Plea-bargaining law: The impact of innocence, trial penalty,
and conviction probability on plea outcomes
Miko M. Wilford, Ph.D.,1 Gary L. Wells, Ph.D.,2 and Annabelle Shestak, M.S.1
1Department of Psychology, University of Massachusetts Lowell
2Department of Psychology, Iowa State University
Miko M. Wilford, Department of Psychology, University of Massachusetts Lowell; Gary L. Wells, Department of Psychology, Iowa State University
This material is based upon work supported by a National Science Foundation Graduate Research Fellowship to the first author under Grant no. 202-18-94-00. A special thanks to the talented research assistants who served as experimenters, confederates, and data coders: Sarah Buck, Monica C. Van Horn, Lauren Anderson, Hannah Balka, Kelly Connolly, Kirsten Cownie, Shelby Forsythe, Taylor M. Harrison, Katharine L. Hughes, Diana Karavida, Megan Kellogg, Elizabeth Larsen, Landon Momberg, Caisa E. Royer, Desirae Runyon, Leah Speed, Deanna Steinbach, Anna Turosak, Carter Cotrupi, Nicole Dalton, Annmarie Khairella, and Renee Lajoie. The authors would also like to thank Andrew M. Smith for his advice regarding the analyses.
Correspondence concerning this article can be sent to Miko M. Wilford, Department of Psychology, University of Massachusetts Lowell, 113 Wilder St., Ste. 300, Lowell, MA 01854.
Objectives: Despite the prevalence of plea-based convictions, we know relatively little about factors that influence the decision to plead. Two experiments were conducted to examine the effects of guilt/innocence status, individual-difference variables, trial penalty, and the likelihood of conviction on rates of plea acceptance.
Methods: Both studies used an adaptation of the cheating paradigm (Russano, Meissner, Narchet, & Kassin, 2005): participants are randomly induced to cheat during the study while others are not. All participants are later accused of cheating. In Experiment 2, trial penalty was manipulated by altering the potential punishment if the plea were rejected, while holding the punishment if the plea were accepted constant. We also manipulated likelihood of conviction by telling participants how likely the professor in charge of the study thought they were of being found guilty of cheating.
Results: In both experiments, guilty participants were significantly more likely to accept plea offers than the innocent. Logistic analyses examining the impact of individual-difference measures on plea outcomes failed to produce consistent evidence of moderating effects. In Experiment 2, manipulating conviction probability affected plea acceptance only among the innocent. The trial-penalty manipulation had no significant effect on plea outcomes among the innocent or the guilty. Reasons for accepting the plea differed between the innocent and the guilty, whereas their reasons for rejecting followed similar patterns.
Conclusions: Overall, this research highlights several avenues for further research aimed at improving the current system of pleas. It also illustrates the ease with which false guilty pleas can be attained.
Keywords: adjudication, false guilty pleas, legal processes, plea bargaining
Plea-bargaining law: The impact of innocence, trial penalty, and conviction probability on plea outcomes
“… criminal justice today is for the most part a system of pleas,
not a system of trials” (Lafler v. Cooper, 2012, p. 11)
With the above quote, the U.S. Supreme Court acknowledged and underscored a shocking trend in criminal prosecution. Prior to the 1980s, approximately 20% of federal criminal convictions were obtained via a courtroom trial process (Oppel, 2011). By 2006, this number had dropped to 5% (Burke, 2007; Ross, 2006), and even more recently, the Bureau of Justice Statistics (2015) documented that only 2.6% of federal criminal convictions were the result of trials. Instead, as Justice Kennedy wrote for the majority in Lafler v. Cooper (2012), all except a very small percentage of criminal convictions are obtained within a system of pleas.
Scholars have long-posited that the plea system could lead to false guilty pleas, but a number of obstacles have made it difficult to observe such cases (Fisher, 2000). First, entering a guilty plea requires a waiver of certain avenues for appeal—without an appeal, the possibility of uncovering an erroneous conviction diminishes greatly (Stephens, 2013). Second, those who are permitted to appeal will face greater challenges because their guilty plea (even if it omitted a guilt admission) can be used as evidence of guilt during the appeals process. Third, many of the organizations involved in exonerating the wrongfully convicted receive far more cases than they can investigate. Therefore, they apply stringent criteria to choose the cases they pursue, which sometimes means excluding defendants who pled guilty (Redlich, 2010a). Fourth, the availability of exonerating evidence, primarily DNA, is often limited to severe crimes (e.g., murder, rape), and guilty pleas are less common in severe crimes (Bureau of Justice Statistics, 2015). Despite these obstacles, the National Registry of Exonerations (2015) reported that nearly 50% of the exoneration cases documented in 2015 involved a false guilty plea. In fact, exoneration cases involving false guilty pleas have been steadily increasing for the last seven years (Innocence Project, 2017; National Registry of Exonerations, 2015). This rise might be due to our increased ability to discover exonerating evidence rather than an increase in false guilty pleas per se, but the fact remains that guilty pleas do lead to wrongful convictions that often go undetected.
Studying the Decision to Plead
Why would people choose to accept plea offers for crimes they did not commit? Answering this, and related questions, is difficult. Most empirical studies on pleas are vignette studies in which people are asked what they would do in a hypothetical situation (Bordens, 1984; McAllister & Bregman, 1986; Redlich & Shteynberg, 2016; Redlich, Wilford, & Bushway, 2017; Zimmerman & Hunter, 2018). Although these studies have made important contributions to our understanding of pleas, questions have been raised about the extent to which people’s decisions in purely hypothetical situations reflect what they would do if the decision involved personal stakes. Thus, when describing the paradigm used in the current experiments, we use the term “real stakes”. Although participants do not experience the consequences presented in this paradigm, the participants believe them to be real at the time they make their decision to accept or reject the plea offer. In the tradition of classic experimental social psychology, we have created conditions that capture experimental realism (Aronson, Wilson, & Brewer, 1998). Although we do not claim that the magnitude of the consequences in our paradigm resemble those that face a criminal defendant charged with a serious crime, it nevertheless is the case that the consequences are perceived to be real rather than hypothetical (Redlich et al., 2017).
In two experiments, we used a modified version of a paradigm that has been used to study false confessions—commonly called the cheating paradigm (Russano, Meissner, Narchet, & Kassin, 2005). This paradigm provides a method of randomly assigning participants to innocence or guilt. The changes we made to the cheating paradigm were needed to highlight some of the critical differences between pleas and confessions. For instance, an important feature of the modified paradigm was that a confession was not required of participants who accepted the plea.1 This is consistent with the fact that Alford and nolo contendere pleas do not require a guilt admission and are accepted in most jurisdictions (Redlich & Ozdogru, 2009). Alford pleas allow defendants to maintain their innocence while accepting a plea deal, and nolo contendere (or no contest) pleas do not require defendants to make any statements as to their guilt or innocence while accepting a plea offer (Hudson v. United States, 1926; North Carolina v. Alford, 1970). Whereas Alford and nolo contendere pleas are less common than the more traditional or standard guilty pleas in which the defendant does admit guilt, they are not rare occurrences. In 2004, a Bureau of Justice Statistics survey recorded a combined 2,553 (17.6% of those surveyed) Alford and nolo contendere pleas. Redlich and Ozdogru (2009) used these numbers to estimate the total population of state inmates who had entered one of these pleas and estimated that 207,181 inmates would fall within these categories.
Omitting the need to confess was additionally important as a recent study, relying on a similar paradigm, has demonstrated that different processes may underlie confession and plea decisions (Wilford & Wells, 2018). Also, unlike a confession, accepting a plea deal involves an explicit disclosure of consequences and benefits, with multiple opportunities for the defendant to affirm his decision or change his mind (Redlich, 2010a). In contrast, confessions obtained through explicit promises or threats are legally inadmissible. Hence, confessions are much harder to exclude when lawfully-obtained (Kassin, 1997). Consequently, some have argued that accepting and entering a plea agreement may represent a rational decision, even in the face of factual innocence (Bibas, 2004). Research also shows that among factually-innocent defendants, false confessions and pleas often co-occur (Kassin, 2012; Redlich, 2010a; 2010b). One likely explanation is that, having falsely confessed, factually-innocent defendants find themselves compelled to take a plea due to the overwhelming likelihood of conviction caused by their confessions. As a result, adapting the paradigm to the study of pleas meant we had to exclude the undue influence of confessions on the plea process.
The Plea Discount (or Trial Penalty)
Plea deals are predicated on an explicit guarantee of reduced sentences or charges as compared to those that would otherwise be expected upon conviction at trial. Most jurisdictions currently provide no guidance as to the appropriate size of this discount or penalty. Hence, prosecutors have considerable discretion in determining the size of the plea discount or trial penalty for each case (Fisher, 2000; Gazal-Ayal, Turjeman, & Fishman, 2013). Analyses of real cases indicate that those convicted at trial received sentences that were 29.6% longer than the sentences of defendants in similar cases who entered plea agreements (Bushway & Redlich, 2012). In analyzing federal cases, Burke (2007) found that the average federal sentence resulting from a plea conviction was only one-third the sentence that those convicted at trial typically received. Thus, whereas trial penalties (applied through sentence and/or charge reductions) clearly exist, it is still unclear what impact these penalties have on plea outcomes.
