The Sixth Amendment of the U.S. Constitution guarantees every citizen the right to a trial by an impartial jury in criminal matters, and the Seventh Amendment extends the right to a jury to civil matters. But what is meant by the term “impartial jury”? What can one expect from a group of individuals entrusted with making the utmost important decisions regarding individual liberty, criminal guilt, and, at times, even life or death for fellow members of their society? The American legal system presumes that jurors can put individual biases aside and render a fair and impartial verdict based on the facts alone.
However, how realistic is this tabula rasa notion of the jury? Are jurors actually capable of overcoming bias, life influences, and personal experiences that may improperly influence their decision-making? Or are legal skeptics justified in claiming that “[people] do not, once selected as a juror, suddenly become the objective, fair-minded individuals the courts ask them to be. They bring to the jury their own backgrounds and prejudices and their own processes for making decisions” (Horowitz, 1998, p.332)?
The ultimate task of a juror is to encode information, process it, deduce implications from it, reason as to the information's function, purpose and validity in the grand scheme of the case, and then communicate these results effectively with other jurors. From start to finish, the juror's task is primarily a cognitive one, and as Kuhn, Weinstrock, and Flaton (1994, 289) note, the job of the modern American jury is “one of the most cognitively complex tasks that ordinary citizens are routinely asked to perform.” As such, how individual jurors process, think and reason is of great interest in learning about how juries reach decisions. Jurors are not expected to consider extralegal factors (i.e. race, gender, age, class, ethnicity, religious persuasion, physical attractiveness of the defendant) in their decision-making.1
In all circumstances, the law presumes jurors will process relevant information rationally and without extraneous non-evidentiary influences. But do they? And, if not, does their style of cognitive processing help to predict their behavior?
A growing body of social science literature calls into question the notion of the tabula rasa juror and whether jurors actually ignore extralegal influences. Studies have documented the influence of a host of extralegal factors on juror decisions, mainly focusing on characteristics of the defendant. Relevant external extralegal factors have included the defendant's race (Sargent & Bradfield, 2004; Sommers & Ellsworth, 2001), age (Bergeron & McKelvie, 2004; Mueller-Johnson, Toglia, Sweeney, & Ceci, 2007; Warling & Badali-Peterson, 2001), religious conviction (Johnson, 1985), gender (DeSantis & Kayson, 1997; Fisher, 1997; McCoy & Gray, 2007), physical attractiveness (Stewart, 1985), occupation (Loeffler & Lawson, 2002), and ethnicity (Perez, Hosch, Ponder, & Trejo, 1993).
The Attraction Leniency Bias
One of the most compelling bodies of evidence showing that jurors are influenced by extralegal defendant characteristics is that of the defendant's physical characteristics, confirming the ancient Greek philosopher Sappho's adage that “what is beautiful is good.” In 1982, Monahan and Loftus concluded that research assessing the effect of a defendant's physical attractiveness on jurors' decision-making was among the most consistent effects noted in the literature at the time. And as early as 1974, Efran showed that, in the criminal context, physically attractive defendants were evaluated with less guilt certainty and less severe punishment. In the context of civil trials, Kulka and Kessler (1978) found that attractive plaintiffs were awarded higher compensatory damages. Stewart (1980, 1985) reported evidence of an attraction-leniency bias in which attractive criminal offenders were given less punitive sentences by jurors. And Stewart (1985) found a linear relationship between attractiveness and sentencing: the higher the defendant's attractiveness, the less severe his sentence.
The consistency of these effects within the literature has also withstood the test of time. Erian, Lin, Patel, Neal, & Geiselman (1998) provided jurors with a mock rape scenario in which attractiveness of the defendant varied. Attractiveness was significantly related to sentencing. Darby & Jeffers (1998) found that attractive defendants were convicted less often, punished less severely, rated happy, likeable, trustworthy, and less responsible for charges being brought against them than were less attractive defendants. Further, these effects have not only been found to operate within jurors but also within judges (Downs & Lyons 1991; Zebrowistz & McDonald, 1991). Mazzella and Feingold's (1994) meta-analysis concluded that, on average, it is advantageous for defendants to be physically attractive.
Overall, then, the attraction-leniency bias has been shown to operate across decision makers (judges and jurors), types of crime, and trial type (criminal and civil), albeit it is moderated by the seriousness of the crime (Beckham & Spray & Pietz, 2007) and the strength of evidence against the defendant (Erian et al., 1998).2
In view of evidence that jurors are, at least occasionally, influenced by extralegal factors and at times give improper weight to such factors in their decision-making, one might ask whether certain personality types might be more susceptible to extralegal influences of the type described above.
Individual Differences Applied to Juror Decision-making
Substantial research has shown how individual differences in personality, processing style, and cognitive ability may affect juror performance. The best known concerns the authoritarian personality type, who tend to be highly punitive (Kassin & Wrightsman, 1988) and more conviction-prone (Bray & Noble, 1978; De La Fuente Solana, Garcia, & Martin Tamayo, 1998), and those who “believe in a just world” and feel that individuals get what they deserve (Lerner, 1980).
Cognitive Experiential Self Theory (CEST)
There is one socio-cognitive theory of individual differences in decision-making that has received little attention in the juror context. Cognitive Experiential Self Theory (CEST) is the outcome of a line of work undertaken by Epstein and his colleges (e.g. Epstein, 2003; Donovan & Epstein, 1997, Epstein, Pacini, Denes-Raj, & Heier, 1996; Denes-Raj & Epstein, 1994). CEST is a broad theory of personality that supposes people process information through two independent but parallel cognitive systems, one a pre-conscious experiential/intuitive system and the other a conscious rational/analytical system (Epstein, 2003).3 Experiential processing is an effortless system that is propelled by what feels good. It emphasizes “vibes,” emotions, and stereotypical thinking from past events (Epstein, 1991). The experiential system is also associated with solving problems that rely on life lessons and experience that elude articulation and logical analysis (Epstein, 1991) and is particularly prone to logical shortcuts or heuristics (Denes-Raj & Heier, 1996). In his early work, Epstein demonstrated that the experiential system is credited with the ability to form interpersonal relationships, creative thoughts and emotionality (Epstein, 1991). It has also been associated with beliefs in the paranormal (Genovese, 2005). The rational system, on the other hand, is intentional and effortful. It is logic-driven; it is active, conscious, and requires justification via evidence (Epstein, 1991). The rational system is the source for scientific and technological achievements, and it makes possible complex generalization, comprehension of cause and effect, and high levels of abstraction and complex reasoning (Epstein, 1991).
In a series of experiments, Epstein and his associates have provided empirical support for CEST's core assumptions and construct validation (Denes Raj & Epstein 1994; Epstein, 2003; Epstein & Donovan, 1997; Epstein et al., 1996). For example, Denes Raj & Epstein (1994) found that when participants were offered an opportunity to win $1 by drawing a red bean from a dish with various colored beans, participants occasionally neglected the dishes with better winning ratios of red beans and chose instead from bowls with larger absolute numbers of red beans, even though the odds were against them. These results were consistent with CEST's dual-processing assumption, indicating that, at times, participants' experiential systems override their rational systems. Subjects claimed that although they knew (rationally) the probabilities in the dishes with more absolute red beans were against them, they felt (experientially) they had a better chance when there were more red beans in the dish.
