Satisfaction, Legitimacy, and Guilty Pleas: How Perceptions and Attorneys Affect Defendant Decision-Making

Abstract

Though almost all convictions are the result of guilty pleas, little research has examined influences on defendant plea decisions. First, we explore how attorney satisfaction and court legitimacy affect defendant choices. Using an experimental approach, we then assess the impact of a defense attorney’s evaluation of a plea offer on the defendant’s willingness to accept a plea, and whether this is moderated by defendants’ individual views. Findings demonstrate that higher attorney satisfaction and higher court legitimacy are associated with higher willingness to accept a plea. Defendant decision-making is also impacted by their attorney’s evaluation of the plea and this influence is moderated by the defendant’s satisfaction with their representation. In line with previous research, we also find differences across guilt indicating that guilty and innocent defendants may have different decision-making processes. Results suggest that guilty plea research should continue to evolve with increased attention to the defendant and their counsel.

KEYWORDS:

• Plea-bargaining
• guilty pleas
• defense attorney satisfaction
• court legitimacy
• defendant decision-making

Acknowledgments

We would like to thank all of our participants who gave us their valuable time to complete our survey. This paper also would not have been possible without the assistance of Ms. Hannah Schlangen. We would also like to thank Dr. Teddy Wilson for his comments on a previous draft and Dr. Stephen Utych for his feedback.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 As outlined by Bushway and Redlich (2012), a trial sentence has uncertainty, with two potential outcomes: 1) Conviction with sentence (X), or 2) Acquittal with no sentence (0). Where there is a probability of conviction (P) and a probability of acquittal (1-P). The expected value of the trail is then, E(Trial) = P(X) + (1-P)0 = P(X). Although the probability of conviction can be measured objectively (e.g., #convictions/#trial cases; Bushway & Redlich, 2012) it could also be captured perceptually (e.g., asking respondents to estimate, on a scale from 0 to 100%, the probability of convict if the case were to go to trial; Bushway et al., 2014).

2 Although the model has primarily been applied to judges, prosecutors and defense attorneys, it has been described as “a formal theoretical (i.e. mathematical) model that starts from the perspective of the defendant” (Bushway et al., 2014, p. 726, emphasis added).

3 In Padilla v. Kentucky (2010), the Court held that failing to advise a client regarding collateral consequences could be considered ineffective assistance of counsel and in Missouri v. Frye (2012), the court stated that attorneys must inform their clients of every offer presented by the prosecution. Lafler v. Cooper (2012) affirmed that providing unsound advice during the plea process can be ineffective assistance of counsel and critically, the Court confirmed that the final decision to accept or reject a plea offer rests with the defendant (McCoy vs. Louisiana, 2018).

4 This theoretical argument stressing the importance of perceptions in decision-making parallels similar transitions in emphasis within criminological theory. For example, decades ago deterrence scholars restated deterrence as a perceptual theory of sanction threats and transitioned from objective to subjective measures of sanction risk (Geerken & Gove, 1975; Waldo & Chiricos, 1972; see also Paternoster, 2010). An analogous shift also transpired within strain theory, when Agnew (2001) argued that the stress literature had evolved to recognize that individuals differed in their subjective evaluations of an objective strain.

5 Related studies have also indicated that perceptions of fairness (as measured by asking survey respondents if they believed they would be treated fairly by courts if they were charged with a crime, or were a plaintiff in a civil lawsuit) are incredibly important for predicting support for the court (Benesh, 2006; Cann & Yates, 2008; Olson & Huth, 1998; Tyler, 1988).

6 The vast majority of respondents (97%) reported driving in the last year, over half reported driving above the speed limit, and the idea of coming sharply around a corner while driving is a crime that even highly law-abiding individuals could envision themselves committing.

7 We considered operationalizing change in WTAP as percentage change, however, a number of individuals initially rejected the offer and thus had a zero value for the initial acceptance. Given this, we are unable to calculate a percentage change.

8 To evaluate whether court legitimacy and defense attorney quality were two separate latent constructs as expected, exploratory factor analysis was used. All 20 items were analyzed together with Promax rotation. Results indicated that they are two distinct factors (Eigenvalues 5.54 and 2.36) with no observed problematic cross-loadings. Factor loadings are provided in the Online Appendix Table A. Given that all items loaded onto distinct factors, the sample size is large, and internal consistency is high, all items with factor loadings greater than .30 were retained. Though there is no agreed upon threshold for factor loadings, .30 was selected because scholars have suggested this threshold may indicate practical significance (Hair, Tatham, Anderson, & Black, 1998). Further, a factor can be considered reliable if it has at least four loadings greater than .60, which both scales satisfy (Field, 2005). Sensitivity analysis was also conducted retaining only items with factor loadings greater than .50 and results were substantively unchanged.

