Markov Chain Models of Jury Deliberation with Ordered Verdict Options develops a formal model of jury deliberation as an absorbing Markov chain. The article shows how a jury’s initial vote configuration can be translated into exact probabilities of final verdicts by modeling deliberation as a sequence of probabilistic transitions among possible voting states until unanimity is reached. It validates the model against observed criminal jury data, extends it to cases with multiple ordered verdict options such as lesser-included offenses, and argues that the framework can be used both to study small-group decision-making more generally and to estimate how trial procedures or trial errors affect verdict probabilities in criminal cases.

---

What the article studies

This article addresses a basic but consequential question: when a jury begins deliberating from a particular starting point, what verdict is it likely to reach? Rather than treating deliberation as a black box, the article models it as a dynamic process in which jurors move among possible vote configurations over time until the group reaches unanimity. The focus is on juries, but the article presents the framework as a more general model of small-group deliberation that can be adapted to other collective decision settings.

The article is especially important within the Fair Trial Analysis research program because it supplies a formal bridge between individual juror preferences and collective jury verdicts. That bridge is essential if fairness is to be measured in terms of changed verdict probabilities rather than intuition alone.

How the model works

The article models deliberation as an absorbing Markov chain. In the simplest binary setting, the jury’s state is defined by the number of jurors currently favoring guilt. Some states are transient, meaning deliberation continues, while unanimous states are absorbing, meaning the jury has effectively reached a final verdict and will remain there. The model then uses a transition probability matrix to calculate the probability that the jury will end in each absorbing state from any given starting point.

A key feature of the model is that its transition probabilities are not random in a simplistic sense. They reflect three substantive features of jury deliberation described in the article: individual-level variation, the social influence of larger factions, and a leniency bias created by the legal presumption favoring the defendant in criminal trials. In the binary model, these ingredients generate the familiar result that the probability of conviction rises nonlinearly as the initial guilty faction grows.

The article then extends the model beyond simple guilty/not-guilty decisions. In cases with ordered verdict options—for example, not guilty, manslaughter, second-degree murder, and first-degree murder—the jury’s state is represented as a vector of vote counts across options rather than a single number. This allows the model to analyze how deliberation proceeds when jurors may move between adjacent options rather than only between guilt and acquittal.

What the article finds

The article finds that the Markov chain model closely replicates observed jury behavior in binary-verdict settings. Using data from 2,303 observed jury deliberations, it compares the model’s predicted verdict probabilities with real-world outcomes and reports that the model closely reproduces the observed S-shaped relationship between the initial size of the guilty-verdict faction and the probability of a guilty verdict. The validation figure on page 20 visually shows the predicted curve closely tracking the observed data points.

The article also shows that the framework performs well in settings with multiple verdict options. Drawing on data reported by Hastie, Penrod, and Pennington, it uses the model to predict verdict frequencies under different decision rules, including unanimity, 10-2, and 8-4 rules. The predicted verdict counts closely match the observed verdict distributions, and the reported goodness-of-fit tests indicate no significant deviation between model predictions and observed outcomes in those study conditions.

The article’s figures and tables help make this concrete. Figure 3 presents a two-dimensional Markov chain for a three-person jury choosing among three ordered outcomes, showing how legal one-step moves connect vote states. Table 1 then demonstrates that, when the model is fed starting distributions from prior experimental work, it predicts deliberated verdict patterns with substantial accuracy.

Why this article matters

This article matters because it provides a transparent and analytically tractable way to model what happens between the beginning and end of deliberation. Existing research has long shown that initial jury preferences strongly influence outcomes, but this article gives that relationship a formal structure that yields exact verdict probabilities rather than rough intuition.

That contribution is important both academically and practically. Academically, it places jury decision-making within a broader class of well-understood stochastic systems and connects legal decision-making to political science research on deliberation, collective behavior, and dynamic systems. Practically, it provides the kind of formal tool needed to estimate how much procedural changes, evidentiary rulings, or trial errors alter the probability of conviction, conviction on lesser offenses, or death sentences.

Applications to trial fairness

The article explicitly ties the model to applied legal analysis. It explains that, once researchers estimate verdict preferences in the relevant jury pool, the model can be used to infer verdict probabilities under different trial conditions and thereby estimate how much a trial error or omission changed the likelihood of conviction or punishment. In that way, the article supplies a formal component of the broader Fair Trial Analysis effort to measure harmfulness, prejudice, and fairness in criminal trials more rigorously.

The article also shows how the model can illuminate questions about lesser-included offenses. By extending earlier research from individual preferences to deliberated verdicts, it demonstrates that the verdict options made available to juries can materially change final outcomes. That is useful both for litigators deciding whether certain instructions help or hurt their case and for courts evaluating whether a mistaken lesser-included-offense instruction was harmful.