According to CEST (Epstein, 2003), the experiential and rational systems operate in parallel and are interactive. However, the experiential system operates much faster and is thought to be unconscious. Thus, a sudden flash of anger that makes an employee want to argue with the boss may be nullified by the slower-acting rational system that brings to awareness the costs of doing so. Epstein (2003) showed that when people were asked to list the first three thoughts that came to mind in provocative situations, the first thought reported by most subjects, regardless of their overall proclivity on the rational-to-experiential spectrum, was often in the experiential system (e.g., desire to express anger). By the third thought, however, there was usually a corrective reaction that was in the rational system (e.g., a realization that the costs of expressing the anger could be counterproductive). Epstein interpreted these results to indicate that initial spontaneous impulsive responses were representative of the experiential system and were often later overridden by more constructive responses through the rational system.
Additionally, Epstein and Donovan (1997) employed CEST theory to analyze the famous Linda conjunction problem, adding further support for the theory. The Linda conjunction problem asks participants to read a scenario involving the fictitious character Linda and then to reason statistically. Linda is described as a bright, 31-year-old, single woman who majored in philosophy in college and was concerned about social justice. Participants are asked to rank order the probability that Linda is a feminist, a bank teller, and a feminist bank teller. Although the joint occurrence of two events occurring cannot be more likely than the probability of either of the separate events occurring (i.e., the “conjunction rule”), participants consistently rank the conjunction of bank teller and feminist as more probable than bank teller alone, therefore engaging in a conjunction error. Epstein reported that despite the fact that many people were able to identify the correct rule for solving Linda-type problems in general, many elected not to apply it to the Linda problem specifically. This was interpreted to mean that although participants “know better” they often prefer to behave according to the demands of the experiential system. Thus, in certain situations it is presumed the experiential mode can override processing in the rational mode, even when the latter is equally available for processing.
CEST, JUROR DECISION-MAKING, AND PRESENT STUDY
Given that the experiential-rational processing distinction represents an explanatory framework that has been construct-validated, it is reasonable to ask what effect these processing differences have (if any) on mock juror decision-making.
Lieberman and his colleagues (Krauss, Lieberman & Olson, 2004; Lieberman, 2002; Lieberman, Krauss, Kyger & Lehoux, 2007) have explored the relationship between CEST and juridical decisions in several studies. In one study, participants were motivated to think either experientially or rationally about a civil case involving an accident. In the rational condition, non-vivid language and evidence was presented in an emotionally neutral manner, while in the experiential condition, vivid words were used and mock jurors were exposed to evidence with graphic depictions. The experientially motivated jurors gave higher awards to plaintiffs when the defendant wasn't attractive than when they were – with a mean difference between conditions of about $500,000. Lieberman showed that, when motivated to think experientially, jurors were more prone to incorporate extralegal information (attractiveness) into their decision-making. Similarly, Krauss et al. (2004) have shown that jurors in a rational mode weigh mathematical and actuarial expert testimony more heavily, while experientially motivated jurors were more influenced by clinical testimony. Recently, Lieberman and his associates have replicated these findings (Lieberman et al., 2007).
To date, the Lieberman and Krauss studies have been alone in applying CEST to juror decision-making. And while these studies yielded interesting results, they were primarily concerned with whether jurors, once manipulated to process in an experiential mode, would exhibit predictable reasoning trends. Their work is not concerned with potential jurors' natural processing tendencies – that is, they are predicated on behaviors that accompany manipulations in processing styles of their mock jurors as opposed to the processing styles they naturally prefer. In actual trials, jurors are exposed to conflicting manipulations of evidence by attorneys for both sides, so having purely experientially or purely rationally motivated juries in the real world is unlikely, even unrealistic, because although one side may attempt to draw jurors' attention to emotional aspects of the evidence, the other side may try to persuade them to consider the evidence in a logical, unemotional way. Jurors bring to this contest their own natural processing proclivities, which ultimately influence how they resolve such conflicting presentations. The present study sought to determine whether jurors who were naturally influenced by their experiential system were more susceptible to extralegal biases than their rationally influenced counterparts, and it sought to do this in the context of a criminal trial.
ASSESSING NATURAL PROCESSING STYLE
A key tenet of CEST is that the rational and experiential systems are not mutually exclusive, but rather coexist in an almost symbiotic cognitive cohabitation, where behavior is the result of the interaction between the two systems (Epstein, 2003). CEST theory assumes that each individual has the potential to engage in either analytic or experiential processing and does so depending on a variety of factors, including the customary way of responding to a particular problem. (For instance, a math problem will generally be approached in a rational mode, and interpersonal problems in an experiential mode.) Moreover, the extent of analytic or experiential processing is a function of the level of emotional involvement (situations rich in emotional personal involvement are often overtaken by the experiential system), and the amount of relevant experience in similar situations or with similar problems (the more familiar a situation, the more likely the experiential system will intuitively handle it as it has always been handled before) (Epstein, 2003). Additionally, a crucial factor that plays a role in the degree to which one relies on one system over the other is individual preference (Epstein, 2003; Epstein et. al. 1996). Although the two systems are adaptive, independent, and yet interactive, people differ in the degree to which they tend to rely on either one in the course of their decision-making due to their personal preference (Epstein, 2003). Assessing individual processing trends across various stimuli and over time has revealed individuals' natural processing style (Epstein, 2003; Epstein et. al. 1996). Therefore, to classify individuals as primarily an experiential or primarily a rational processor is supported by both theory and empirical findings.
To assess quantitatively the degree to which experientiality and rationality differ within individuals, the Rational Experiential Inventory (REI) was developed (Epstein et al., 1996; Pacini & Epstein, 1999). The REI is a 40-item self-report measure of individual differences in intuitive-experiential processing, and analytical-rational thinking. The inventory is based on a coupling of the Need for Cognition scale (Cacioppo & Petty, 1982) used to measure rational processing and the Faith in Intuition scale (Epstein, et al., 1996) used to measure experiential processing. Because the REI uses two separate scales to account for the orthogonal two-dimensional structure of CEST, the inventory provides a separate score for experiential and rational processing for each individual. The REI has been shown to have high total scale reliabilities (Epstein & Pacini, 1999), and has been construct-validated using such measures as the Basic Belief's Inventory (Catlin & Epstein, 1992), the Big 5 (Goldberg, 1990), the Emotional Expressivity Scale (Kring, Smith, & Neale, 1994), the Categorical Thinking scale (Epstein & Meier, 1989), the Ego Strength and Defensiveness scale (Epstein, 1983), and the Conservative Ideology and Attitudes towards Criminals scale (Pacini & Epstein, 1999). Epstein et al. (1996) found that the underlying scale used to measure experiential processing, the so-called “faith in intuition scale” (FI) produced strong correlations with heuristic responses. Participants with high FI scores responded through the use of heuristics more so than other participants, and were more likely to regard heuristic responses as logical. The results of these studies support CEST's assumptions of both rationality, and experientiality (Pacini & Epstein, 1999).