Most vignette studies have found a positive relationship between increases in trial penalty and likelihood of plea acceptance (Bordens, 1984; Gregory, Mowen, & Linder, 1978, Exp. 1; McAllister & Bregman, 1986a; Zimmerman & Hunter, 2018). Only one study has examined the impact of the trial penalty in a real-stakes environment (Dervan & Edkins, 2013).2 Participants were accused of cheating (using an adaptation of the cheating paradigm) and were offered a plea wherein they could admit their guilt and forfeit their study credits. If they refused the plea offer and were found guilty by an academic review board, the board would forfeit their credit and enforce enrollment in an ethics course. The researchers varied the duration of the ethics course to manipulate the size of the plea discount. Whereas the plea acceptance rates were impacted by this change (in the expected direction) the difference was not statistically significant, nor did the magnitude of the effect differ between the innocent and the guilty. Other studies have examined the effect of the trial penalty on the offers or recommendations made by attorneys, producing mixed results (Kramer, Wolbransky, & Heilbrun, 2007; McAllister & Bregman, 1986a; 1986b).
Probability of Conviction
Research examining the impact of the probability of conviction on plea outcomes has relied exclusively on hypothetical scenarios.3 McAllister and Bregman (1986a) manipulated both trial penalty and conviction probability, asking participants to report whether they would accept a plea. Participant-defendants were more likely to accept a plea as the probability of conviction increased. More recently, Zimmerman and Hunter (2018) also found an increase in plea acceptance as conviction likelihood increased.
Other vignette studies have manipulated both guilt status and conviction probability, finding that acceptance generally increases as probability of conviction increases, though the pattern differs somewhat between those who imagined they were innocent versus guilty. Specifically, Bordens (1984) found that the increase in the likelihood of pleading among the guilty occurred when the probability of conviction reached 50% and remained constant with further increases, while the same increase only occurred among the innocent when the probability of conviction reached 90%. In contrast, Tor et al. (2010) found that the likelihood of pleading among the guilty was highest when conviction probability was between 50-70% and decreased significantly as the probability increased beyond 70%; among the innocent, in contrast, likelihood of pleading rose sharply as the probability of conviction reached 70% and continued to rise further. Other studies have demonstrated a similar effect of conviction probability on the likelihood of pleading, though the threshold probability necessary to increase plea acceptance varied across studies (Helm & Reyna, 2017; Helm et al., 2018).
The dominant theory on plea decision-making, the shadow-of-the-trial model (Bibas, 2004; Bushway & Redlich, 2012; Bushway, Redlich, & Norris, 2014), predicts that decisions to accept a plea should be driven by the expected value of the plea offer. The plea sentence should be less than the probability of conviction multiplied by the expected trial sentence if it is to be accepted. Thus, we would expect prosecutors to offer relatively small plea discounts (or trial penalties) when conviction probability at trial is high, and larger discounts when the probability drops. However, this model has been criticized due to its simplicity (Bibas, 2004). Most notably, researchers have questioned the model’s validity given its omission of guilt status as a predictor of plea outcomes. Numerous studies have shown that guilty individuals are more likely to accept a plea than innocent individuals, even when the probability of conviction (or evidence strength) is kept constant (Bordens, 1984; Dervan & Edkins, 2013, Gregory et al., 1978; Redlich & Shteynberg, 2016; Tor et al., 2010; Wilford & Wells, 2018). This difference could be due, at least in part, to the phenomenology of innocence (Kassin, 2005). Innocent defendants–particularly those who have stronger just world beliefs—overestimate the transparency of their innocence to others, which can lead them to make decisions that are contrary to their best interests (e.g., falsely confessing to escape a noxious interrogation).
Individual Decision-Making Factors
Very few studies to-date have evaluated individual difference factors that lead people to enter plea agreements. Until recently, research that has examined individual differences in plea decision-making have centered around two primary factors: age and ethnicity. Specifically, minority defendants are likely to receive plea offers that include relatively smaller charge or sentencing discounts (Kutateladze, Andiloro, & Johnson, 2016), are often advised by defense counsel to accept these less favorable deals (Edkins, 2011), and are warier of accepting plea offers (Albonetti, 1990). Younger defendants (particularly juvenile defendants) are often overly influenced by short-term benefits of pleading (Daftary-Kapur, & Zottoli, 2014; Redlich & Shteynberg, 2016), and are more likely to follow counsel advice to plead (Redlich & Goodman, 2003; Viljoen, Klaver, & Roesch, 2005). Research on other factors: mental health, cognitive disability status (Redlich et al., 2010), gender (Redlich et al., 2010; Viljoen et al., 2005), and education (Viljoen et al., 2005; Redlich et al., 2010) is also emerging, but is in its infancy.
To-date, no published study has examined the influence of personality structure or belief systems on plea decision-making. Nonetheless, relevant theory suggests that differences in belief systems (e.g. the belief in a just world) and personality traits (e.g. neuroticism, openness to experience) could affect defendants’ decisions (Appelt, Milch, Handgraaf, & Weber, 2011). As noted earlier, belief in a just world – a cognitive bias positing that people get what they deserve – is believed to underlie the phenomenology of innocence (Kassin, 2005). Similarly, higher neuroticism has been associated with lower propensity for risk taking (i.e. loss aversion), while openness to experience was predictive of a higher propensity for risk taking (Lauriola & Levin, 2001).
Several studies have also sought to describe defendants’ self-reported reasons for accepting plea agreements. In describing the conclusions of these studies, it is important to distinguish among studies that asked convicted individuals to reflect on their real decisions, and those that surveyed research participants who responded to vignettes (but who never faced real criminal charges). In the former, prosecutorial pressure (Bordens & Basset, 1985; Malloy, Shulman, & Cauffman, 2014), expediency (Bordens & Basset, 1985, Redlich et al., 2010), perceived likelihood of conviction and subsequent sentence (for adults, Albonetti, 1990; but not adolescents, Viljoen et al., 2005), indirect pressures (e.g., family suffering; Bordens & Basset, 1985), and remorse were often reported by defendants. In addition, some false plea defendants have cited the perception that their situation was hopeless, or said they pled to protect another person (Redlich et al., 2010).
In experimental research, few studies have directly examined participant-defendants’ reasons for pleading guilty. Consequently, many of the rationales reported by defendants in field studies have yet to be observed in experimental research. Instead, studies have highlighted participant-defendants’ impressions of the strength of the case against them, their fear of a conviction, and the overall likelihood of conviction as factors influencing plea outcomes (Bordens, 1984; Helm & Reyna, 2017; McAllister & Bregman, 1986; Redlich & Shteynberg, 2016; Zimmerman & Hunter, 2018). Similarly, recent research has demonstrated the influence of perceived fairness of the plea offer on the likelihood of pleading (Redlich & Shteynberg, 2016; Tor et al., 2010). In these experiments, guilty participants often perceived plea offers as inherently fairer than did innocent participants (Redlich & Shteynberg, 2016; Tor et al., 2010). Given the difficulty in separating factually-innocent and guilty defendants who enter plea agreements, as well as quantifying their conviction probabilities, we are also unable to evaluate the influence of these factors on defendants’ decisions to plead in the real world. Our study attempts to bridge this gap, by evaluating the influence of individual differences as well as guilt status, conviction probability, and trial penalty on the rationales participant-defendants provide for their real-stakes plea decisions.
The Current Research
The current research presents a situation in which participants believe they face actual consequences (Russano et al., 2005). Participants were asked to engage in a problem-solving task with an experimental confederate. Half of the participants were induced to cheat on an individual problem and half were not. All participants (whether innocent or guilty) were later accused of cheating and warned of the seriousness of the accusation. They were faced with a choice—agree to work twenty hours in the research lab over the next month (accept the plea deal) or risk a more severe consequence if charged later (reject the plea deal).
The plea offer used in the current research is analogous to real-world pleas that include community service; these deals have become more common as a method of diverting offenders from overcrowded prisons (Subramanian, Moreno, & Broomhead, 2014). Accepting the plea in our experiment was costlier (20 hours of work) than in other plea studies that have used a real-stakes paradigm. Dervan and Edkins (2013) and Gregory et al. (1978, Exp. 2) offered participants a plea in which they would forgo their compensation for participating in the study. Russano et al (2005) allowed participants to keep study credits but required attendance at another study session.
To address gaps in existing research, we examined several individual difference variables, such as belief in a just world, as well as the Big Five personality traits, to see if they had a moderating impact on plea outcomes among the innocent versus the guilty. We also collected participants’ rationales for their plea decisions.
Trial penalty and probability of conviction manipulations. In Experiment 2, we added two additional manipulations: we manipulated the trial penalty and the probability of conviction. We use the term trial penalty (rather than plea discount) because participants were faced with one of two probabilistic punishments if the plea was rejected and they were found guilty later. Thus, the manipulation impacted the severity of the punishment participants would face if convicted after a “trial”. Probability of conviction was manipulated by telling participants that their likelihood of being found guilty of cheating (if they rejected the plea) was either somewhat likely (25%) or extremely likely (80%). These numbers were chosen to maximize the difference between the two conditions while also preserving the plausibility of the manipulation. Hence, Experiment 2 conditions were designed to test for main effects and interactive effects of the trial penalty, probability of conviction, and guilt status within a real-stakes paradigm. In both studies, we predicted that although guilty participants would accept plea deals at a higher rate than innocent participants, a significant proportion of the innocent would nonetheless accept plea deals. In Experiment 2, we predicted that an increase in conviction probability, as well as an increase in the magnitude of the trial penalty, would increase plea acceptance (as the shadow-of-the-trial model would predict). But, we also predicted that the effect of conviction probability and the trial penalty would differ between the innocent and the guilty.
