (p. 173). Despite the importance of damage awards, juries are often at sea about the amounts that should be awarded, with widely differing awards for cases that seem comparable. Applying a new model of damage award decision making and testing predictions from the model, our aim in this work is to increase understanding of the challenging translation of qualitative representations of the gist of damages into dollars.

To test the model, we systematically varied the size, context, and meaningfulness of numerical comparisons or anchors. We were able to elicit large differences in award amounts by varying theoretically motivated predictors; effects were robust across the two cases. Each of the predicted factors had a significant impact on award amounts, and their effects were multiplicative as demonstrated by significant interactions. That is, we obtained effects on pain-and-suffering awards of arbitrarily low versus high anchors. However, as anticipated, providing meaningful reference points for award amounts, as opposed to only providing arbitrary anchors, had a larger and more consistent effect on judgments.

If people had verbatim representations of award amounts in mind, even inaccurate amounts, it would not be so easy to influence those amounts by orders of magnitude simply by mentioning the cost of courtroom renovations or the median income. Consistent with the Hans–Reyna model, people seem to have vague ordinal judgments of damages that they map onto vague ordinal judgments of dollar amounts (Hans & Reyna, 2011). The ordinal judgments of damages were not arbitrary: They correlated with severity of injuries, pain, and suffering, as shown with actual jury awards (Bovbjerg, Sloan, & Blumstein, 1989; Eisenberg et al., 2002; Vidmar & Hans, 2007). The ordinal judgments of damages are hypothesized to be mentally represented as qualitative gists, consistent with evidence from research on number judgments in many tasks (Peters, 2012; Reyna & Brainerd, 2008; Siegler, Thompson, & Schneider, 2011; Zikmund-Fisher, 2013).

As predicted, perceptions of the gist of dollar amounts were influenced by the content (low or high anchors) and context (low or high reference points that make numbers seem relatively small or large) of numerical comparisons (e.g., McAuliff & Bornstein, 2010; Pettibone & Wedell, 2000; Reyna, 2008; Ungemach et al., 2011). It is interesting to note that context had a greater biasing effect on awards when anchors were “meaningless” with respect to damages, such as the cost of courtroom renovations. Meaningful anchors, numbers that provided a sense of the magnitude of potential award amounts, such as median income, were less influenced by comparisons than meaningless anchors were.

The results also confirmed another prediction of the model, that jury awards would be more variable in the meaningless anchor condition because the gist of the award amount (as low or high) would be more transparent in the meaningful anchor condition. As predicted, subjects’ awards centered more tightly around the relevant anchor in the meaningful anchor condition (see Figure 2). Meaningful anchors communicated a sense of the magnitudes of potential awards, implying that amounts were low or high, so that people could more clearly understand the gist of the numbers.

In contrast, correlations examining the relationship of different numeracy scales to award amounts showed no association between subjective or objective numeracy and the mean level of recommended awards or the variability of awards. Indeed, higher numeracy was associated with the perception that the award task was harder rather than easier, contrary to what might be expected on the basis of traditional dual-process approaches. However, once other factors were controlled for in regression analyses, such as perceived difficulty of the task and severity of the injury, numeracy emerged as a significant predictor of variability in award judgments.

Our results support more recent conceptions of numeracy, need for cognition, and inhibition that distinguish such effects of individual differences from effects of other factors that affect judgment, such as mental representations of the gist of numbers (e.g., Liberali et al., 2012; Stanovich et al., 2012). Prior research has contrasted numeracy and rational deliberation with emotional and intuitive factors, such as outrage and cognitive biases (e.g., Hastie, 2011; Kahneman et al., 1998). Our theoretical approach builds on this prior research by acknowledging such factors. However, we emphasize a different kind of intuition, grounded in parallel processing of gist, alongside verbatim analysis of numbers, that is associated with insight and advanced cognition (e.g., Reyna, 2012a). The construction of a causal narrative from case facts (e.g., Pennington & Hastie, 1992) helps jurors derive a gist representation of damages and who is responsible for them. Using this theoretical approach, we attempt to explain both systematicity and variability in juror damage awards.