To recap, we sought to determine whether processors influenced primarily by the experiential system (E-processors) are more susceptible to extralegal biases (such as defendant attractiveness) than are those influenced by the rational system (R-processors). This expectation follows from the oft-reported finding that the experiential system is generally associated with a reliance on intuitive hunches, emotional feelings, heuristics and personal schema, and generally poorer reasoning (Marks, Hine, Blore & Phillips 2007), including extralegal biases associated with the presentation of evidence (Lieberman, 2002; Krauss et al., 2004). Thus, it was hypothesized that E-processors would be more prone to pay attention to defendant attractiveness in a criminal trial than would R-processors. And because E-processors are by their nature more likely to be compelled by intuitive or gut feelings, we anticipated that they would be more likely to report that legally irrelevant information would affect their guilt determination than would R-processors because such information appeals to their intuitive and emotional sense.
Support for these hypotheses would be reflected if: (a) experientially dominant processors exhibit a greater disparity between their guilt assessment and sentencing of attractive defendants compared with less attractive defendants – conversely, rationally dominant processors should sentence and convict less attractive and attractive defendants similarly; (b) experientially dominant processors report a higher incident of influence with regard to whether various extralegal factors would affect their verdicts: and (c) experientially dominant processors should be more likely to exhibit a stereotype bias by indicating that less attractive defendants are more likely to be the “type of person” who could commit a crime.
The study sample consisted of 169 undergraduates who were recruited from psychology courses at Cornell University (111 female, 58 male), and participated in the study in exchange for extra credit. Seventy-one per cent were Caucasian, with the other groups being Hispanic (15%), Black (4%), Asian-American (6%), Arab (3%), and other (1%). Ninety-one per cent of the sample were between the ages of 19 and 22. The sample size satisfied an estimate of the statistical power required to detect a medium effect size with more than four predictors (Cohen, 1992).
Following the informed consent notification, participants were instructed to navigate through five phases of the study in sequential order: (1) the defendant's status report, (2) the case summary, (3) closing arguments summary, (4) jury instructions, and (5) two questionnaires, one dealing with guilt and sentencing feelings, and the final one to assess processing style.
All study materials were presented to participants via an interactive website which was individually accessed by participants.4 Participants were initially directed to an informed consent page where they were asked to enter the last three digits of their student ID number. The combination of odd and even numbers was used to randomly sort participants into one of eight groups (e.g., even-odd-odd = group #1; odd-even-odd = group #2, etc.). Participants sorted into groups one through four were assigned to the “attractive defendant condition,” with participants in groups five through eight assigned to the “less attractive defendant condition.” Eight photos (four for the attractive defendant category and four for the less attractive defendant category) were used in an effort to reduce unmeasured effects due to additional characteristics that a defendant might possess other than attractiveness (e.g., degree of baby-face-ness; Berry & Zebrowitz, 1988).
Defendant's Status Report
Participants were first presented with a defendant profile that provided personal information about the defendant, including name, racial identification, gender, hair and eye color, height, weight, date of birth and various descriptive information. Most prominently, the profile included a high-resolution color photograph of the defendant. Photographs were taken from the Michigan Offender Tracking Information System (OTIS), an online public database that provides internet users with information about a wide variety of offenders who are, or were, under the supervision of the Michigan Department of Corrections.5
Prior to conducting the study, a sample of 18 offenders from the OTIS database were selected for attractiveness ratings, nine who generally seemed to fit an attractive condition and nine who generally seemed to fit a less attractive condition. To control for race and age, all of the photos chosen were of white males within a 5-year age range. All photographs were then systematically pre-tested for attractiveness. After removing one photo from the original set of 18 because he was smiling (to control for facial expression effects; see Abel & Watters, 2005) and two additional photos due to unexpected pre-test results, the four photos with the highest average attractiveness ratings were assigned to the attractive condition and the four photos with the lowest average ratings were assigned to the less attractive condition. The four photos with the highest attractiveness ratings had a mean of 8.2 and the four with the lowest attractiveness ratings had a mean of 6.2. All defendant profiles were identical across participants; the only feature that varied was the photograph that was displayed with each profile.
Following the presentation of the profile, participants were presented with the trial summary from an aggravated assault case. The summary provided an outline of the charges, the evidence that had been levied against the defendant by the state, and the evidence supporting the defendant's self-defense claim. The text of an appellate case was used to prepare this summary, because it was brief enough to keep participant interest, yet provided enough factual information to create a sense of reality.
The case was selected from The Supreme Court of Ohio Decision and Announcement Search.6 The names of all involved parties were changed, and the case text was shortened to four single-spaced pages but modified to increase the level of evidentiary ambiguity. The prosecution alleged that the defendant attacked his ex-girlfriend with a hammer causing serious bodily harm by using undue force. The case against the defendant included (1) testimony from his ex-girlfriend alleging that the defendant attacked her, (2) blood analysis on a hammer found at the scene which matched both the defendant and the victim, and (3) testimony that the defendant fled town for a month after the attack. The defendant claimed self-defense and alleged that his ex-girlfriend accosted him with a knife when he went to get some tools he had left at her house. In support of his version of the facts, (1) two of his family members testified to the existence of stab wounds on the defendant's arm, (2) a knife was found at the crime scene and (3) the presence of his blood on the hammer was alleged to be the result of the knife wound.
To ensure that the defendants' image remained fresh in the participants' minds, after the trial summary stage, a trial report was provided that highlighted the main arguments from both the prosecution and the defense but also re-displayed prominently the color photograph of the defendant. The trial report contained a recap of the best five arguments for each side of the case. They were labeled as “key arguments from the prosecution” and “key arguments from the defense.” The prosecution argued: (1) there was no physical evidence presented that the defendant was stabbed; (2) the only witnesses that testified on the defendant's behalf were family members; (3) the defendant's motive was revenge, after he expressed he wanted to rekindle their romance but she did not; (4) the force used was too excessive to be merely warding someone off; (5) the defendant fled town for 1 month after the attack and fled the scene without seeking help. The defense argued that: (1) the defendant was viciously attacked by his ex-girlfriend, who was brandishing a knife and ultimately stabbed him, he feared for his life and was acting in self-defense; (2) two eyewitnesses testified that the defendant did in fact have stab wounds on the day of the attack; (3) a knife was found at the scene of the crime several feet from the defendant's ex-girlfriend; (4) the defendant's ex-girlfriend was heavily intoxicated; and (5) the defendant's ex-girlfriend had a motive to attack the defendant, she was jealous that the defendant was seeing another woman.
Following the presentation of the case summary, participants were presented with closing arguments from the defense and prosecution as well as jury instructions. The jury instructions provided the guiding principles that the jurors were to apply to the evidence in determining guilt for the charge of aggravated assault. The elements necessary to prove aggravated assault were explained (the state must prove beyond a reasonable doubt that (1) the defendant caused bodily injury to another; (2) the defendant caused the injury through the use of a deadly weapon; and (3) the defendant acted purposely), as well as the three elements required to assert the self defense (i.e. (1) the defendant did not create the situation giving rise to the assault; (2) the defendant had a reasonable belief that he was in danger; and (3) there was no opportunity to escape). The aggravated assault and self-defense instructions were modeled after the New Jersey Criminal Code, N.J.S.A. 2C:12-1b(2) and the case State v. Martin, 21 Ohio St.3d 91, 93 (1986). There was no group deliberation phase as our hypotheses focused on individual processing proclivities, leaving to subsequent research questions of interaction, given positive results at this stage.