Participants. One hundred and sixty-five undergraduate students enrolled in introductory courses at a large Midwest university participated in this experiment in exchange for course research credit (97 females and 68 males). The participants averaged 19 years of age with a range of 18-45 years.
The complexity of this research paradigm resulted in several necessary exclusions. Twenty-three of the 165 study participants (13.9%) were omitted from all data analyses. Of these, eight were removed due to suspicion. Participants excluded due to suspicion accurately described one of two possible elimination criteria during debriefing. The criteria included: 1) the confederate-participant’s involvement with the study, or 2) the study’s purpose as examining how people would react to an accusation and subsequent deal. An additional five people in the guilty condition had to be excluded for refusing to provide the confederate with their answer, thereby making them innocent despite their random assignment to guilt. Four other people were discounted due to early suspension of the study given their evident emotional distress during the accusation process. The remaining six people were excluded because they: possessed research lab experience (n = 2),4 were non-native English speakers (n = 2), participated in a similar study (n = 1), or the experimenter made a significant error (n = 1). The final study sample was N = 142 (71 participants per experimental cell).
Materials. Many of the materials (e.g., problem solving packets, personality questionnaires, etc.) used in this study were adapted from previous research and can be made available upon request to the corresponding author. The study also included a number of demographic and individual difference measures.
Global Belief in a Just World Scale. The Global Belief in a Just World Scale is a 7-item scale; each item was presented with a 6-point, Likert-type scale ranging from 1 (strongly disagree) to 6 (strongly agree). This scale was created to measure people’s endorsement of the belief that in general, people get what they deserve (Lipkus, 1991).
Big Five-Aspect Scale. The five-factor model of personality has been validated by numerous researchers in a countless number of studies in various contexts (John, Naumann, & Soto, 2008; McCrae & Costa, 1987). More recently, researchers have identified a narrower aspect-level of personality (DeYoung, Quilty, & Peterson, 2007). Each of the Big Five personality traits subsumes two aspects: neurotic volatility, neurotic withdraw, agreeable politeness, agreeable compassion, conscientious orderliness, conscientious industriousness, extraverted enthusiasm, extraverted assertiveness, open openness, and open intelligence—this model of personality has been validated through factor analysis (DeYoung et al., 2007). The Big Five Aspect Scale is designed to measure all ten of these personality aspects. It includes one hundred items (10 items for each facet) measured on Likert-type scales from 1 (strongly disagree) to 7 (strongly agree). Given the breadth of this measure, we felt it was a good tool to begin uncovering individual characteristics that could moderate plea outcomes among a population of adults.
Procedure.The procedure utilized in this study was adapted from previous research (Russano et al., 2005). Participants were told that the researchers were interested in examining how people completed problems both individually and in teams. A confederate posing as another participant waited outside the laboratory with the real participant. After providing informed consent and completing an initial questionnaire, the experimenter5 provided the confederate and participant five minutes for a rapport-building session.
After the rapport session, the experimenter returned with two packets of individual logic problems and one packet of team logic problems. Participants were instructed to work together on the team problems only, and to solve the individual problems alone. Experimenters then left the room while the problems were solved. Participants randomly assigned to the guilty condition were induced to cheat by the confederate on the second individual problem—hereafter referred to as the triangle problem. If the participant resisted the confederate’s initial request for help on the problem, the confederate would request help up to two more times. The confederates never asked participants in the innocent condition for help on the individual problems.
After participants completed the Big Five Aspects scale, the experimenter returned, stating that an issue arose while scoring the logic problems. The experimenter asked the confederate to exit the room with her; three minutes later, she returned with the confederate and asked the participant to follow her to a separate room. While alone with the participant, the experimenter explained that the participant and the confederate had the same wrong answer on the triangle problem. She stated that such a match is statistically improbable unless the two shared answers on that problem—a violation of study instructions.
The experimenter told participants that the professor in charge of the study (who was not present for the session) had been contacted to determine how to proceed. She then revealed that the suspected conduct could be considered academic dishonesty. Once the seriousness of the situation had been explained, she stated that the professor wanted the situation to be remedied in some way. To ensure the participant fully understood the impact of cheating (and the importance of following study instructions), the professor requested that s/he be asked to work in the lab for 20 hours.
The information participants were provided presented them with two basic options:
Option 1: Sign a statement affirming your agreement to work in the lab for 20 hours over the next four weeks and the accusation will be dropped
Option 2: Refuse to sign the statement and face a possible charge of academic dishonesty through the Dean of Students Office
The experimenter then handwrote a statement for participants to sign acknowledging their acceptance of the agreement. The statement said, “I agree to work 20 hours on the Problem Solving with Personality study by (one month after that day’s date).” If participants did not sign the statement after it was placed in front of them, the experimenter waited and reiterated the request that they sign up to two more times (to mimic the behavior of prosecutors). The experimenter then exited the room to update the professor.
After the experimenter returned with a final questionnaire, participants were administered a funnel debriefing in which they were asked a number of questions while being gradually probed for suspicion as the deception in the study was progressively explained. Participants were initially told that the professor provided these questions to the experimenter to gain a better understanding of the situation. Upon leaving, all participants were provided with information regarding on-campus student counseling services in case the study caused lingering stress.
Plea outcomes. If plea-bargaining is an effective justice-system tool, then the rates of plea acceptance should draw a clear line separating the innocent from the guilty. And, in fact, most guilty participants (80.3%) did accept the plea offer, 95% CI [69.6%, 87.9%]. Moreover, guilty participants accepted the plea deal at a reliably higher rate than did innocent participants, P1 – P2= 28.2%, 95% CI [12.7%, 41.9%]. This finding replicated the pattern found in other high-stakes plea studies (Dervan & Edkins, 2013; Wilford & Wells, 2018). Innocent participants were also more likely to accept the plea offer than to reject it (at a rate of 52.1%, 95% CI [40.7%, 63.3%]); this proportion is much higher than the ideal 0% of innocent pleas, P1 – P2= 52.1%, 95% CI [39.6%, 63.3%]. It is also on par with Dervan and Edkins (2013; 56.4%) and Wilford and Wells (2018; 40.7%), though distinct from Gregory et al. (1978, Exp. 2) who recorded no false guilty pleas (though they only had 8 innocent participants).
Plea decision rationales. Participants were subsequently asked to report why they chose to reject or accept the plea deal. Their responses were coded into categories and are described in Table 1. To test whether the pattern of responses differed significantly between the innocent and the guilty, we conducted two omnibus chi-square analyses. Prior to these tests, categories with expected count totals of fewer than five had to be collapsed with the “Miscellaneous” category. The top four reasons participants provided for accepting the plea were preserved, and all other responses were joined with Miscellaneous responses. Miscellaneous answers varied greatly but included responses such as: “I cannot prove I am innocent”, and “I was confused”, or “I don’t know why I signed”. The top four reasons were characterized as: to avoid worse consequences, it was the best alternative, felt trapped/pressured, and because of actual guilt. The chi-square analysis indicated that the pattern of these responses differed somewhat between the innocent and the guilty, although this difference did not reach statistical significance, X2(4, N =80) = 8.57, p = .073, V = .32. The differences appeared to be driven primarily by the fact that innocent participants provided far more Miscellaneous responses than guilty participants. Guilty participants also, unsurprisingly, cited their own guilt more frequently than innocent participants.
Because fewer participants rejected the plea, only three reasons for rejecting were maintained, and all other response categories were lumped with Miscellaneous answers. Participants’ Miscellaneous reasons for rejecting the plea included things such as doubting the seriousness of the claim as well as believing the punishment was unfair. This analysis provided no indication that the reasons for rejecting a plea differed between the innocent and the guilty, X2(3, N =47) = 1.14, p = .768, V = .15. The predominant reason for rejection in both groups was innocence—a significant proportion of guilty individuals did not label their behavior as cheating, and consequently, cited their innocence. In sum, it appears that the innocent and the guilty differed somewhat in their reasons for accepting a plea, but were similar in their reasons for rejecting a plea.
Individual differences. A hierarchical logistic regression was conducted with each individual difference measure to determine whether it interacted with guilt-innocence to moderate the rates of plea acceptance. Each individual difference variable was measured with multiple items (as described in the Materials)—responses to these items were averaged to produce a more reliable measure (refer to Table 2 for descriptive statistics and reliability for each individual difference measure). Step one of each regression included both guilt-innocence and the theoretically-relevant individual difference measure (e.g., belief in a just world, each Big Five Aspect). Entering both variables at step one of the analyses helped to ensure that any covariance of the two (by chance) would be excluded from the model. Step two included the interaction variable, which was computed by multiplying guilt-innocence and the individual difference variable.
Due to the novelty of these analyses, no correction for multiple comparisons was applied—to ensure that all potentially interesting interactions among plea outcomes and these individual difference variables were preserved for further examination in Experiment 2. That said, if applying a strict correction for multiple comparisons (e.g., Bonferroni), none of the individual difference measures significantly moderated plea outcomes. If applying the standard α-value < .05 criterion (without a post-hoc correction), only two individual difference variables emerged as having a potentially moderating effect on plea outcomes. Nevertheless, in light of the multiple comparisons, interpretations of these analyses must be made with caution.
Strong endorsement of belief in a just world was related to plea outcomes among the guilty, but not the innocent, exp(β) = .27 [.08, .89], Wald = 4.63, p = .031 (see Figure 1). The main effects of guilt status and belief in a just world were not significant. This moderating pattern remained significant when controlling for Big Five Aspects, exp(β) = .27 [.08, .90], Wald = 4.56, p = .033. Guilty participants who possessed strong just-world beliefs were less likely to reject the plea than guilty participants who did not possess strong just-world beliefs; belief in a just world had no significant impact on plea outcomes among the innocent.