This sample of college undergraduates showed relatively high levels of numeracy, and so it would be worthwhile to examine the impact of numeracy with a broader sample and with a broader range of meaningful numbers. However, computational facility with numbers is not sufficient for good judgment using numbers, as demonstrated, for example, when people higher in computational numeracy rate numerically inferior monetary gambles as more attractive than superior gambles (Reyna et al., 2009). These results underline the importance of studying not just computational numeracy, which is likely to influence variability in award judgments, but also number sense and meaningful interpretation of the gist of numbers, which is likely to influence understanding whether awards are small or large (Liberali et al., 2012). Hence, future research should separate two kinds of numeracy that are both relevant to jury damage awards: knowing the numerical difference between 100,000 dollars and a million dollars from understanding whether a million dollars is a small or large number in the context of the plaintiff’s injury and the defendant’s liability.

Consistent with prior research on jury decision making and with the Hans–Reyna model, perceptions of pain and suffering (severity of injury) and of the negligence of the defendant (liability) predicted the amount of damage awards (Eisenberg, Hans, & Wells, 2007; Greene et al., 2001; Wissler, Evans, Hart, Morry, & Saks, 1997). This finding is inconsistent with a legal approach that assumes that jurors reliably separate liability and damages decisions (Wissler, Rector, & Saks, 2001). Liability was not entirely fused with pain and suffering—regressions showed that each contributed unique variance to damage awards. Future research should examine whether jurors’ causal narratives integrate perceived pain and suffering with negligence to produce a holistic impression of damages.

Pain and suffering along with negligence were significant predictors of damage awards even when alternative explanatory factors such as perceived economic damages and motivation to perform the award task were controlled for (see Table 4). Therefore, although these decisions were hypothetical, they seemed to elicit some degree of sympathy for victims and a sense of the defendant’s wrongdoing, and these reactions drove damage awards in the ways they do with actual juries (Hans, 2014).

Although our theory-testing experiment has demonstrated the success of the overall approach, we recognize the obvious limitations of conducting research on damage awards by using relatively brief written materials and employing a predominantly college student sample (Penrod, Kovera, & Groscup, 2011). Pain and suffering judgments were made in isolation, without giving subjects an opportunity to determine economic damages, as real juries do. That was necessary in this first test of the theory. However, because these judgments can interact (Robbennolt & Hans, in press), as we continue our research, it will be useful to explore pain and suffering assessments alongside related decisions. In future work, we plan to expand the case and trial materials and use a broader range of study subjects. We also intend to incorporate other ways of varying meaningfulness and the symbolic value of numbers (Zelizer, 1994).

Even at this early stage of development, the Hans–Reyna model of jury decisions about damage awards has some promising implications for policy. The American jury system rests on the assumption that ordinary people can make valid and reliable judgments about such matters as damage awards. Some prior literature on damages challenges this assumption because it suggests that damage awards, especially pain and suffering and punitive damages, are unpredictable and arbitrary. On the basis of this literature, some have called for policy reforms that minimize the role of the jury when numerical judgments and technical knowledge are required (Sunstein et al., 2002). However, the Hans–Reyna model suggests that there is an underlying coherence in people’s judgments about the severity of plaintiffs’ damages and liability of defendants; at the level of the gist, award judgments make sense. Therefore, if these results hold up in further tests, their implication is that jury awards are not unpredictable and arbitrary, countering arguments to minimize the role of the jury.

Specific difficulties arise, according to the model and consistent with our results, when people try to attach a number to their gist of the case. As demonstrated by our results, people need meaningful reference points to judge the gist of award magnitudes. The content and context of numbers is not necessarily about the merits of the case; the merits determine whether injuries are perceived as small or large. The model distinguishes mental representations of the gist of the case from the gist of the numbers, and focuses on aligning both representations. The importance of meaningfulness helps to explain an apparent disconnect between the incoherence found in experimental work on punitive damages (Sunstein et al., 2002) and the coherence of actual jury punitive damage awards in which the worst offenders are punished the most severely (Eisenberg et al., 2002). Our results also suggest that attorneys can influence how the gist of numbers is perceived by providing meaningful anchors and reference points. Indeed, there is evidence that they have done so in jury trials (Hans & Reyna, 2011).