Finally, participants completed a survey that was presented in two parts. Part I consisted of 11 items that inquired about the participant's opinions, feelings, and guilt determination regarding the case scenario. Aside from the questions on guilt assessment and punishment, all questions employed a Likert scale for participant responses. Participants were initially asked whether they would find the defendant guilty or not based on the information provided. Those participants who found the defendant guilty were subsequently asked to choose what sentencing they would recommend from a list of possible sentence lengths, less than 2 to over 12 years (the range of years was modeled after average sentences for aggravated assault from the Bureau of Justice Statistics of State Court Sentencing of Convicted Felons (Bureau of Justice Statistics, 1996)). Although jury sentencing for non-capital cases is not the norm in America, it has recently been championed by commentators as more democratic, less sensitive to electoral pressure, and better suited to current theories of punishment than sentencing by a judge alone (Hoffmann, 2003; Iontcheva, 2003; Lanni, 1999). Currently six American states (Arkansas, Kentucky, Missouri, Oklahoma, Texas, and Virginia) involve juries in the sentencing determinations of non-capital offences. King and Noble (2005) has reported that juries determine approximately 4,000 felony sentences per year, including battery, aggravated robbery, assault and rape. Thus, although juror sentencing is not normative, it is not uncommon, and influential legal scholars have recommended it for wider use.7
Next, each participant was provided with three pieces of information that were not part of the original trial and were asked to indicate how much each piece of information, if true, would change their rating of the defendant's guilt. Each participant was presented with extralegal information using branching technology in the computerized presentation of the case. The particular extralegal items provided to each participant were a function of whether that particular participant found the defendant guilty or not guilty (there were six pieces in total but only three given to each verdict condition). These pieces of extralegal evidence included, for example, “the defendant volunteers at a soup kitchen in his spare time,” and “the defendant has a drug addiction for which he does not seek treatment.” All six pieces of information represented extralegal information that, from a legal standpoint, should have had no bearing on the jurors' guilt assessment (i.e., although perhaps such character information may be considered by some courts to play a role in sentencing mitigation, they should play no role in determining guilt).
A third set of questions gauged perceptions of the defendant (e.g., how clean did the defendant appear?; how dangerous would you say the defendant appeared?) and was followed by a direct inquiry of how attractive each participant found the defendant. These questions were included to validate the predetermined attractiveness levels of the defendants in each condition and to determine whether R-processors and E-processors viewed attractiveness similarly. Finally, participants were asked to what extent they believed the defendant appeared to be “the type of person” who would commit the alleged crime.
Part II of the data collection consisted of the administration of the most recent version of the REI (Pacini & Epstein, 1999). It was concluded that administering the REI at the conclusion of the study, while risking being influenced by performance in the guilt-determination phase of the study, was considerably smaller than the risk that would result from administering it before that phase, as answering questions about rationality could sensitize participants to this dimension in determining guilt and sentences.
Thus, the offender details and trial information, including the defendant's image, were presented to participants in the following way:
As anticipated, participants' ratings of defendant attractiveness were highly negatively correlated with pre-sorted attractiveness results r(167) = −0.738, p < .01, indicating that the lower participants rated the photographs on attractiveness, the more likely the photos were to be in the less attractive condition. Thus, the means for the reported attractiveness in both conditions conformed to anticipated results (attractive M = 2.95, less attractive M = 1.43). Further, there was complete non-overlap between attractive and less attractive photos for all eight photographs used: the averages for the four less attractive photographs ranged from 1.05 to 1.59, and the averages for the four attractive condition photographs ranged from 2.90 to 3.00. As a preliminary check on the possibility that participants' processing style may have influenced attractiveness perception, paired t-tests were performed on the means of the defendant attractiveness ratings for each of the photographs across processing style. No significant differences were found, indicating the photographs were deemed equally attractive to E- and R-processors. Initial exploration of the data did not reveal any significant trends for sex.
Mode of Analysis
Traditionally REI results have been analyzed in one of two ways. The first is to use correlation and regression analysis on the unadjusted continuous rational and experiential scales, termed continuous scale analysis (Burns & D'Zurilla, 1999; Epstein et al., 1996; Pretz & Totz, 2007). The second way has been to use median population splits to create four groups or quadrants (high experiential/low rational [Q1], high experiential/high rational [Q2], low experiential/low rational [Q3], low experiential/high rational [Q4]) (Berger, 2007; Shiloh, Salton & Sharabi, 2000) termed quadrant sorting analysis. However, for reasons described below, the primary mode of analysis undertaken in the present study required the development of a novel scale.8
Processing Style Influence (PSI) Score and PSI Category Measures
Since the main hypothesis of the present study posits the influence of one system over the other, a measure was created to capture this, if it in fact occurs. The experiential and rational scales were split at the population medians as in the quadrant method. As explained below, the distance from the median on each scale in relation to the other's distance from its median was used to index the influence of one system over the other within a given individual. This measure capitalizes on the possibility that one system can override the other, as CEST theory postulates and as experimental data suggest (see, for example, Berger, 2007). We refer to this measure as a Processing Style Influence (PSI), score and it can be indexed simply by summing for each mock juror the relative positions of both processing styles from their individual medians:
Because we posited that it will be those individuals influenced more by the experiential system than by the rational system who will be the most susceptible to extralegal biases, an index of the extent to which one type of processing trumps the other, as a function of the distance of each from its median, is required. Neither the traditional continuous scale measures nor the four median split quadrant measures are satisfactory for testing this possibility.9
The PSI was developed to make finer predictions. The advantage of PSI scoring lies in its splitting the participants in Q2 and Q3, depending on their behaviors (see contrast between the traditional quadrant method in Figure 1 and the PSI method in Figure 2). By drawing a diagonal across the quadrant grid through the origin, we can hypothesize that participants in Q2 will be more likely to behave like those in Q4, and the same is true for Q3, now Q3b and Q2b are included with the Q4 participants on the basis of which group these individuals are expected to reason most like. Together this group (Q4, Q2b, Q3b) represents a group of participants who had lower E scores than R scores (i.e., they would be assumed to be more influenced by the rational system). The same can be done for Q1, merging it with Q2a and Q3a, to create a measure for experiential-motivated participants who are high experientials/low rationals. Thus, using the PSI measure, the quadrants Q2 and Q3 can be split between Q1 and Q4 based on reasoning tendencies, leading to targeted predictions that cannot be made on the basis of the traditional quadrant assignment. Further, since the PSI generates a numerical measure, the effect of the degree of difference between the two systems in any given individual can be included in the hypothesis testing.
PSI scores were used in both linear and categorical analyses. For the latter, the PSI scores were used to sort participants into primarily experientially influenced processors (E-processors), and primarily rationally influenced processors (R-processors). PSI categories were also used. A negative PSI score reflects placement in the primary R-processor category and a positive score reflects placement in the primary E-processor category. Both PSI scores and categories were used in the regression analysis.
ANOVAS, logistic and ordered logistic regressions, t-tests, Z-tests, multiple regressions, chi-squares and correlation analyses were all utilized. A p-value < .05 was considered statistically significant. Multiple contrasts were adjusted with Bonferroni procedures. And for purposes of both theory and practice, effect sizes were considered either moderate (d-values 0.20–0.50) or larger (Cohen, 1988).