There was a main effect of intelligence, as measured as an aspect of openness, exp(β) = 3.80 [1.73, 8.36], Wald = 10.98, p = .001 (see Figure 2). Further, the main effect was qualified by an interaction. Innocent individuals with higher open-intelligence were generally more likely to reject the plea than innocent individuals lower in open-intelligence, exp(β) = .29 [.11, .82], Wald = 5.52, p = .019. These values did not change when controlling for other individual difference measures. Thus, higher intelligence was related to lower plea acceptance, but only for the innocent. Open-intelligence had no discernable impact on plea outcomes among the guilty.
The false plea rate in Experiment 1 was so high that guilty individuals were just 1.54 times more likely to accept a plea than innocent individuals. In fact, only a few discernable differences emerged between the innocent and the guilty across a number of measures. Logistic regression analyses demonstrated minimal effects of individual differences on plea outcomes. Specifically, high just-world beliefs corresponded to an increase in plea acceptance among the guilty, and high open-intelligence correlated with a decrease in plea acceptance among the innocent. One of the few other factors found to potentially separate the innocent from the guilty was an omnibus chi-square analysis indicating that the innocent and the guilty are driven by different factors when accepting a plea, though the analysis was not significant. The scant impact of individual differences overall is not too surprising in light of the situational context. Social psychologists have long acknowledged how the power of the situation can mask innate individual differences (Haney & Zimbardo, 2009).
Experiment 2 was designed to build on the results of Experiment 1. In addition to testing the replicability of the previous effects, Experiment 2 incorporated two additional manipulations. Specifically, we examined the impact of altering the magnitude of the trial penalty as well as manipulating participants’ beliefs about the probability of conviction if they were to reject the plea offer. These two manipulations represent core elements in the shadow-of-the-trial model (Bibas, 2004; Bushway & Redlich, 2012; Bushway et al., 2014). The predictions were that (a) increasing the perceived probability of conviction at trial should increase plea acceptance, and (b) increasing the trial penalty should increase plea acceptance. Moreover, we were also interested in how the guilt/innocence manipulation, which is fully crossed with the probability-of-conviction and trial-penalty manipulations in Experiment 2, might moderate the impact of these manipulations.
Determining whether the trial-penalty manipulation and the probability-of-conviction manipulation are different in magnitude for the innocent than for the guilty is very important. The shadow-of-the-trial model, however, is silent on the question of what role guilt or innocence might play. Nevertheless, it is quite possible that the innocent and the guilty perceive these variables differently, because their own guilt or innocence could provide different frames of reference for thinking about the problem. For instance, innocent individuals could perceive potential punishments as more severe than guilty individuals due to their belief that they are undeserving of any punishment. Beliefs like the phenomenology of innocence (Kassin, 2005) could also cause the innocent to perceive their likelihood of being convicted as lower than guilty individuals. Further, our aims were to (a) determine whether the overall plea acceptance rates observed in Experiment 1 would replicate, (b) determine whether the innocent and the guilty are truly motivated by different factors when accepting a plea deal, and (c) test the reliability of the significant moderating effects of belief in a just world and open-intelligence found in Experiment 1. All of the Experiment 1 individual difference measures were included in Experiment 2, although we only report analyses for the measures that previously produced significant effects. We focus only on the replicable effects, because the inclusion of multiple variables (especially without a priori hypotheses) can often produce false positives. Thus, we felt any significant results of the individual difference variables should be found in both experiments to be considered reliable.
Participants. Three hundred and seventy-nine undergraduate students at a large Midwest university received research credits in exchange for their participation in the study. The study sample was 61% female. Participants ranged from 18 to 28-years-old with a mean age of 19 years.
Experiment 2 employed the same exclusion criteria as Experiment 1. Forty total participants (10.6%) were omitted from subsequent data analyses. Of these, the most common reason for exclusion was early suspension of a study session due to emotional distress (n = 16). Nine others were excluded for resisting the confederate’s requests for help, despite their random assignment to guilt. Six participants were omitted because they were not native English speakers (an eligibility requirement). Only two participants were excluded due to their suspicion of the study (using the same standards as Experiment 1). The remaining participants were excluded due to: possessing advanced research experience (n = 4), experimenter error (n = 2), or prior experience in a similar study (n = 1). The final study sample was N = 339 with a range of 38 to 49 participants per experimental condition.
Design. This study employed a 2 (innocent or guilty) x 2 (trial penalty: less strict or more strict) x 2 (probability of conviction: somewhat likely or extremely likely) between-participants design.
Materials. The materials used in this experiment were identical to those in Experiment 1 (i.e., Big Five Aspects Scale, Global Belief in a Just World) with the exception of a few additional post-manipulation questions to examine the effects of the probability and discount manipulations.
Procedure. The procedure for Experiment 2 differed from Experiment 1 in four ways. First, the rapport-building session was shortened from five to three minutes to provide the extra time needed for the additional study manipulations. Second, the authority in charge of arbitrating the accusation (if the plea was rejected) changed from the Dean of Students Office to the Department of Psychology’s Human Research Ethics Review board. This board was fictitiously described as having procedures for determining when students are guilty of cheating in research studies, as well as possessing the authority to impose sanctions on students found guilty. This change was made due to some particularly emotional reactions to the possibility of punishment from the Dean of Students Office—an office that many students associate with serious cases. Third, participants faced one of two possible consequences if they rejected the plea and were found guilty of cheating later. Participants in the less-strict-penalty condition were told that they may be required to write a 15-page APA-formatted research paper on the ethics of research. Participants in the more-strict-penalty condition were told that they may face indefinite academic probation and an F on the research portion of their grade (in whatever class to which the research credits were designated to apply); in most courses, this punishment would lower their overall grade by 5-10%. Within the university code of conduct regarding academic dishonesty, completing assignments that highlight the misconduct are common sanctions for those found guilty of academic dishonesty. These sanctions are also typically considered the least severe, among a range of outlined sanction options, including “conduct probation”. Although listed as the second-to-least severe, it was the strictest of the enumerated sanctions that would not result in further increases in participant distress, and thus, in further exclusion of participants. The final difference was that participants were provided with estimates on the likelihood that they would be found guilty of cheating. Specifically, experimenters reported that the professor estimated, based on his experience dealing with similar cases, that the odds of being found guilty (probability of conviction) were either somewhat likely (around 25%) or extremely likely (around 80%).
Plea outcomes. To ensure that all potential statistical interactions were captured, we performed a logistic regression analysis for which participants’ plea acceptance or rejection was the outcome of interest, and their guilt status as well as the probability of conviction and trial penalty were treated as potential modifiers. More specifically, we used a hierarchical binary logistic regression model in which main effects were entered on the first block, two-way interactions on the second block, and the three-way interaction was entered on the third and final block. The presence of higher-order interaction terms renders all lower-order terms conditional main effects, necessitating the hierarchical structure of effects (see Table 3 for the proportions of plea acceptance in all eight experimental conditions).
Neither the three-way interaction nor any of the two-way interactions were significant, Bs < 0.51, SEs > 0.49, Wald’s χ2(1) < 0.88, ps > .35. As in Experiment 1, guilt status had a significant impact on plea outcomes with guilty participants being more likely to plead than innocent participants, B = 1.11, SE = 0.25, Wald’s χ2(1) = 19.53, p<.001, eB = 3.04 (95% CI [1.86, 4.99]). The odds of a plea were 3.04 times more likely from guilty participants (3.89:1) than from innocent participants (1.38:1).6 In addition, the probability of conviction also significantly influenced participants’ plea decisions, B = 0.55, SE = 0.25, Wald’s χ2(1) = 5.02, p = .03, eB = 1.74 (95% CI [1.07, 2.82]). The odds of a plea were 1.74 times greater when the probability of conviction at trial was high (2.84:1) compared to when the probability of conviction at trial was low (1.74:1). The impact of the magnitude of the trial penalty on plea outcomes was not significant, though this could be explained by participants’ unexpected perceptions of the severity of the two punishments. The plea acceptance rate among the innocent was 58.1%, 95% CI [50.7%, 65.3%], and among the guilty the plea acceptance rate was 80.2%, 95% CI [73.6%, 85.6%]. These proportions are extremely close to those found in Experiment 1.
Probability of conviction. Although the two-way interaction was not significant, plea acceptance outcomes produced by the probability-of-conviction manipulation followed a pattern suggesting that innocent participants were more influenced by this manipulation than the guilty participants, as predicted (B = -0.47, SE = 0.51, Wald’s χ2(1) = 0.88, p =.35, eB = 0.62, 95% CI [0.23, 1.68]). Examination of the raw proportions indicates that the likelihood of conviction had a negligible impact on guilty participants’ willingness to accept the plea with those being told conviction was somewhat likely (collapsing data from the two trial penalty conditions) accepting the plea deal 78.2% of the time, and those told that their chances of conviction were very likely accepting 82.5% of the time, P2 – P1 = 4.3%, 95% CI [-7.9%, 16.2%]. In contrast, the probability of conviction had a significant impact on the innocents’ propensity to accept the plea deal (collapsing across the trial penalty conditions) with 49.4% accepting in the somewhat likely condition and 67.1% accepting in the extremely likely condition; P2 – P1 = 17.6%, 95% CI [2.9%, 31.3%].