In sum, this work introduces a new type of factor that can be studied in jury decision making, namely, the perception of the meaningful gist of damage awards. The model predicts that mental representations of the gist of damages will be ordinally consistent but numerically variable. These results suggest that jurors are not responding randomly without regard for the facts of the cases, and they are not necessarily confused or incompetent with respect to the level of damages they think is warranted. Rather, their judgments of the gist of damages take content and context into account, shifting with time and place as would be appropriate in the real world. The difficulty arises in translating the gist of damages into the precise numbers needed in the law. Future research should examine additional factors that influence the gist of damages and of awards, and how meaningful judgments can be facilitated in real-world jury decision making.

Appendix

Spearman Correlations Among Award Amounts or Deviations, Individual Differences, and Case Perceptions for Control Group

Variable	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19
1. Award	—
2. Lowest award	.92**	—
3. Highest award	.92**	.83**	—
4. Scaled award	.96**	.89**	.89**	—
5. Absolute mean deviation	.04	.07	.04	.08	—
6. Injury severity	.25*	.20	.19	.25*	−.06	—
7. Pain and suffering	.25*	.27*	.17	.28**	.04	.64**	—
8. Objective numeracy	−.06	−.02	.04	−.07	−.19	.04	−.09	—
9. Subjective numeracy	−.05	−.04	.01	−.08	−.08	−.07	−.12	.36**	—
10. Cognitive reflection	−.12	−.12	−.01	−.13	−.14	−.21	− .33**	.38**	.34**	—
11. Need for cognition	.02	−.02	.06	−.00	−.14	.17	.01	.24**	.32**	.22**	—
12. Task difficulty	−.27	−.27*	−.22	−.25*	−.11	.12	.01	.28*	.20	.04	.24*	—
13. Task motivation	.14	.12	.19	−.11	.07	.34**	.30**	−.15	.03	−.25	.10	−.11	—
14. Task cognitive effort	.25*	.23*	.17	.20	.10	.28**	.31**	−.18	.04	−.31	.06	.03	.50**	—
15. Plaintiff perception	.13	.13	.09	.14	.10	.37**	.43**	−.09	−.03	− .24*	.13	.10	.20	.23	—
16. Defendant perception	−.02	−.06	−.01	−.08	−.04	−.05	−.04	.06	.19	.13	−.07	.07	−.19	−.16	−.09	—
17. Defendant negligence	.11	.14	.10	.13	.17	.06	.16	−.00	.01	−.16	.15	.02	.11	.13	.16	−.26*	—
18. Defendant cause injuries	.06	.05	.05	.07	.09	.43**	.18	.13	.14	−.02	.25*	.12	−.02	.13	.11	−.07	.40**	—
19. Punishment as a factor	−.22*	−.23*	− .26*	−.17	.09	.18	.16	−.32**	− .24*	−.33**	−.12	−.01	.17	.11	.18	−.14	−.05	−.05	—
20. Economic damages as a factor	−.04	−.00	−.00	−.10	.05	−.08	.09	.08	−.20	−.25*	− .25*	−.07	−.03	−.05	.04	−.01	−.06	−.06	.30**

Note. Injury severity and pain and suffering were rated on a scale, ranging from 1 (low) to 7 (high), task difficulty, motivation, and cognitive effort were rated from 1 (a small amount) to 7 (a great deal), perception of the plaintiff and defendant were rated from 1 (extremely negative) to 7 (extremely positive), the defendant’s negligence and degree to which the defendant caused the injuries were rated from 1 (not at all) to 11 (extremely), and the extent to which the desire to punish the defendant and to which economic damages were taken into account as a factor were rated from 1 (not at all) to 7 (a great deal). Sample size was 90 for all correlations, except defendant cause injuries (n = 89), task difficulty (n = 88), and economic damages as a factor (n = 89).

^*p < .05.

^**p < .01.