We begin with some descriptive data before describing the results of multivariate analyses. The median processing style scores across the entire sample were higher on rationality than experientiality (77/100 on rationality vs. 67/100 on experientiality), as expected of a university sample. Females on average scored three points lower on rationality than males [t(167)= 2.11, p < .05]; however, there were no significant gender differences for experientiality scores. The rationality scale was negatively correlated with guilt determinations [r(167) = −0.25, p < .01], indicating the higher the rationality score, the more likely the participant was to acquit the defendant, regardless of attractiveness. Consideration of extralegal factors was also negatively correlated with rationality: [r(167) = −0.17, p < .05], the higher the rationality score the less likely the participant was to report that extralegal factors would affect their verdict.
Overall, no attraction leniency bias was found for participants' guilt determinations across the full sample of defendants [c2(1, N = 169) = 0.168, p >.60]. To test whether E-processors would convict less attractive defendants at higher rates than R-processors, a 2 × 2 ANOVA (defendant attractiveness × PSI category) was performed on mock jurors' verdicts. Despite the lack of a main effect for defendant attractiveness [F(1,165) = 0.12, p > .70] there was a significant main effect for processing style: E-processors were 15% more likely to convict than were R-processors, F(1,165) = 5.26, p < .03), d = 1.22. Interestingly, although E-processors did not differentially convict attractive and less attractive defendants, E-processors and R-processors did differ in the rates at which they convicted less attractive defendants (Z-test, p < .05). E-processors were 22% more likely to convict a less attractive defendant than R-processors (89% vs. 67%), d = 1.65, whereas E-processors' conviction rates for attractive defendants did not differ from those of R-processors (7% difference, p > .10) (Table 1). Chi square 2 × 2 (processing style × guilt determinations) cross-tabular distributions supported these results: conviction ratios statistically differed within the group of less attractive defendants [c2(1, N = 85) = 6.38, p < .02] between E- and R-processors but not within the group of attractive defendants [c2(1, N = 84) = 0.54, p = .314]. Figure 3 and Table 1 depict these results.
Table 1. Guilt Determinations by Processing Style and Defendant Attractiveness
Stat Diff btwn Means
The statistical difference between cells was calculating using paired proportion contrast Z-tests.
n = 48
n = 36
Less Attractive Defendant
n = 46
n = 39
**p < .05
Stat Diff btwn Means
**p < .05
n = 94
n = 75
To test whether E-processors would sentence less attractive defendants to longer incarcerations than would R-processors, a 2 × 2 ANOVA (defendant attractiveness × processing influence) was performed on the mock juror sentencing recommendations. The results provide support for the hypothesis that E-processors differentially sentenced less attractive and attractive defendants, treating attractive defendants relatively leniently and less attractive defendants significantly more harshly. The model revealed no significant sentencing effect for either attractiveness or processing style alone; however, the interaction between processing style and the attractiveness of the defendant was significant, F(1,165) = 5.46, p <. 03). E-processors sentenced the less attractive defendants to approximately 22 more months (or 67% longer sentences) in prison than attractive defendants (d = 0.57). In contrast, there was no statistical difference between the amount of time R-processors sentenced attractive and less attractive defendants (Table 2).
Table 2. Sentencing (in years) by Processing Style and Defendant Attractiveness
Stat Diff btwn Means
n = 48; (2.5)
n = 36; (3.2)
Less Attractive Defendant
n = 46; (3.2)
n = 39; (3.5)
***p < .05
Stat Diff btwn Means
***p < .01
To test whether the degree to which the experiential system over-influences the rational system is predictive of an extralegal bias, an ordered logistic regression was performed on sentencing. The independent variables included in the model were the continuous PSI scores, defendant attractiveness, and their interaction. The results mimicked the ANOVA: neither the PSI measure itself nor the defendant's attractiveness achieved significance, but their interaction did (β = 0.05, Z = 2.3, p < .02). Substantively, this finding indicates that the greater the degree to which the experiential system trumps the rational system, the greater the disparity between sentencing attractive and less attractive defendants (see Figure 4).
Although the two types of processors did not sentence differentially for attractive defendants, E-processors sentenced unattractive defendants to 4.5 years versus only 2.9 years by R-processors. This reinforces our claim that the quadrant and continuous scales that in the past have been used in CEST research lack sensitivity to reveal processing differences that result from relative dominance of one type of processing over the other. Neither of those traditional scales would reveal this finding (see Appendix A).
Extralegal Factor Consideration
Participants were asked a series of three questions gauging how likely various pieces of extralegal information would be to affect their guilt verdicts. To create an overall measure of extralegal bias, the results from the three extralegal bias measures were summed for each verdict condition, the scores ranging from 0 (least likely extralegal used in verdict) to 15 (the most likely). An ordered logistic regression was performed on the combined extralegal measure with the PSI scores as the independent variable. The results indicated that the more likely the experiential system was to influence the participant over the rational system, the more likely they were to report that extralegal factors would affect their verdicts (β = 0.02, Z = 2.0, p < .05) (see Figure 5). As can be seen in Appendix A, traditional statistical analyses involving both the continuous scale and quadrant approaches failed to reach significance on participants' extralegal considerations, further supporting the strength of the PSI measure.
Using the PSI method, E-processors were 7% more likely to report that extralegal factors would affect their verdicts than were R-processors (p < .01, d = 0.51. Paired t-tests revealed that this trend was statistically significant for five of the six measures (see Table 3). Out of all the measures, both groups reported the extralegal fact that “the defendant has a drug addiction for which he does not seek treatment” would be most likely to affect their decision of guilt; E-processors, however, were 14% more likely than R-processors to find this factor influential (p < .05) and they were also 12% more likely to report that the extralegal factor “the defendant collects welfare and unemployment benefits” would likely affect their guilt determination (p < .05).
Table 3. Extralegal information measure responses by processing style
Reported values based on the means of 1–5 Likert scale responses (1, not at all likely to change my verdict; 5, positively would change my verdict).
% increase = the percentage that experientially influenced thinkers are likely to consider a particular extralegal factor above the rationally influenced processors.
t-tests performed on means underlying the reported percentages.
While in college the defendant was on the dean's list
***p < .01
The defendant goes to church almost every Sunday
**p < .05
The defendant volunteers at a soup kitchen during his free time
**p < .05
The defendant lost custody of his son in a court proceeding
The defendant collects welfare and unemployment benefits
Overall, those who found the defendant not guilty were 12% more likely to report extralegal factors would cause them to change their verdict from not guilty to guilty than were those who found the defendant guilty to report that the additional extralegal factors would change their verdict from guilty to not guilty (p < .01). Given that some extralegal factors were defense oriented (e.g., doing charity work) and others were prosecution oriented (e.g., prior drug conviction), this suggests that guilt-finders were more steadfast in their verdicts, or at the very least that the extralegal factors that favored the defense were less potent than those that favored the prosecution. Below we argue that evidence of drug use was the primary factor driving this result.
To test for a potential stereotype bias, all participants were asked how likely the defendant appeared “to be the type of person who could commit a crime like this.” The responses ranged from 1 to 5, from “very unlikely” to “very likely.” A 2 × 2 ANOVA was performed on the stereotype bias measure (processing style categories, defendant attractiveness and their interaction). Processing style [F(1,165) = 5.36, p < .03] and defendant attractiveness [F(1,165) = 17.70, p < .001) were both significant as main effects whereas their interaction was not. Although experiential and rational processors were both more likely to report the less attractive defendant was more likely to fit the criminal stereotype, the degree to which E-processors indicated this was 11% higher than R-processors (p < .05). Overall, E-processors were 8% more likely to report the defendant looked like the type of person who could commit the crime regardless of attractiveness, and overall the less attractive defendants were rated 16% more likely by both processing groups to be the “type of person” who could have committed the crime.
