Abstract
We investigate joint probabilistic choice rules describing the behavior of two decision makers, each facing potentially distinct menus. These rules are separable when they can be decomposed into individual choices correlated solely through their respective probabilistic choice rules. Despite its significant interest for the study of peer effects, influence, and taste variation, a complete characterization of these rules has remained elusive (Chambers, Masatlioglu, and Turansick, 2021). We fully characterize separable choices through a finite system of inequalities inspired by Afriat's theorem. Our results address the possibility of entangled choices, where decision makers behave as if they do not communicate, yet their choices are not separable. More generally, we establish that separable joint choice restrictions can be factored into individual choice restrictions if only if at least one decision maker's probabilistic choice rule uniquely identifies the distribution over deterministic choice rules. The no communication condition and the individual restrictions are no longer sufficient in the absence of this uniqueness. Our results offer robust tools for distinguishing between separable decision-making and behaviors beyond mere peer effects such as imitation and cheating.
Introduction
In social sciences, understanding how groups make decisions is fundamental to analyzing peer effects, social influence, and shared preferences. A critical challenge lies in distinguishing between "separable" choices—where correlation between two parties arises strictly from shared underlying tastes—and non-separable behaviors, where correlation stems from active communication, imitation, or coordination.
This paper provides the first complete mathematical characterization of separable joint choices. By bridging concepts from theoretical physics (such as Bell inequalities) and consumer theory (Afriat-like inequalities), the research demonstrates that the standard "no communication" condition is surprisingly insufficient to guarantee independent behavior. This phenomenon, termed "entangled choice," reveals a gap in our understanding of joint decision-making that has significant implications for both theory and empirical research.
The Problem of Separability and Entanglement
A joint choice rule is considered separable if two decision-makers (DMs) act as if they are in separate rooms: they do not communicate, but their choices may appear correlated because they share a similar background or preference distribution.
Historically, researchers assumed that if DMs satisfied the "no communication" condition—meaning their individual choices do not depend on the menu offered to the other person—their behavior was inherently separable. However, the authors identify a "gap" where DMs satisfy this marginality condition yet their joint behavior cannot be explained by any separable model. This is "entanglement," a concept adapted from quantum mechanics to describe a system where the whole is more than the sum of its independent parts, even in the absence of direct interaction.
Methodology
The authors utilize a rigorous framework combining linear algebra and functional analysis to model these behaviors:
- Matrix Representation: Probabilistic choice rules are represented as linear systems ($A\nu = \rho$), where $A$ encodes deterministic choice patterns and $\nu$ is a probability distribution over those patterns.
- Kronecker Products: To model joint behavior, the authors use Kronecker products of individual choice matrices ($A^1 \otimes A^2$), representing all possible combinations of independent individual behaviors.
- Bell/CHSH Inequalities: The research adapts the Clauser-Horne-Shimony-Holt (CHSH) inequalities from quantum physics. These provide a mathematical test for "excess correlation" that marginality alone cannot detect.
- Marginalizable Extensions: Inspired by Afriat’s Theorem in revealed preference theory, the authors develop a method of "virtual replication." They prove that a choice rule is separable if it can be extended to a system with "virtual" clones of the DMs while maintaining the no-communication property.
Key Insights: The Uniqueness Requirement
The most striking insight of the paper is the role of uniqueness. The "no communication" condition only implies separability if at least one DM’s choice rule uniquely identifies their underlying preference distribution.
If both DMs have "partially identified" preferences—meaning multiple distributions could explain their individual behavior—they can effectively "coordinate" on these hidden distributions. This coordination produces entangled choices without any actual communication. In a social context, this looks like "faking" independence while being driven by a hidden common agent or a shared latent strategy.
Formal Results
The paper establishes several landmark theorems for decision theory:
- Theorem 1: Individual behavioral restrictions are sufficient to characterize separable choice if and only if at least one DM has a unique representation of their choice rule.
- Theorem 2: In the general case (where uniqueness is absent), a joint choice rule is consistent with the separable model if and only if it is "marginalizable on average." This means it can be embedded into a larger, symmetric system of virtual decision-makers.
- Proposition 1: The authors formally link CHSH inequalities to the thought experiment of separable choice, proving they are necessary and sufficient for identifying entanglement in specific simple domains.
Conclusion and Future Directions
By providing a finite system of linear inequalities to characterize separable choice, this research makes the theory testable in practice. However, challenges remain. Testing for separability is generally an NP-hard problem, as the number of inequalities grows rapidly with the complexity of the menus.
Future work will involve finding specific "Bell-type" inequalities for high-dimensional choice environments and applying these tools to empirical data. Distinguishing between genuine peer effects and "entangled" coordination (such as cheating or hidden influence) remains a vital frontier for understanding human social behavior.
Summary updated