The trial penalty. Although the trial penalty manipulation did not reliably impact plea outcomes, its potential impact was more apparent among the innocent than the guilty; innocent: P2 – P1 = -6.6%, 95% CI [-20.8%, 0.8%] versus guilty: P2 – P1 = -2.9%, 95% CI [-14.8%, 9.4%], respectively. The effect among the innocent, although not significant, is the opposite of what was hypothesized. Namely, innocent participants in the less-strict penalty condition were actually more likely to accept the plea than were innocents in the more-strict penalty condition. In other words, the stricter sanction seemingly reduced false guilty pleas.
Manipulation checks. To check the validity of the trial penalty manipulation, participants were asked, “How severe did you feel the possible consequences of not agreeing to the statement were?” Responses were measured on a 5-point Likert scale from 1 (Not at all severe) to 5 (Very severe). Interestingly, participants in the less-strict condition actually produced a mean rating of 2.64 (SD = 1.29) on a 5-point Likert scale whereas participants in the more-strict condition produced a mean rating of 1.75 (SD = 1.07); this difference was significant; t(337) = 6.95, M1 – M2 = 0 .89, 95% CI [.64, 1.14], p < .001. Thus, the previous less- and more-strict designations were perceived by participants such that the less-strict condition (15-page paper) was perceived as stricter and the more-strict condition (fail research component and indefinite academic probation) was perceived as less strict (contradicting the characterization of these penalties in the university codes of conduct). Because neither penalty exceeded the midpoint, it appears that participants did not perceive either punishment to be particularly severe. Guilt status also seemed to impact perceptions of the trial penalty such that the guilty perceived the potential penalty as less strict (M = 2.05, SD = 1.16) overall than did the innocent (M = 2.32; SD = 1.34); t(337) = 2.00, M1 – M2 = 0 .27, 95% CI [.00, 0.54], p = .047. Thus, perceived severity of the trial penalty was colored both by the penalty itself as well as the knowledge that one is innocent or guilty and, consequently, more or less deserving of punishment.
To ensure that the probability of conviction manipulation was effective, participants were asked, “Given the evidence in the current situation [if you hadn’t signed the statement], how likely is it that you would have been charged with cheating and consequently lost your research privileges [been required to write the 15-page research paper on the ethics of research]?” with variations depending on the trial penalty condition and whether the participants had agreed to sign the statement. Responses were made with a Likert-type scale from 1 (extremely unlikely) to 10 (extremely likely). The probability of conviction manipulation did have the expected effect; overall, participants in the extremely likely condition generally reported higher values for their perceived probability of conviction (M = 4.31; SD = 3.22) than those in the somewhat likely condition (M = 3.52; SD = 2.75); t(333) = 2.42, M1 – M2 = 0.79, 95% CI [0.15, 1.43], p = .016. Guilt and innocence also had an impact on the perceived probability of conviction such that the guilty reported a higher likelihood of conviction (M = 4.70; SD = 3.08) than the innocent (M = 3.14; SD = 2.74); t(333) = 4.89, M1 – M2 = 1.56, 95% CI [0.93, 2.18], p < .001. This is noteworthy considering that all participants were presented with the same evidence; namely, that they had produced the same wrong answer on a particular problem.
Participant Rationales. As in Experiment 1, after choosing to accept or reject the plea deal, participants were asked to explain their choice. This question produced more variability than in the previous experiment, which precludes us from making definitive conclusions (given the limited number of participants in certain cells; see Table 4). However, we again conducted omnibus chi-square analyses after collapsing any response types with total expected counts of fewer than five into the Miscellaneous category. The analysis of reasons for accepting the plea included nine total response categories: to avoid worse consequences, it was the best alternative, felt trapped/pressured, felt confused or did not know, seemed like the right thing to do, could not prove innocence, to conclude the situation, and because of actual guilt. Many of these response categories had been originally observed in Experiment 1 but were collapsed into the Miscellaneous category due to smaller expected cell counts. The chi-square analysis showed that there was a significant difference in the responses provided by the innocent versus the guilty for choosing to accept the plea bargain, X2(8, N = 233) = 20.31, p = .009, V = .29. This finding does seem to indicate that the pattern in Experiment 1 might have achieved significance if additional participants had been included. However, the pattern of this effect differed somewhat from Experiment 1; instead of producing more variability in their responses, innocent individuals were more likely to feel trapped into accepting the plea. Though, as in Experiment 1, guilty participants regularly cited their own guilt as a reason for accepting the plea whereas innocent individuals never provided this reason. The chi-square analysis examining reasons for rejecting the plea agreement included five response categories. This analysis was again not significant, X2(4, N = 105) = 6.24, p = .182, V = .24. In sum, as in Experiment 1, innocent participants provided a different pattern of responses for accepting the plea than guilty participants, but reasons for rejecting the plea were relatively similar between the two groups.
Individual differences. Hierarchical logistic regressions were conducted to determine whether the individual difference variables that produced significant effects and interactions with guilt-innocence to moderate the rates of plea acceptance in Experiment 1 would be replicated in Experiment 2. The parameters of these logistic regressions were the same as in Experiment 1.
None of the previously significant main effects or interactions from Experiment 1 were replicated in Experiment 2. This lack of replication occurred despite both measures maintaining high reliability (belief in a just world: α = 0.70; intelligent aspect of openness: α = 0.83). The failure to replicate these effects seems to lend further credence to the idea that the plea situation is so compelling that it erodes the effect of individual differences—at least among student adults. It is also possible that the strength of the phenomenology of innocence or guilt overrides the effect of individual differences. Regardless, even if the disappearance of these effects was due to the introduction of the probability of conviction and trial penalty manipulations, this still indicates that these effects are easily minimized by factors that would typically accompany a plea situation.
Overall, many of the findings from Experiment 1 were replicated in Experiment 2. First, an alarming proportion of innocent participants still agreed to accept the plea deal. In both experiments, the overall rate of false guilty pleas exceeded 50%, and the diagnosticity of a guilty plea failed to exceed 1.5. Second, the pattern of responses for accepting the plea differed significantly between the innocent and the guilty—this finding replicates the trend from Experiment 1. Reasons for rejecting the plea, on the other hand, did not differ significantly between the innocent and the guilty in either experiment. Although it remains possible that there are some differences underlying the innocent versus the guilty person’s decision to reject a plea, the current results suggest that these differences are not as robust as those that drive the innocent versus the guilty person’s decision to accept a plea. Thus, the only category in which the results of Experiment 2 differed from those of Experiment 1 was individual differences. None of the previously significant individual difference main effects or interactions were replicated in Experiment 2.
Experiment 2 also extended the findings of this research by examining the impact of two new variables. Probability of conviction significantly impacted the proportion of plea acceptance among the innocent, although the two-way interaction was not significant, with no discernable impact on the guilty. These findings indicate that the innocent may be more influenced by the probability of conviction. The effect of the trial penalty, on the other hand, was unclear due to notable differences between participants’ perception of the severity of the penalties, which contradicted how these punishments are characterized in muniversity codes of conduct.
The current research showed that not only can innocent individuals accept plea deals, but that they can do so at a rate that renders the outcome (accept versus reject) one of low diagnosticity. Of course, diagnosticity is also impacted by base rates of guilt versus innocence, which were likely substantially different in this experiment than in the real world. Participants’ reasons for accepting the plea did differ between the innocent and the guilty, but reasons for rejecting the plea did not. An exploratory analysis of individual difference variables initially produced preliminary support for a moderating impact of belief in a just world and the intelligence component of openness on plea outcomes. However, neither of these effects were replicated in Experiment 2. Hence, overall, there is not much evidence that the individual differences we examined had an impact on participants’ decisions to accept or reject a plea offer.
Plea likelihood. “Ninety-seven percent of federal convictions and ninety-four percent of state convictions are the result of guilty pleas” (Lafler v. Cooper, 2012, p. 11). Given this domination of the justice system, the results of the current research merit some concern. Whereas we uncovered at least one potential method of increasing false guilty pleas, neither of the manipulations impacted true guilty pleas. Increasing the reported probability of conviction increased the proportion of false guilty pleas (collapsing data from the two trial penalty conditions); this effect did not extend to true guilty pleas. However, true plea rates among the guilty were already quite high, introducing the possibility of a ceiling effect, which may also exist in the real world.
Although our conviction-probability manipulation might appear overly simplistic, it is important to note that the extremely-likely condition rate was in the range of real trial conviction rates (the combined federal-level jury trial and nonjury trial conviction rate in 2012 was 83.2%; Bureau of Justice Statistics, 2015). Further, providing an estimate for the probability of conviction did not preclude research participants from drawing their own conclusions regarding their perceived probability of conviction. There are a number of factors, beyond reported probabilities, that could influence participants’ and defendants’ perceived likelihood of conviction. This assumption was clearly demonstrated by the significant effect of guilt status on participants’ perceived probability of conviction. Despite being provided an explicit probability figure, participants’ perceptions of probability were still susceptible to other influences.
The current real-stakes experiment produced results that generally resemble those obtained in hypothetical scenario studies (e.g., Bordens, 1984). While the exact points at which willingness to accept the plea did not replicate, the tendency for the innocent to be less likely to accept a plea until the probability of conviction approached near-certainty, as well as the tendency for the guilty to accept a plea regardless of probability, did. Similarly, Tor et al. (2010) found that participants asked to assume innocence were less likely to reject pleas as probability of conviction exceeded 50%; participants asked to assume guilt maintained steady levels of plea acceptance with little impact of the probability manipulation.