In summary, the seven primary findings from this study were: (1) E-processors convicted less attractive defendants 22% higher than R-processors; (2) E-processors sentenced less attractive defendants on average to 22 months longer prison sentences than they sentenced attractive defendants, whereas R-processors did not differentially sentence as a function of attractiveness; (3) the greater the degree to which the experiential system trumped the rational system in a given participant the more likely that participant was to sentence the less attractive defendant to harsher sentences; (4) E-processors were more likely to report that extralegal factors would have changed their verdicts (e.g. charity work of defendant); (5) the greater the degree to which the experiential system trumped the rational system in a given participant, the more likely that participant was to report extralegal factors would affect their verdicts; (6) there was a trend for E-processors to rate less attractive defendants as more likely to be the “type of person” who crimes; and (7) E-processors also exhibited global tendency to be more conviction prone regardless of defendant attractiveness, overall convicting defendants at 15% higher rate than R-processors.
Although there was no overall attraction-leniency bias found across the entire sample, nor was an attraction-leniency bias within processing style found for guilt determinations, this is consistent with what others have reported. For example, Lieberman (2001) found no interaction between attractiveness and processing style for civil liability but he did find such an effect for damages. And Erian et al. (1998) found attractiveness did not predict guilt determination but did predict sentencing, as was found in the present study, too.
However, although attractive defendants fared no better in general – even if they had experiential processors making their guilt judgments (i.e. they were convicted at the same rates by E- and R-processors) – less attractive defendants did fare worse if their guilt determination was made by an E-processor, being 22% more likely to be convicted by an E-processor than by an R-processor. So rather than an attraction-leniency bias operating among E-processors, an “unattractive harshness” conviction bias was displayed. Although the magnitudes of some of the effects were moderate (e.g. 0.57), they nevertheless would seem to have some real-world importance because they are large enough to portend practically important outcomes (e.g. E-processors sentenced unattractive defendants to 22 months longer prison terms than they sentenced attractive defendants for the same crime).
Extralegal information was shown to be more influential in changing a participant's verdict from not guilty to guilty than vice versa. This result appears to be driven by a general bias across both processing styles against defendants who are drug users. Both E-processors and R-processors were more likely to report that the extralegal fact that the defendant was a drug addict was likely to reverse their not-guilty verdicts. This is consistent with research reporting a general discrimination against drug users (e.g. Day, Ross & Dolan (2003) reporting among a sample of those with hepatitis C – 22% reported perceived discrimination due to their drug use in a variety of settings, including the justice system). This is also consistent with the Federal Rules of Evidence, which aim to keep information from juries about prejudicial defendant drug use while on trial for an unrelated crime (see relevant sections of Federal Rules of Evidence 401; 403).
The present study was designed to explore the possibility that individual processing characteristics of mock jurors might reliably differentiate their verdicts and sentencing, and the results revealed surprisingly strong effects for the most part. Future research might examine the causal basis for this finding. Although it was not designed to test them, the present study does suggest several potential candidates. One possible mechanism might be a tendency among E-processors to employ an “availability heuristic,” which leads them to endorse outcomes most representative of extralegal stimuli, while perhaps leaving them insensitive to the reliability of relevant evidence. (Kahneman & Tversky, 1973). One of the reasons E-processors were more likely to exhibit decision-making errors may be that the availability heuristic leads them to look at a given defendant during a criminal trial, assess that defendant, and then compare that defendant to their preconceived notions of what a criminal “should” or is expected to look like. (Verosky & Todorov, 2010). If the appearance of the defendant does coincide with what the E-processor believes a criminal is expected to look like then it may be less cognitively stressful for them to conclude that the particular defendant before them did indeed commit the crime. This is, of course, a testable hypothesis that future research might tackle.
The results show what many before us have opined (e.g. Horowitz et al., 1998), namely, that jurors' mode of thinking can be a powerful force affecting the consideration of extralegal information. Processing style may not only be a useful construct for the researcher in generating hypotheses, but it may be another tool in the scientific jury selection toolbox for practitioners trying to empanel a jury with a lower likelihood of extralegal bias susceptibility. To date, the process of voir dire, which is predicated on the assumption that jurors may have predispositions that favor one side, has not attempted to examine socio-cognitive processing styles of potential jurors. The present findings, if replicated and extended to group decision-making contexts that resemble actual jury deliberations, could help establish the usefulness of juror selection for specific processing styles or help guide the planned presentation of evidence, given the strategy and evidentiary presentation that is being planned, if attorneys can somehow adapt the questionnaire or a set of oral questions to categorize venire members.
In sum, this study provides evidence that the exploration of individual processing variables may have substantial pay-offs for both researchers and practitioners. The present results reveal the relevance of CEST-defined processing styles when parsed according the PSI measure developed here for studies of juridical decision-making; the PSI score and category methods used appear to have an advantage over the traditional custom of treating the scales as completely separate and continuous or as a 2 × 2 quadrant classification that is impervious to the degree that one type dominates the other within a quadrant (See Appendix A for full comparison of all three methods of analysis.).
Finally, despite what we regard as several strengths of the present design (the use of multiple criminal photographs rather than only one or two that is common in this type of research, the use of actual case materials and actual jury instructions, etc.), there are six areas for future research that can extend the present study.
First, although research has found few differences in mock jury results due to presentation style (i.e. video, live presentation or case description), or as a function of the juror pool (i.e. students vs. the community members; Bornstein, 1999), mock jurors lack one important feature that real jurors face – a consequence for their decisions.10 Thus, it will be important to replicate the present results in quasi-experimental and naturalistic studies, using voir dire pools or community samples from which actual juror pools are drawn and for cases in which actual consequences result.
Second, in the present study there is no sense of how processing style may moderate or even mediate jurors' behaviors during group deliberations. Perhaps a high level of rationality of one juror can override a high level of experientiality of others. The current research is a first step in the process of examining the relationship of processing style to outcomes, and clearly more work will be necessary to answer this and a host of related questions, such as using CEST theory to derive predicted forms of the interaction, specifying a priori the conditions where the rational processing mode would offset the experiential mode versus conditions where the experiential mode would overpower the rational.
Third, it is of interest to determine whether participants can be more reliably sorted into rational and experiential groups based on existing psychometric measures or observed behavior as opposed to answers to the REI questions. For instance, rather than using the REI to sort participants, perhaps answers to such problems as the “Linda conjunction,” odds ratio questions, or responses to other measures such as the Cognitive Reflection Test (CRT) can be used to assess whether individuals lured into judgments on the basis of “gut intuitions” correspond to the REI classification developed in the current study.11
Fourth, there is some evidence that more intelligent individuals express a greater need for cognition (e.g. enjoy grappling with ambiguous evidence and complex ideas); for instance, Cacciopo and Petty (1982) reported a correlation of 0.39 between need for cognition and college entrance examination scores. And there is also evidence that dogmatic and authoritarian individuals are less cognitively complex. This leads to such questions as: are R-processors more likely to score higher on measures of general intelligence; and are E-processors more likely to allow an initial piece of evidence to frame how they think about and interpret subsequent evidence (e.g. Carlson & Russo, 2001; Cialdini, Petty, & Cacciopo, 1981)?