9 Though other studies have provided this probability directly to participants (Bordens, 1984; Bushway et al., 2014; McAllister & Bregman, 1986), we elected to allow them to assess this on their own. This choice allows us to use the respondents’ own views of the court system and is particularly relevant in the present study given that respondents may not be familiar with typical conviction probabilities.

10 The two domains/behaviors from Weber and colleagues (2002) chosen for this study due to their high relevance were health and ethics. We also added four additional relevant behaviors from Gardner and Steinberg (2005, p. 629): riding in a car with someone who has been drinking, trying a new drug you know nothing about, breaking into a store to steal something, and driving over 90 miles per hour at night on the highway.

11 Data were missing on the following variables in the frequencies listed in parentheses: sex (5), age (7), race (6), plea severity discount (10), willingness to accept initial plea offer (4), change in willingness to accept plea offer (8), probability of conviction (5), court legitimacy (18), defense attorney satisfaction (18), risk aversion (9), private attorney (13). A chi-squared test indicated that there is not a significant difference in missing observations across treatment conditions (p = .15).

12 Power analysis indicated that a sample size of 432 is sufficient to detect meaningful differences between treatments at a significance level of p < .05. Our data collection continued beyond the requisite sample size to ensure that the study had sufficient statistical power after accounting for potential missing data and to allow all faculty who were recruited to have the opportunity for their class to participate in the research study.

13 Given the high number of zero values present in the WTAP₁ variable (19% of the sample), we also considered using a Tobit model. However, we decided against this because resent research suggests there may be bias in this estimator (Wilson, Loughran, & Brame, 2020). We also considered categorizing the dependent variable as follows: 5 (> =75); 4 (>50 and <75); 3 (=50); 2 (<50 and >25); 1 (< =25 and >0); 0 (=0), and analyzing this variable using an Ordered-Logit model. We decided against this, however, because recoding the variable as categorical could mask potentially important variation in the dependent variable. However, as a robustness check we analyzed the data using all three of these approaches (OLS, Tobit, and ordered logit) and findings are substantively similar across each modeling strategy.

14 We evaluated the data for the presence of outliers and identified 17 individuals with potential to systematically bias results. However, sensitivity analysis indicated that results were not sensitive to removal of these individuals, so they were retained in the sample.

15 Similar to sensitivity analysis with respect to the first dependent variable, we also considered analyzing this variable using an Ordered Logit Model. To do this we, operationalized WTAP1 and WTAP2 categorically (as described in the previous footnote) and created a “categorical change” variable WTAP2cat-WTAP1cat. Substantive findings were, again, robust with this approach.

16 As an additional robustness check, models analyzing our second dependent variable (Tables 3–5) were also run with the controls from Table 2 included and results were substantively unchanged.

17 Tables 4 and 5 include a variety of 2 and 3-way interactions. We also tested a 4-way interaction between legitimacy or satisfaction and guilt, value of plea, and attorney evaluation. We found no significant 4-way interactions (p > .10) and these are thus not included in tables or discussion.

18 There is also a significant interaction between DAS and receiving a “good” relative to “bad” evaluation (p < .001), though these results are not depicted within the table. Full results available upon request.

19 There is also no significant evidence of a 4-way interaction (evaluation*guilt*3-year*DAS), results available upon request.

20 Online Appendix Table C provides an additional illustration of these findings. Rather than including a three-way interaction between evaluations, court legitimacy, and plea offer (e.g. Good*CL*3-Year) in the model (as within Table 5 Model 3), here two separate models are run: one holding 3-year plea constant, the second holding 7-year plea constant. In both of models, attorney evaluation is then interacted with court legitimacy and a Z-test for equality of coefficients (Paternoster et al., 1998) is conducted (comparing both Good*CL; and Bad*CL) across models. Here, findings are entirely consistent with the results presented in text.

21 In supplemental analysis we implemented the Karlson, Holm, and Breen (2012) method of mediation analysis to examine whether perceptions of defense attorney satisfaction and court legitimacy influenced defendant’s WTAP through their effect on perceived probability of conviction as indicated by the classic shadow model. Findings indicated that these perceptions had direct effects on WTAP and were not entirely mediated by their influence on individuals’ perceived probability of conviction, contradicting the expectations of the shadow model.

Additional information

Notes on contributors

Jacqueline G. Lee

Dr. Jacqueline G. Lee is an Assistant Professor at Boise State University in the Department of Criminal Justice. Prior to graduate school, she also earned a JD, and her research focuses on courts and sentencing, prosecutorial discretion, and the integration of legal and social science research. ORCiD: https://orcid.org/0000-0002-0680-374X

Chae M. Jaynes

Dr. Chae M. Jaynes is an Assistant Professor at the University of South Florida in the Department of Criminology. Her research focuses on offender and criminal justice practitioner decision-making, rational choice, offender reentry and employment, criminal justice policy, and quantitative methodology.

John Ropp

John Ropp is a PhD student at Michigan State University. He completed his M.A. at Boise State University.