Introduction: The Doctrinal Paradox
In the field of judgment aggregation, a group of agents (such as a jury or committee) votes on several interconnected propositions. For instance, in a legal setting, jurors might vote on two premises: (1) "The defendant caused the injury" and (2) "The injury is serious." The conclusion, "The defendant is guilty," is logically dependent on both premises being true.
The doctrinal paradox arises when the majority vote on each individual premise is "Yes," but the majority vote on the conclusion is "No," even if every single juror was logically consistent in their own private vote. This creates a fundamental tension between premise-based and conclusion-based decision-making.
While theoretical "worst-case" scenarios prove this paradox is unavoidable, its real-world frequency has remained poorly understood. Previous studies often relied on simplistic models, like the "Impartial Culture" model where every juror is assumed to flip a fair coin. This paper bridges the gap by using a more robust "semi-random" framework to characterize the likelihood of these paradoxes in realistic settings.
Methodology: A Semi-random Framework
The authors employ a semi-random social choice framework. In this model, an adversary chooses the ground-truth distribution of judgments (allowing for arbitrary correlations), but the actual votes are subject to independent random noise. This approach is significantly more resilient and realistic than standard models.
To analyze the probability of a paradox, the authors translate the logical problem into a geometric one:
- Polyhedra Presentation: The set of voting profiles leading to a paradox is represented as a union of polyhedra in $m$-dimensional space.
- Multivariate Central Limit Theorems: Advanced probabilistic tools are used to calculate the likelihood that a group's total vote "histogram" lands inside these "paradox polyhedra" as the number of agents ($n$) increases.
- Condition Characterization: They define four mathematical conditions ($\kappa_1$ to $\kappa_4$) that determine the asymptotic likelihood of a paradox.
Key Insights: The "Four-Case" Rule
The most significant finding is that the likelihood of a paradox follows one of four distinct mathematical patterns as the group size $n$ grows:
- Zero: The paradox is mathematically impossible for the given setup.
- Exponentially Vanishing ($\exp(-\Theta(n))$): The paradox becomes extremely unlikely very quickly as the group size increases.
- Polynomially Vanishing ($\Theta(n^{-0.5})$): The paradox disappears slowly.
- Constant ($\Theta(1)$): The paradox is robust and persists even in very large groups.
Key Results and Implications
The paper provides a full mathematical characterization of the maximum and minimum semi-random likelihoods. Numerical experiments, involving one million independent trials per distribution, perfectly align with these theoretical predictions. For instance, in specific settings, the probability of a paradox drops on a log-scale as $n$ increases, confirming the exponential decay prediction.
The results suggest that for many common voting rules, the doctrinal paradox effectively vanishes in large groups. This provides a strong theoretical justification for using simple majority rules in large-scale "e-democracy" platforms, as the inherent noise in human judgment can actually mitigate the risk of logical breakdown. Furthermore, the authors demonstrate that the conditions determining which "bucket" a situation falls into are computationally verifiable using Integer Linear Programming (ILP).
Limitations and Future Work
While comprehensive, the study has limitations. The proofs assume "strict positiveness"—that every combination of judgments has at least a non-zero probability. Future research could explore scenarios where certain judgment combinations are strictly forbidden. Additionally, while the paper covers a broad class of "quota rules," it does not encompass every possible aggregation method. Finally, extending this polyhedral method to more complex agendas with multiple layers of logical dependency remains an open challenge.
Summary updated