Fifth, follow-up work will be needed to determine how generalizeable the present effect is. It might be, for instance, that it depends upon the strength of the evidence such that defendant attractiveness becomes especially pertinent for E-processors when evidence is weak than when it is strong. The value of PSI scoring lies in hypothesis-testing because the method generates a numeric measure based on the degree of influence of one processing system over the other in any given individual. It also allows targeted predictions, with increased power, based on grouping individuals as rationally or experientially influenced reasoners, predictions that cannot be made on the basis of the traditional quadrant assignment. However, PSI is not without limitation. Although PSI aids in hypothesis-testing regarding rational and experiential reasoners, for researchers targeting hypotheses related specifically to subgroups of REI participants who either score high on both rationality and experientiality or low on both, the PSI measure will be less helpful since it does not treat these groups as separate and distinct classes of reasoners. Also because no technique has been developed to assess precisely what mode an individual uses to process information at any particular point in time, we must focus on the general reasoning patterns of those who showed a global tendency to be more influenced either by rationality or by emotionality. Thus, our discussion is not intended, and should not be read, as a generalization about the precise process an individual may have used to reach a conclusion. Instead we focus on the pattern that emerged in reasoners who tended to be more influenced by either rationality or emotionality.
Finally, future work assessing various extralegal factors aside from defendant attractiveness will be important to flesh out the boundary conditions whereby processing style influences juror behavior.
Notwithstanding these caveats and areas in need of future work, one thing seems clear from the present study: the previous emphasis by psycholegal researchers on external characteristics of jurors and defendants (race, age, gender, attractiveness, etc.) can be profitably joined with the study of internal processing characteristics.
Although at times and depending on the nature of the case, these factors can be relevant (e.g., if there is an allegation of race, age or gender discrimination). However, in such situations these factors would not be extralegal. In contrast, the present study focuses on a situation in which these factors are not relevant and are therefore extralegal.
Also, when attractiveness is thought to be a component of the crime, for example swindling, conning, or espionage, wherein the defendant “exploits” their good looks, the bias can reverse (Sigall & Ostrove, 1975; see also Wuesch, Castellow, & Moore, 1991).
This dual-mode distinction has taken many names in the psychological science literature, and psychologists over the years have proposed a number of dual-process models of cognition along similar principles to CEST; some examples include: “heuristic system” v. “analytic system,” “associative system” v. “rule-based system,” “system 1” v. “system 2” (see, for example, Guthrie, Rachlinski, & Wistrich, 2007, reviewing this research), a “gist-based” vs. a “verbatim” system (Brainerd, Reyna, Wright, & Mojardin, 2003). Unlike these other dual-process theories, CEST has led to the validation of a number of psychometric measures, such as the Rational Experiential Inventory, developed to identify experiential and rational system influences within individuals. CEST's empirical validation has been acknowledged as “far more extensive than other dual-process models in its description of the principles of the experiential system” (Lieberman, 2002).
The survey was administered through www.questionpro.com which provided a research grant to conduct this study.
Moreover, there is ample research, some dating back a half century, showing that judges' and jurors' verdicts, and punishments are highly correlated. For example, Kalven (1964) analyzed how 600 judges reported they would have ruled in 8,000 civil and criminal cases that had been heard by jurors; he found agreement rates of 79% and 80%, respectively, between judges and juries in criminal and civil cases. Wissler, Hart, and Saks (1999) reported similar assessments of severity among jury-eligible mock jurors, lawyers, and civil trial judges' assessments of injury severity and general damages in personal injury cases (though judges awarded smaller general damages). Robbennolt (2002) compared jury-eligible adults and trial judges' awards of compensatory and punitive damages in a mock personal injury case. There were differences in awards of compensatory but not punitive damages. And finally, Guthrie, Rachlinski, and Wistrich (2001) reported that magistrate judges were susceptible to the same decision-making biases that influence laypersons decision-making. Taken together, these analyses suggest that judges' decision-making is often similar to actual and mock jurors in terms of verdicts and punishments. Thus, the present focus on juror outcomes may have implications for how judges themselves would behave in a similar case.
For an analysis of the data under the continuous scale method and quadrant sorting method see Appendix A.
For example, when testing interactions between rationality and experientality, using the continuous scale approach, an individual with a score of 90 on rationality and 60 on experientiality would be treated the same but oppositely to an individual with a score of 60 on rationality and 90 on experientiality. However, this is an empirical question that ought not be preordained by the measurement scale employed. Similarly, using the quadrant method, an individual sorted into the high rational/high experiential quadrant would have scored above the median on both scales. However, such an individual may have scored one point above the median in rationality and 20 points above the median in experientiality, and would actually be 19 points higher above the median in the latter but would not be treated any differently than someone 30 points above the median on both scales. Likewise for an individual who is below the median in both experientiality and rationality but more so in one than the other. Our hypotheses depend on finer-grained scale sensitivity, which led to the creation of the PSI. If we were making hypotheses using the quadrant sorting method, we would predict that Q1 jurors would be susceptible to the biases such as a defendant's attractiveness, because they are more likely to be influenced by the experiential system; the quadrant sorting method would also predict that Q4 jurors would be least likely to exhibit such bias. However, according to the quadrant sorting method, it is unclear what to expect from a mock juror who is neither rational nor experiential or someone who is both. Thus, the quadrant method does not lead to clear expectations regarding Q2 or Q3 jurors. As a result, hypothesis testing with the quadrant method is primarily concerned with Q1 and Q4 and has little to say about Q2 and Q3. The principle hypotheses in this study would simply be that we would expect greater evidence of reasoning errors in Q1 than in Q4. This is fine as far as it goes, but it is silent about other combinations.
We are not claiming that mock juror simulations are irrelevant to actual juror performance, as numerous studies have demonstrated similar outcomes. Often the effects found in simulations are replicated in research addressing the real-world judicial process (see, for example, Taylor & Hosch's (Taylor and Hosch, 20042004) finding that ethnicity predicted sentencing length in actual cases, and Zeborwistz & McDonald's (1991) finding that attraction-leniency bias operates in actual small claims cases brought before judges in much the same way it has been found to operate among mock jurors). Our point is merely that mock jurors' processing styles may be moderated by the consequences of jurors' decisions and other jurors' counter-arguments, and the only way we will know if this occurs is to replicate mock juror studies with actual jurors.
The CRT distinguishes between intuitive and deliberative processing styles, on the basis of three questions: (1) A bat and ball cost $1.10 in total. The bat costs $1.00 more than the ball. How much does the ball cost? (2) If it takes 5 minutes to make five widgets, how long would it take 100 machines to make 100 widgets? (3) In a lake there is a patch of lily pads. Every day the patch doubles in size. If it takes 48 days for the patch to cover the entire lake, how long would it take for the patch to cover half the lake? Each of the questions has an incorrect answer that immediately pops into mind (see Guthrie, Rachlinski, & Wistrich 2007).