In sum, the probability of conviction appeared to influence plea outcomes, but only among the innocent. Further, innocent individuals reported lower perceived probabilities of conviction than guilty individuals. Thus, prosecutors who attempt to secure more pleas by citing the average probability of conviction, or their personal conviction rate, might be doing more harm than good. The current research indicates that such a maneuver is more likely to impact the innocent than the guilty, because the innocent’s perceived conviction probabilities will be relatively lower. Although one could argue that prosecuting attorneys are more likely to interact with guilty defendants than innocent ones (unlike in the current research), our results still indicate that innocent individuals will be more vulnerable to these types of strategies given their relatively lower expected probability of conviction. Consequently, employing this strategy could increase total plea convictions—unfortunately, this increase could stem from a growth in false guilty pleas rather than true guilty pleas.
The trial penalty. We did not find a reliable effect of the trial penalty on plea outcomes among the guilty or the innocent. Follow-up analyses revealed that the distinction between our trial penalty conditions was not as great as we had anticipated when designing the study. Our manipulation was a consequence of thorough discussion regarding what punishments could plausibly be administered after a cheating charge of this nature along with an examination of university codes of conduct. Unfortunately, college students despise writing far more than we had predicted. That said, further research should better assess the impact of the trial penalty by systematically manipulating the penalty’s magnitude in a significant quantitative (rather than qualitative) manner. Dervan and Edkins (2013) followed this formula by telling participants they could be required to attend an ethics seminar or course that met three hours per week for three week or an entire semester. Notably however\, they also failed to find a significant effect of their trial penalty manipulation.
Whereas the complexities with the trial-penalty manipulation render its potential magnitude unclear, it is clear that further research in this area is needed. With the growing prevalence of mandatory minimum sentences, discrepancies between the penalties offered during pleas and the ones threatened at trial can be dramatic (U.S. Sentencing Commission, 2011). Further, our finding that perceptions of severity regarding the trial penalty differ between the innocent and the guilty unveils the possibility of an even greater innocence effect (Kassin, 2005). Prosecutors attempting to secure pleas by increasing the magnitude of the trial penalty could be dramatically increasing the innocent’s perceptions of the penalty’s severity, but only slightly altering the guilty’s perceptions of severity. In fact, the debate between the plea discount and trial penalty label could depend on the guilt status of the defendant. Guilty defendants could be more prone to viewing pleas as a method of securing discounts on the sentence they would otherwise serve; innocent defendants, on the other hand, could view any increase in the punishment they face as a penalty for rejecting a plea offer. Thus, increasing trial penalties or plea discounts could have little effect on pleas among the guilty, but could have a profound impact on the innocent.
Plea decision making. The combined results of the current research highlight weaknesses of the shadow-of-the trial model. The significant differences in the rates of plea acceptance between the innocent and the guilty (despite the probability of conviction and trial penalty being held constant [Exp. 1] or manipulated equally across conditions [Exp. 2]) cannot be accounted for by the current shadow-of-the-trial model. The current model does not account for a predictive effect of guilt status. Further, the difference in the impact of conviction probability on the guilty versus the innocent could not be explained by the current model, which is neutral with regard to actual innocence. In other words, guilt or innocence should only impact plea outcomes when the differences in status correspond to differences in the probability of conviction. For instance, in cases in which a guilty defendant faces a low probability of conviction, the shadow-of-the-trial model would predict that defendant would reject plea offers unless the offer was extremely favorable. However, we did not find that conviction probability influenced plea decisions among the guilty, only the innocent.
Thus, perhaps the shadow-of-the-trial model is missing the same component that normative models of decision-making missed for a number of years—the importance of reference points (Tversky & Kahneman, 1981). It is possible that innocent individuals view punishment qualitatively differently than guilty individuals. The finding that innocent participants viewed the potential punishments as more strict than guilty participants could illustrate their tendency to not only perceive sentence disparities (between the plea offer and potential trial sentence) as penalties, but more broadly, as losses. Whereas guilty individuals might perceive these offers as providing discounts that confer gains to the punishment they are aware they deserve for their actions. Thus, innocent participants might be more resistant to accepting pleas due to the general aversion people have to loss. Further research should examine the impact of plea-relevant variables on both the innocent and the guilty.
Participant-defendant rationales. The data also indicated that innocent and guilty individuals are driven by similar factors when rejecting a plea. In contrast, reasons for accepting the plea differed more substantially between the innocent and the guilty in Experiment 1 and significantly so in Experiment 2. However, it is important to note that this pattern could also be due to the variations in samples. It is possible that the reasons for rejecting a plea did not differ significantly between the innocent and guilty because there were fewer overall participants included in the analysis. In both experiments, the total number of participants who accepted the plea significantly outnumbered those who rejected. Regardless, results seem to indicate that there is more variation in reasons for accepting a plea than there is for rejecting a plea. This finding could have interesting implications from a legal reform standpoint.
The pattern of responses participants provided for their plea decisions converged well with Bordens and Bassett’s (1985) taxonomy of factors representing reasons why people plead guilty. They interviewed actual offenders and defendants and found that many reported being driven by remorse, sentence-related motivations, expediency, and the likelihood of conviction. In their conclusion, Bordens and Bassett (1985) argued that all of the factors they identified followed a central theme—pressure. Other studies have echoed this sentiment with Viljoen et al. (2005) and Redlich et al. (2010) also finding pressure to be a significant factor in people’s decision to plead guilty. Our results showed similar reasons (e.g., remorse: guilty, sentence-related motivations: avoiding worse consequences, expediency: to conclude situation, etc.) which could also fall under the theme of pressure.
Individual differences. Although none of the individual difference variables moderated plea diagnosticity consistently, these null effects are nonetheless valuable to understanding plea-bargaining. Namely, that the power of the situation may be so strong as to overcome many of the behaviors that would typically be endorsed by people high in certain traits. In other words, the effect of individual differences such as belief in a just world, intelligence, neuroticism, or agreeableness might be overshadowed by the power of the pressure to plead guilty. It is also possible that the phenomenology of innocence/guilt overshadows other individual differences. Thus, the impact of individual differences within average adult ranges, might be minimal. This finding is in line with other research that has demonstrated weak or mixed effects of gender (Davidson & Rosky, 2015; Sommers, Goldstein, & Baskin, 2014; Viljoen et al., 2005), cognitive ability or years of education (Viljoen et al., 2005; Redlich et al., 2010), and the disappearing (or, at least shrinking) effect of age as participants reach adulthood (Redlich et al., 2010; Viljoen et al., 2005). Future research should continue measuring the impact of individual differences across legally-relevant variables (e.g., juvenile status, mental health, etc.).
Limitations. The penalties that face criminal defendants who are offered a plea deal are far harsher than working twenty hours in a research lab. However, the same could be said for the penalties that those same defendants may face if convicted at trial. Thus, these two limitations should theoretically push the plea acceptance rate in opposite directions and cancel each other out (to a degree). The more severe the penalty for accepting the plea, the lower the plea rate, but the more severe the penalty faced if convicted at trial, the higher the plea rate. Further, the plea deal offered in this study, twenty hours of lab work, imposed the harshest penalty in a plea simulation study to-date. But, we did not alter the terms of the deal if participants rejected the initial offer—in other words, the bargaining component of the plea process was not represented in the current research. Though some have argued that the term plea-bargaining is a misnomer anyway—defendants have such little power and so few resources that prosecutors often dictate terms that can be accepted or will be withdrawn (Rakoff, 2014).
The timeline for the current research was much shorter than real-world plea procedures. Suspects are typically not offered a plea deal until they have been formally charged with a crime and questioned by law enforcement. Thus, the amount of time between the formal accusation and the initial plea offer can vary from days to months. A number of ethical considerations precluded us from extending the study timeline. It is, therefore, currently unclear whether reducing the timeline increased the pressure to plead, and thus, false guilty pleas. However, there are reasons to believe that reducing the timeline could decrease false guilty pleas. Prolonged pre-trial detention (due to unaffordable or denied bail) reportedly increases the likelihood of pleading (Rakoff, 2014). Further, real world cases have also involved short-term plea deals (e.g., 24-hour only offers; also known as exploding offers) meaning that time pressure can be as much a factor in the real world as it was in the current research (Gross, 2015; Zottoli, Daftary-Kapur, Winters & Hogan, 2016).
Conclusion. In Lafler v. Cooper (2012) , the Court ruled that, due to the ubiquity of plea-bargaining and its growing role in due process, defendants who have secured legal counsel possess a Constitutional right to that counsel performing effectively amid the plea bargain process. “… the right to adequate assistance of counsel cannot be defined or enforced without taking account of the central role plea bargaining plays in securing convictions and determining sentences” (Lafler v. Cooper, 2012, p. 11). This decision, as Justice Antonin Scalia wrote for the dissent, “… opens a whole new field of constitutionalized criminal procedure: plea-bargaining law…” (Lafler v. Cooper, 2012, p. 1 of dissent). Researchers should ultimately reveal what policies can protect the innocent from plea convictions.
The current research has demonstrated that plea outcomes among the innocent are more easily impacted than plea outcomes among the guilty. Any impact of the probability of conviction and trial penalty manipulations on the guilty was negligible. In contrast, the probability of conviction had a significant impact on the proportion of innocent pleas. This finding is extremely important from a policy standpoint. It demonstrates that plea-bargaining reform could significantly reduce the number of false guilty pleas without a comparable reduction in true guilty pleas. Further research should continue to demonstrate how “plea-bargaining law” can be written to preserve the process for the guilty while protecting the innocent.
