Mathematical Structures in Group Decision-Making on Resource Allocation Distributions

Noah E. Friedkin, Anton V. Proskurnikov, Wenjun Mei, Francesco BulloScientific 2019 · DOI

Abstract

Optimal decisions on the distribution of finite resources are explicitly structured by mathematical models that specify relevant variables, constraints, and objectives. Here we report analysis and evidence that implicit mathematical structures are also involved in group decision-making on resource allocation distributions under conditions of uncertainty that disallow formal optimization. A group's array of initial distribution preferences automatically sets up a geometric decision space of alternative resource distributions. Weighted averaging mechanisms of interpersonal influence reduce the heterogeneity of the group's initial preferences on a suitable distribution. A model of opinion formation based on weighted averaging predicts a distribution that is a feasible point in the group's implicit initial decision space.

Introduction

Resource allocation—the division of a finite "pie" among competing options—is a fundamental organizational challenge. Traditionally, this task is approached through either formal optimization (e.g., linear programming) or informal group consensus. While formal methods rely on explicit constraints and objectives, the "social algorithm" of informal discussion often appears chaotic or purely political. This study bridges this gap by investigating whether informal group decisions are governed by implicit mathematical structures and whether these decisions can be predicted using networked models of opinion dynamics.

Methodology and the Friedkin-Johnsen Model

The researchers conducted five experiments with groups of 3–4 participants, progressively relaxing formal constraints from classic linear programming problems to purely informal allocations, such as faculty hiring and state budget decisions.

To model the social process, the study employed the Friedkin-Johnsen (FJ) model of social influence. Unlike simpler models that assume eventual total consensus, the FJ model accounts for individual "stubbornness"—the degree to which participants remain attached to their initial positions. The model integrates three key components:

  1. $X(0)$: The array of initial individual preferences.
  2. $W$: An influence matrix, derived from participants' self-reported influence weights.
  3. $X(\infty)$: The final settled-upon opinions.

Data was gathered from three disjoint samples of undergraduate subjects at the University of California, Santa Barbara, totaling 302 participants across 87 groups.

The Implicit Polytope and Geometric Decision Space

A central finding of this research is the emergence of the implicit polytope. A group’s initial range of opinions (the minimum and maximum values for each resource category) defines a bounded convex polytope. This geometric structure acts as a "sandbox" that constrains the negotiation process.

Figure 6: Visualization of the implicit decision space (polytope) formed by initial group preferences (A) and the subsequent convergence of opinions within that space (B).

Even without explicit instructions regarding limits, groups rarely settle on values outside the range defined by their members' initial preferences. This implicit polytope represents a physical manifestation of "bounded rationality," where the group limits its consideration to the subspace defined by collective initial expertise and biases.

Key Results

The FJ model demonstrated remarkable predictive accuracy across all experimental conditions, even without parameter fitting:

  • Predictive Power: In Experiment 2, the correlation ($\rho$) between the model’s predicted final positions and observed positions reached 0.978 ($p < 0.0001$).
  • Constraint Adherence: In informal allocation tasks (Experiments 4 and 5), between 85.6% and 89% of final positions were located within the group's implicit decision space.
  • Consensus Efficiency: Group discussion significantly reduced the "hazard rate" of incorrect solutions in formal tasks compared to individuals working in isolation.

Conclusion and Future Directions

The study concludes that group discussion acts as a weighted averaging mechanism. Final consensus is not a random compromise but a mathematically predictable result of interpersonal influence. While these findings are robust in controlled lab settings, future work should explore field settings with exogenous disturbances and utilize Natural Language Processing (NLP) to extract influence weights directly from discussion transcripts. Additionally, the model could be expanded to account for "creative" shifts where groups move beyond the initial convex hull of opinions.

Summary updated

Abstract

Optimal decisions on the distribution of finite resources are explicitly structured by mathematical models that specify relevant variables, constraints, and objectives. Here we report analysis and evidence that implicit mathematical structures are also involved in group decision-making on resource allocation distributions under conditions of uncertainty that disallow formal optimization. A group's array of initial distribution preferences automatically sets up a geometric decision space of alternative resource distributions. Weighted averaging mechanisms of interpersonal influence reduce the heterogeneity of the group's initial preferences on a suitable distribution. A model of opinion formation based on weighted averaging predicts a distribution that is a feasible point in the group's implicit initial decision space.
BibTeX
@inproceedings{friedkin2019mathematical,
  author = {Noah E. Friedkin and Anton V. Proskurnikov and Wenjun Mei and Francesco Bullo},
  title = {{Mathematical Structures in Group Decision-Making on Resource Allocation Distributions}},
  booktitle = {Scientific},
  year = {2019},
  doi = {10.1038/s41598-018-37847-2},
}
Figure