Central to the present analyses is the assumption of greater validity of the PSI measure over the two traditional measures. One means of establishing validity is by making (and finding support for) a priori predictions that the traditional measures cannot make. As seen, the PSI measure led to the generation and confirmation of a number of hypotheses. Below we show that the traditional measures do not lead to these same hypotheses, nor do they provide support for some of them. However, before delving into these differential predictions, a general concern needs to be raised about the PSI reconfiguration of a four-quadrant theory into a dual-process theory. One might imagine that if the four original quadrants are each posited to have their own unique characteristics, then reconfiguring them into a two-process system will result in less fine-grained expectancies. However, the demonstration of increased significance and power that resulted from using the PSI suggests otherwise. To assess the difference between using the PSI score and using either of the two conventional dependent measures of classifying participants (continuous scale, quadrant), all analyses were also run using the two conventional methods.
Occasionally, the quadrant and continuous scale methods led to significant findings that were not significant when the PSI score was used, and the conventional methods occasionally failed to yield reliable results when the PSI measure did. However, for most analyses, the different measures converged, and often when they did not, increased significance and reliability were noted in the PSI measure. The real superiority of the PSI method, however, is that it not only leads to the same hypotheses as the conventional methods, but it also leads to the generation and testing of hypotheses that could not be tested using the other methods. Below are the results of the conventional analyses compared with the PSI measure used here.
As can be seen in Chart 1, the quadrant hypothesis is somewhat supported by our results since there is evidence that Q1 mock jurors did exhibit reasoning errors whereas Q4 jurors did not. However, there are significant results for Q2 and Q3 that are hard to make sense of theoretically because of the wide range of participant scores in these groups. When Q2 and Q3 participants are sorted into Q1 and Q4 based on their reasoning characteristics, the results fall neatly, almost totally as expected, into place (see Chart 2). As can be seen, all of the reasoning errors uncovered are linked to E-processors, with no results too ambiguous to categorize or too varied to interpret.
– 21% more likely to convict less attractive defendants. (guilt)
No significant findings.
Appendix A Chart 2
E- processors (all comparisons are to R-processors)
(No reasoning errors found)
– 15% more likely to convict overall across all attractiveness levels.
– 22% more likely to convict less attractive defendants.
– E-Processors sentenced less attractive defendants to 22 more months.
– E-Processors were 7% more likely to report extralegal factors in general would change their verdict. 14% were more likely to report a drug addiction would change their verdicts and 12% more likely to report receiving welfare would change their verdict.
– E-Processors were 11% more likely to report unattractive defendants appear to be the type of people that commit crimes.
Appendix A Chart 3
No interaction effect.
No significant findings.
Q1 (High E Low R)– No significant findings.
Q2 (High E High R)– 21% more likely to convict less attractive defendants.
E_Processors: 15% more likely to convict overall regardless of defendant attractiveness.
Q3 (Low R Low E)– No significant findings.
E_Processors: 22% more likely to convict less attractive defendants.
Q4 (High R Low E)- No significant findings.
E-Processors: sentenced less attractive defendants to 22 more months in prison.
Interaction between R and E scales was a significant predictor of sentences.
Q3– significant attraction leniency effect (sentenced less attractive defendants to approximately 31 more months in prison).
Logistic regression revealed the degree to which the experiential system trumps the rational system (PSI Score) is a significant predictor of differential sentencing between attractive and less attractive defendants.
E scores alone were a marginally significant predictor of sentences.
Q4– No significant findings.
E-Processors : were 7% more likely to report extralegal factors in general would change their verdict, 14% were more likely to report a drug addiction would change their verdicts and 12% more likely to report receiving welfare would change their verdict.
As for the continuous scales, the only significant results were found with regard to sentencing (see Chart 3). Consistent with the PSI results, the interaction of the pure rationality and experientiality scores was found to be a significant predictor of sentences. Pure experientiality scores alone were also a marginally significant predictor of sentences.
What follows are the full descriptions and results for the two traditional methods of analysis.
The continuous scale method was tested in a logistic regression on guilt determinations using defendant attractiveness, rationality scale, experientiality scale, defendant attractiveness × rationality scale, experientiality scale × defendant attractiveness, rationality scale × experientiality scale and a three-way interaction among experientiality scale × rationality scale × defendant attractiveness. None of the results were significant. The conventional quadrant method, however, did yield results similar to those found with the PSI measure, although several of the trends were only marginally significant: contingent logistic regressions were performed for each processing quadrant. For, Q2, the results were marginally significant (β = 1.34, Z= 1.70, p < .10); this group convicted less attractive defendants at a 21% higher rate than attractive defendants.
An ordered logit regression using the experientiality and rationality continuous scales was performed with sentencing as the dependent variable and the independent variables included the rationality scale, the experientiality scale, defendant attractiveness, the interaction between rationality and defendant attractiveness, and the interaction between experientiality and defendant attractiveness. No results reached significance. However, the quadrant method did lead to two significant results: Q3 (β = 2.44, Z = 2.74, p < .01) and Q1 (β = 1.81, Z = 2.76, p < .01).
An ordered logit regression using the experientiality and rationality continuous scales was performed with sentencing as the dependent variable, and the independent variables included the rationality scale, the experientiality scale, defendant attractiveness, the interaction between rationality and defendant attractiveness, and the interaction between experientiality and defendant attractiveness. The interaction between the rationality scale and the experiential scale (β = –0.004, Z = –1.97, p < .05) was significant, and the experiential scale alone was a marginally significant predictor (β = 0.266, Z = 1.74, p < .10), no other factors achieved significance. Similar contingent ordered logit regressions were performed for each of the median split quadrants, where attractiveness was tested as a predictor of sentencing. Significant results were obtained for Q3 (β = 1.40, Z = 2.04, p < .05). There was a marginally significant result for Q1 (β = .96, Z = 1.84, 0.266, p < .07).
Extralegal Consideration Bias
In juxtaposition to the significant findings using the PSI measure to predict extralegal biases, statistical analyses involving both the continuous scale and quadrant approaches failed to reach significance on participants' extralegal considerations.
An ordered logit regression using the experientiality and rationality continuous scales was also performed where stereotype bias was the dependent variable and the independent variables included the rationality scale, the experientiality scale, defendant attractiveness, the interaction between rationality and defendant attractiveness, the interaction between experientiality and defendant attractiveness, and the interaction between the rationality and experientiality scales. No terms achieved significance.
Contingent ordered logit regressions were performed for each of the median split quadrants, where attractiveness was tested as a predictor for the stereotype bias measure. Significant results were obtained for the Q3 (β = 2.44, Z = 2.74, p < .01) and Q1 (β = 1.81, Z = 2.76, p < .01).
In sum, the continuous scales yielded only one reliable effect (p = .049), which was that sentencing was a function of the interaction between the rational and experiential scales regardless of defendant attractiveness. This suggests that as an individual decreases in the scores on both scales, the higher is their recommended sentence regardless of defendant attractiveness. No interaction effects between processing style and defendant attractiveness were revealed. The quadrant sorting method led to three significant effects: Q3 processors differentially sentenced less attractive and attractive defendants, assigning less attractive defendants to approximately 31 months 6 days longer on average. Q3 and Q1 were both significantly likely to report that the less attractive defendants appeared to fit a criminal stereotype (i.e. appeared to be the type of person who would commit the crime). Comparing these results to the main results, discussed supra, it is apparent that the PSI measure allows for finer-grained hypotheses testing and analyses.