Albonetti, C. A. (1990). Race and the probability of pleading guilty. Journal of Quantitative Criminology, 6, 315-334. doi: 10.1007/BF01065413
Appelt, K. C., Milch, K. F., Handgraaf, M. J., & Weber, E. U. (2011). The Decision Making Individual Differences Inventory and Guidelines for the Study of Individual Differences in Judgment and decision-making Research. Judgment and Decision Making, 6(3), 252-262.
Aronson, E., Wilson, T. D., & Brewer, M. B. (1998). Experimentation in social psychology. In D. T. Gilbert, S. T. Fiske, & G. Lindzey (Eds.), The handbook of social psychology (pp. 99-142). New York, NY: McGraw-Hill.
Bibas, S. (2004). Plea bargaining outside the shadow of a trial. Harvard Law Review, 117(8), 2463-2547. doi: 10.2307/4093404
Bordens, K. S. (1984). The effects of likelihood of conviction, threatened punishment, and assumed role on mock plea bargaining decisions. Basic and Applied Social Psychology, 5, 59-74. doi: 10.1207/s15324834basp0501_4
Bordens, K. S., & Bassett, J. (1985). The plea bargaining process from the defendant’s perspective: A field investigation. Basic and Applied Social Psychology, 6, 93-110. doi: 10.1207/s15324834basp_0602_1
Bureau of Justice Statistics. (2015). Federal justice statistics, 2012 – statistical tables (NCJ Report No. 248470). Washington D.C.: U.S. Department of Justice.
Burke, A. S. (2007). Prosecutorial passion, cognitive bias, and plea bargaining. Marquette Law Review, 91, 183-211. Retrieved from http://scholarship.law.marquette.edu/cgi/viewcontent.cgi?article=1023&context=mulr_conferences
Bushway, S. D., & Redlich, A. D. (2012). Is plea bargaining in the ‘shadow of the trial’ a mirage? Journal of Quantitative Criminology, 28(3), 437-454. doi: 10.1007/s10940-011-9147-5
Bushway, S. D., Redlich, A. D., Norris, R. J. (2014). An explicit test of plea bargaining in the “shadow of the trial.” Criminology, 52(4), 723-754. doi: 10.1111/1745-9125.12054
Daftary-Kapur, T., & Zottoli, T. M. (2014). A first look at the plea deal experiences of juveniles tried in adult court. International Journal of Forensic Mental Health, 13, 323-336.
Davidson, M. L., & Rosky, J. W. (2015). Dangerousness or diminished capacity? Exploring the association of gender and mental illness with violent offense sentence length. American Journal of Criminal Justice, 40, 353-376. doi: 10.1007/s12103-014-9267-1
Dervan, L. E., & Edkins, V. A. (2013). The innocent defendant’s dilemma: An innovative empirical study of plea bargaining’s innocence problem. Journal of Criminal Law and Criminology, 103, 1-48. doi: 10.2139/ssrn.2071397
DeYoung, C. G., Quilty, L. C., & Peterson, J. B. (2007). Between facets and domains: 10 aspects of the Big Five. Journal of Personality & Social Psychology, 93(5), 880-896. doi: doi:10.1037/0022-3522.214.171.1240
Edkins, V. A. (2011). Defense attorney plea recommendations and client race: Does zealous representation apply equally to all? Law and Human Behavior, 35, 413-425. doi: 10.1007/s10979-010-9254-0
Fisher, G. (2000). Plea bargaining’s triumph. The Yale Law Journal, 109(5), 857-1086. doi: 10.2307/797483
Gazal-Ayal, O., Turjeman, H., & Fishman, G. (2013). Do sentencing guidelines increase prosecutorial power? An empirical study. Law and Contemporary Problems, 76(131), 131-159.
Gregory, W. L., Mowen, J. C., & Linder, D. E. (1978). Social psychology and plea bargaining: Applications, methodology, and theory. Journal of Personality and Social Psychology, 36,1521-1530. doi: 10.1037/0022-35126.96.36.1991
Gross, S. R. (2015, July 24). The staggering number of wrongful convictions in America. The Washington Post. Retrieved from https://www.washingtonpost.com/opinions/the-cost-of-convicting-the-innocent/2015/07/24/260fc3a2-1aae-11e5-93b7-5eddc056ad8a_story.html
Haney, C., & Zimbardo, P. G. (2009). Persistent dispositionalism in interactionist clothing: Fundamental attribution error in explaining prison abuse. Personality and Social Psychology Bulletin, 35, 807-814. doi: 10.1177/0146167208322864
Helm, R. K., Reyna, V. F., Franz, A. A., Novick, R. Z., Dincin, S., & Cort, A. E. (2018). Limitations on the ability to negotiate justice: attorney perspectives on guilt, innocence, and legal advice in the current plea system. Psychology, Crime & Law, 1-20. doi: 10.1080/1068316X.2018.1457672
Helm, R. K., & Reyna, V. F. (2017). Logical but incompetent plea decisions: A new approach to plea bargaining grounded in cognitive theory. Psychology, Public Policy, and Law, 23(3), 367. doi: 10.1037/law0000125
Innocence Project (2017). Our impact. Retrieved from https://25years.innocenceproject.org/ impact/
John, O. P., Naumann, L. P., & Soto, C. J. (2008). Paradigm shift to the integrative Big Five trait taxonomy: History, measurement, and conceptual issues. In O. P. John, R. W. Robins, & L. A. Pervin (Eds.), Handbook of personality: Theory and research (pp. 114-158). New York, NY: Guilford Press.
Kassin, S. M. (1997). The psychology of confession evidence. American Psychologist, 52, 221-323.
Kassin, S. M. (2005). On the psychology of confessions: Does innocence put innocents at risk? American Psychologist, 60, 215-228. doi: 10.1037/0003-066X.60.3.215
Kassin, S. M. (2012). Why confessions trump innocence. American Psychologist, 67, 431-445. doi: 10.1037/a0028212
Kassin, S. M. & Norwick, R. J. (2004). Why people waive their Mirandarights: The power of innocence. Law and Human Behavior, 28, 211-221. doi: 10.1023/B:LAHU.0000022323. 74584.f5
Kramer, G. M., Wolbransky, M., & Heilbrun, K. (2007). Plea bargaining recommendations by criminal defense attorneys: Evidence strength, potential sentence, and defendant preference. Behavior Sciences and the Law, 25, 573-585. doi: 10.1002/bsl.759
Kutateladze, B. L., Andiloro, N. R., & Johnson, B. D. (2016). Opening Pandora’s Box: How does defendant race influence plea bargaining?. Justice Quarterly, 33, 398-426.
Lafler v. Cooper, 132 U.S. 1376 (2012).
Lauriola, M., & Levin, I. P. (2001). Personality traits and risky decision-making in a controlled experimental task: An exploratory study. Personality and Individual Differences, 31(2), 215-226.
Leo, R. A. (1996). Miranda’s revenge: Police interrogation as a confidence game. Law & Society Review, 30(2), 259-288. doi: 10.2307/3053960
Lipkus, I. (1991). The construction and preliminary validation of a global belief in a just world scale and the exploratory analysis of the multidimension. Personality and Individual Differences, 12(11), 1171-1178. doi: 10.1016/0191-8869(91)90081-L
Lynch, D. R., & Evans, T. D. (2002). Attributes of highly effective criminal defense negotiators. Journal of Criminal Justice, 30, 387-396. doi: 10.1016/S0047-2352(02)00153-8
Malloy, L. C., Shulman, E. P., & Cauffman, E. (2014). Interrogations, confessions, and guilty pleas among serious adolescent offenders. Law and human behavior, 38(2), 181.
McAllister, H. A., & Bregman, N. J. (1986a). Plea bargaining by defendants: A decision theory approach. The Journal of Social Psychology, 126(1), 105-110. doi: 10.1080/00224545. 1986.9713576
McAllister, H. A., & Bregman, N. J. (1986b). Plea bargaining by prosecutors and defense attorneys: A decision theory approach. Journal of Applied Psychology, 71(4), 686-690. doi:
McCrae, R. R., & Costa, P. T. Jr. (1987). Validation of the five-factor model of personality across instruments and observers. Journal of Personality and Social Psychology, 52(1), 81-90. doi: 10.1037/0022-35188.8.131.52
National Registry of Exonerations. (24 November, 2015). Innocents who plead guilty. Retrieved from http://www.law.umich.edu/special/exoneration/Documents/NRE.Guilty.Plea. Article1.pdf
Oppel, R. A. Jr. (2011, September, 26). Sentencing shift gives new leverage to prosecutors. The New York Times. A1. Retrieved from http://www.nytimes.com/2011/09/26/us/tough-sentences-help-prosecutors-push-for-plea-bargains.html?_r=0
Pan, J., & Kaiser, M. G. (2003). Preliminary proceedings: Guilty pleas. Annual Review of Criminal Procedure, 32, 1-35. Retrieved from LexisNexis.
