The paradox of conviction probability: Mock defendants want better deals as risk of conviction increases.

Abstract

We examined how probability of conviction affects the maximum plea sentence mock defendants will accept. Hypothesis: Relying on Prospect Theory (Kahneman & Tversky, 1979), we hypothesized that, relative to the expected value of trial, participants would need increasingly better sentences as conviction probability increased and would settle for sentences worse than the expected value of trial when probability was very low. Method: We manipulated conviction probability and potential trial sentence in a series of three between-subjects experiments, with Amazon Mechanical Turk participants assigned to the role of guilty defendants. Participants were majority White (75-82%) and non-Hispanic (92-94%); approximately half (45-51%) identified as female. Study 1 (N ϭ 681) explored the effects of conviction probability (.05, .15, .50, .85, .90) and potential trial sentence (5, 20 years) on the maximum sentence accepted in exchange for a plea. Study 2 (N ϭ 343; X age ϭ 37.5) clarified results of Study 1 for the upper range of probabilities for two potential trial sentences (5, 10 years). Study 3 (N ϭ 1,035; X age ϭ 37.6) replicated the effects of probability (.05, .10, .15, .50, .85, .90) and potential trial sentence (5, 10 years). Results: Across all three studies, participants wanted increasingly better deals (relative to the expected value of trial) as conviction probability increased. For example, in Study 3, when probability of conviction was 0.90, plea sentences were, on average, 58% better than the expected value of trial; in contrast, when the probability was 0.05, sentences that were nearly 4 times the expected value of trial were acceptable. Conclusions: The most commonly used model of plea decision-making, Shadow of the Trial (SOT) (Mnookin & Kornhauser, 1979), assumes a direct and constant linear relationship between conviction probability and plea sentence. In contrast, our data suggest that the way conviction probability affects mock defendants' appraisals of plea offers may change across the probability spectrum. These results can facilitate development of a more comprehensive model of plea decision-making. Public Significance StatementCriminal defedants may become increasingly risk seeking as likelihood of conviction increases, requiring progressively better plea offers before pleading guilty. These findings can be used to update existing models that psycho-legal researchers rely on to understand plea decision-making. In addition, knowledge of these data can better equip attorneys for plea negotiation by improving their understanding of how defendants process information about conviction probability.

Developing a model of guilty plea decision-making: Fuzzy-trace theory, gist, and categorical boundaries.

“…In terms of Postulate 1, results of Study 1 clearly showed that the maximum offers participants were willing to accept were similar within groupings of meaningfully similar conviction probabilities (which we refer to as gist-based bands) and were different between these groupings. Given that these data replicate the patterns found by Bartlett and Zottoli (2021), we take these results as persuasive evidence that participants were influenced by categorically meaningful distinctions in POC rather than by the precise numerical changes that would be expected by the SOT model (and by prospect theory).…”

“…Because participant estimates of conviction probability did not drop below approximately .30, this study did not obtain direct observations of systematic deviations at the low end of the probability spectrum. However, among those studies that have manipulated conviction probability directly, deviations from SOT model predictions have been found at both the high and low ends of the probability spectrum (e.g., Bartlett & Zottoli, 2021; Clatch & Borgida, 2021, Study 3; Petersen et al, 2022). For example, Bartlett and Zottoli (2021, Study 3) used vignettes to manipulate POC directly, via attorney communication, across six levels, including three at the low end of the range (.05, .10, .15, .50, .85, .90) and asked participants for the maximum sentence to which they would plead guilty.…”

“…Indeed, Bartlett and Zottoli’s (2021, Study 3) hypotheses as to the deviations from E(trial) at the upper and lower end of the probability spectrum were grounded in prospect theory’s probability function (see Kahneman & Tversky, 1979), which expects participants to overweight small probabilities and underweight large probabilities. However, although their hypotheses were supported, the authors discussed other plausible mechanisms that could give rise to the same data (e.g., hyperbolic discounting; Ainslie & Haslam, 1992) and stressed the importance of research aimed at clarifying the underlying psychology.…”

“…This assumption was not merely theoretical. Prior to conducting Study 1, we pilot-tested four of Bartlett and Zottoli’s (2021, Study 3) conditions as forced-choice studies using the median plea sentences they reported as our explicit plea offers. We found that acceptance rates exceeded 50% for the two highest probability conditions.…”

Guilt status influences plea outcomes beyond the shadow-of-the-trial in an interactive simulation of legal procedures.

“…Thus, although the effect of penalty discrepancy is not moderated by guilt status, the impact in terms of plea acceptance probability is disproportionate between the innocent and the guilty. These variations are made more interesting by recent research demonstrating that the impact of conviction probability on plea outcomes (or the perceived value of plea offers) among guilty defendants is nonlinear in nature (Bartlett & Zottoli, 2021). Consequently, it remains possible that the combination of conviction probability and plea discount influence innocent versus guilty defendants differently at certain points.…”