Rakoff, J. S. (2014, November 20). Why innocent people plead guilty. The New York Review of Books, 61(18). Retrieved from http://www.nybooks.com/articles/2014/11/20/why-innocent-people-plead-guilty/
Redlich, A. D. (2010a). False confessions, false guilty pleas: Similarities and differences. In G. D. Lassiter & C. Meissner (Eds.) Interrogations and confessions: Current research, practice, and policy (49- 66). Washington, D. C.: APA Books. doi: 10.1037/12085-003
Redlich, A. D. (2010b). The susceptibility of juveniles to false confessions and false guilty pleas. Rutgers Law Review, 62, 943-957. Retrieved from http://pegasus.rutgers.edu/ ~review/vol62n4/Redlich.pdf
Redlich, A. D., & Goodman, G. S. (2003). Taking responsibility for an act not committed: the influence of age and suggestibility. Law and Human Behavior, 27, 141
Redlich, A. D., & Ozdogru, A. A. (2009). Alford pleas in the age of innocence. Behavioral Sciences and the Law, 27, 467-488. doi: 10.1002/bsl.876
Redlich, A. D., & Shteynberg, R. V. (2016). To plead or not to plead: A comparison of juvenile and adult true and false plea decisions. Law and Human Behavior. Advance online publication. doi: 10.1037/lhb0000205
Redlich, A. D., Summers, A., & Hoover, S. (2010). Self-reported false confessions and false guilty pleas among offenders with mental illness. Law and Human Behavior, 34, 79-90. doi: 10.1007/s10979-009-9194-8
Redlich, A. D., Wilford, M. M., & Bushway, S. (2017). Understanding guilty pleas through the lens of social science [Special anniversary issue]. Psychology, Public Policy and Law, 23(4), 458-471. doi: 10.1037/law0000142
Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton University Press.
Ross, J. E. (2006). The entrenched position of plea bargaining in United States legal practice. The American Journal of Comparative Law, 54, 717-732. doi: 10.2307/20454559
Russano, M. B., Meissner, C. A., Narchet, F. M., & Kassin, S. M. (2005). Investigating true and false confessions within a novel and experimental paradigm. Psychological Science, 16(6), 481-486. doi: 10.1111/j.0956-7976.2005.01560.x
Shakespeare, W. (1998). Hamlet. S. Barnet (Ed.). New York, NY: Penguin Group.
Sommers, I., Goldstein, J., & Baskin, D. (2014). The intersection of victims’ and offenders’ sex and race/ethnicity on prosecutorial decisions for violent crimes. Justice System Journal, 35(2), 178-204. doi: 10.1080/009826X.2013.869153
Stephens, R. (2013). Disparities in postconviction remedies for those who plead guilty and those convicted at trial: A survey of state statutes and recommendations for reform. The Journal of Criminal Law and Criminology, 103(1), 309-342.
Subramanian, R., Moreno, R., & Broomhead, S. (2014). Recalibrating justice: A review of 2013 state sentencing and corrections trends. New York: Vera Institute of Justice, Center on Sentencing and Corrections.
Tor, A., Gazal-Ayal, O., & Garcia, S. M. (2010). Fairness and the willingness to accept plea bargain offers. Journal of Empirical Legal Studies, 7, 97-116. doi: 10.1111/j.1740-1461.2009.01171.x
Tverksy, A., & Kahneman, D. (1981). The framing of decisions and the psychology of choice. Science, 211(4481), 453-458. doi: 10.1126/science.7455683
U.S. Sentencing Commission (2011). Mandatory minimum penalties in the federal criminal justice system. Retrieved from http://www.ussc.gov/research/congressional-reports/2011-report-congress-mandatory-minimum-penalties-federal-criminal-justice-system
Viljoen, J. L., Klaver, J., & Roesch, R. (2005). Legal decisions of preadolescent and adolescent defendants: Predictors of confessions, pleas, communication with attorneys, and appeals. Law and Human Behavior, 29(3), 253-277. doi: 10.1007/s10979-005-3613-2
Wilford, M. M., & Wells, G. L. (2018). Bluffed by the dealer: Distinguishing false pleas from false confessions [Special section on Guilty Pleas]. Psychology, Public Policy & Law, 24(2), 158-170. doi: 10.1037/law0000165
Yandell, B. (1979). Those who protest too much are seen as guilty. Personality and Social Psychology Bulletin, 5, 44-47. doi: 10.1177/014616727900500109
Zimmerman, D. M., & Hunter, S. (2018). Factors affecting false guilty pleas in a mock plea bargaining scenario. Legal and Criminological Psychology, 23(1), 53-67. doi: 10.1111/lcrp.12117
Zottoli, T. M., Daftary-Kapur, T., Winters, G. M., & Hogan, C. (2016). Plea discounts, time pressures, and false-guilty pleas in youth and adults who pleaded guilty to felonies in New York City. Psychology, Public Policy, and Law, 22(3), 250-259. doi: 10.1037/law0000095
Significant two-way interaction between belief in a just world and guilt-innocence
Significant two-way interaction between openness-intelligence and guilt-innocence
The Reasons Provided for Plea Decisions in Experiment 1
|Reasons for Accepting||Innocent||Guilty||Reason for Rejecting||Innocent||Guilty|
|Avoiding Worse Consequences||29.4% (10)||23.9% (11)||Innocent||51.6% (16)||43.8% (7)|
|Best Alternative||23.5% (8)||37.0% (17)||No Proof||12.9% (4)||6.3% (1)|
|Trapped/Pressured||14.7% (5)||10.9% (5)||I’ll Fight This||9.7% (3)||12.5% (2)|
|Guilty||0.0% (0)||13.0% (6)||Miscellaneous||25.8% (8)||37.5% (6)|
|Miscellaneous||32.4% (11)||15.2% (7)|
Note: The % frequency of reasons provided for acceptance and rejection of the plea deal among the guilty versus the innocent research participants. The number (n) of participants providing each reason appear in parentheses.
Descriptive statistics and reliability measures for Experiment 1 individual difference measures
|RangeLow High||Mean (SD)||Cronbach’s α|
|Belief in a Just World||1.43||5.43||3.60 (.71)||.78|
Note. N = 142. All items for which strong agreement would imply lower endorsement of the relevant trait were reverse-coded. The measures were then averaged and aggregated into indices. The belief in a just world index included seven items, the self-esteem index included ten items, and all of the Big Five Aspect indices included ten items. All items were measured on 7-point scales except belief in a just world, which was measured on a 6-point scale.
Proportion of plea acceptance across all eight experimental conditions in Experiment 2
|Innocent||25% Chance of Conviction||80% Chance of Conviction|
|Less Strict Penalty||More Strict Penalty||Less Strict Penalty||More Strict Penalty|
|52.3 (23)[37.9, 66.2]||46.5 (20)[32.5, 61.1]||70.5 (31)[55.8, 81.8]||63.4 (26)[48.1, 76.4]|
|Guilty||25% Chance of Conviction||80% Chance of Conviction|
|Less Strict Penalty||More Strict Penalty||Less Strict Penalty||More Strict Penalty|
|81.6 (31)[66.6, 90.8]||75.5 (37)[61.9, 85.4]||82.1 (32)[67.3, 91.0]||82.9 (34)[68.7, 91.5]|
Note. The frequency appears in parentheses (n); the confidence intervals for each proportion appear in the second row.
The Reasons Provided for Plea Decisions in Experiment 2
|Reasons for Accepting||Innocent||Guilty||Reason for Rejecting||Innocent||Guilty|
|Avoiding Worse Consequences||38.4% (38)||37.3% (50)||Innocent||75.0% (54)||54.5% (18)|
|Best Alternative||24.2% (24)||23.1% (31)||Miscellaneous||13.9% (10)||18.2% (6)|
|Trapped/Pressured||20.2% (20)||10.4% (14)||Unfair Punishment||4.2% (3)||9.1% (3)|
|Confused/Don’t Know||4.0% (4)||4.5% (6)||No Proof||4.2% (3)||6.1% (2)|
|Seemed Right||4.0% (4)||6.7% (9)||No Time||2.8% (2)||12.1% (4)|
|Can’t Prove Innocence||3.0% (3)||1.5% (2)|
|Conclude Situation||3.0% (3)||2.2% (3)|
|Guilty||0.0% (0)||13.4% (18)|
|Miscellaneous||3.0% (3)||0.7% (1)|
Note: The % frequency of reasons provided for acceptance and rejection of the plea deal among the guilty versus the innocent research participants. The number (n) of participants providing each reason appear in parentheses.
1Other studies that have employed a plea bargain or deal manipulation (Dervan & Edkins, 2013; Gregory et al., 1978, Exp. 2; Russano et al., 2005) did require a confession as a term of pleading or accepting the deal.
2Russano et al. (2005) included a deal manipulation in their study, however, this manipulation was employed in an interrogation context in which participants were verbally offered the terms of a deal in exchange for signing a confession. While this can sometimes occur in the real world, confessions secured via a deal are typically deemed inadmissible. Further, participants in the Russano et al. (2005) study had to trust their verbal agreement would be honored, whereas participants in the current research signed a statement explicitly agreeing to the imposed punishment. Gregory et al. (1978, Exp. 2) created a high-stakes plea simulation to examine the impact of guilt status on plea outcomes, but their study did not incorporate any other manipulations (e.g., trial penalty, conviction probability).
3Although Dervan and Edkins (2013) did provide participants with a probability of conviction value (80-90%), this value was kept constant across participants.
4We chose to exclude everyone with psychological research lab experience due to the increased likelihood that they would be familiar with studies involving deception. We also felt it was possible that they would not have perceived the consequence of working in the lab as negatively as most people.
5Experiment 1 included eight female experimenters. Experiment 2 included twelve experimenters: nine were female and three were male. All of the experimenters (for both experiments) underwent extensive training including three ~90-minute sessions supervised by the lead author.
6If the descriptive odds of accepting the plea bargain are used to compute the odds ratio, it will differ slightly from the value produced by the logistic regression model. This discrepancy is due to the logistic regression model computing values for which the effects of all other model predictors are removed.