Feedback dynamics in matching networks drive behavioral differentiation despite overlapping objectives

Technical Summary: Feedback Dynamics in Matching Networks Drive Behavioral Differentiation Despite Overlapping Objectives

Problem Statement
Many bipartite social networks (e.g., dating apps, hiring markets, housing searches) exhibit pronounced asymmetries in selectivity: one group can afford to be highly selective, while the other is forced to accept less desirable matches. While such imbalances are often attributed to intrinsic differences between groups (e.g., gender ratios, evolutionary psychology, or cultural norms) or exogenous scarcity, this paper investigates whether such asymmetries can emerge endogenously from the matching dynamics themselves. Specifically, the authors ask whether a feedback loop, inherent to the structure of repeated matching networks, can drive two groups with nearly identical objectives and population sizes toward diametrically opposite behaviors regarding selectivity.

Methodology
The authors develop a continuous-time mathematical model of repeated one-to-one matching between two groups (denoted as "males" and "females" with sizes $M$ and $N$). The model is built on five core properties:

1. Two non-overlapping groups attempt to match across groups.
2. Utility is non-transferable (no prices or salaries adjust match attractiveness).
3. Agents form matches repeatedly (matches do not permanently remove agents from the pool).
4. Agents have "soft limits" on matching rates, defined by a target matching rate ($c_i$ or $d_j$).
5. Preferences are idiosyncratic (uncorrelated in the base model), meaning an agent's acceptance probability depends on their own selectivity, not the specific identity of the partner.

Model Dynamics:
Agents adjust their selectivity (modeled as an acceptance probability, $a_i(t)$ or $b_j(t)$) to achieve their target matching rate. The evolution of selectivity is governed by differential equations where the rate of change is proportional to the difference between the agent's experienced matching rate and their target rate.

• If the experienced rate is below the target, the agent becomes less selective (acceptance probability increases).
• If the experienced rate exceeds the target, the agent becomes more selective (acceptance probability decreases).
• The system is constrained such that acceptance probabilities remain within $[0, 1]$.

The authors analyze this system using both analytic techniques (proving existence and uniqueness of equilibrium) and numerical simulations (exploring heterogeneous populations and varying parameters like encounter rates and population ratios).

Key Contributions and Results

1. Existence and Uniqueness of Equilibrium:
   The paper proves that barring a specific borderline case where the total target matching rates of both groups exactly balance ($\sum c_i = \sum d_j$), the system possesses a unique, globally attractive equilibrium. Regardless of initial conditions, the selectivity of every agent converges to a fixed value determined by the system parameters.

2. Emergence of Extreme Asymmetry:
   In the case of homogeneous groups (all males have target rate $c$, all females have $d$), the authors derive an analytic expression for the equilibrium. They demonstrate that even a minuscule difference in the aggregate target rates (e.g., $Mc > Nd$) drives the group with the higher aggregate demand to become entirely non-selective (acceptance probability $\to 1$), while the other group becomes highly selective.

   • Example: If $M=N=100$ and the female target rate is slightly lower than the male target rate, females may end up accepting only 1% of encounters, while males accept 100%.

3. Robustness to Heterogeneity:
   Through simulations with heterogeneous populations (where target rates are drawn from overlapping distributions), the paper shows that the phenomenon persists. Even when the distributions of individual goals are almost indistinguishable, the equilibrium selectivity distributions become highly polarized.

   • Key Finding: Group membership is a stronger determinant of behavior than individual goals. Individuals in the "in-demand" group with high target matching rates (who would theoretically be "promiscuous") end up being highly selective, while individuals in the "supply" group with low target rates end up being non-selective.

4. Threshold Behavior and Sensitivity:

   • Encounter Rate ($K$): The polarization effect is contingent on sufficiently frequent encounters. If the encounter rate is low (below the target matching rate), the feedback mechanism weakens, and the "in-demand" group cannot afford to be selective.
   • Population Ratio: The system exhibits threshold behavior. The polarity of selectivity reverses abruptly only when the ratio of group sizes crosses the inverse ratio of their target matching rates. Small deviations from this critical threshold do not significantly alter the outcome.

5. Extension to Variable Attractiveness:
   The model is extended to include heterogeneous attractiveness factors. The results indicate that while attractiveness inequality affects the specific values of acceptance probabilities, it does not qualitatively change the polarization phenomenon; the structural feedback still drives one group toward high selectivity and the other toward low selectivity.

Significance and Claims
The paper claims that structural feedback mechanisms embedded in matching networks can generate large behavioral asymmetries that do not necessarily reflect equally large differences in underlying preferences or group compositions.

• Theoretical Implication: The authors argue that observed behavioral differences in bipartite markets (such as the disparity in selectivity between men and women on dating apps) should not be automatically attributed to intrinsic differences in mating psychology or social roles. Instead, a significant portion of this asymmetry may be an emergent property of the matching dynamics themselves.
• Mechanism: The core mechanism is a feedback loop: as one group becomes more selective, it reduces matching opportunities for the other group, forcing the latter to become less selective to maintain their target matching rate. This loop amplifies tiny initial imbalances in population size or average goals.
• Applications: The model provides a framework for understanding selectivity in various repeated matching environments, including online dating, hiring interviews, apartment hunting, and academic conference submissions. It suggests that interventions altering the encounter rate (e.g., platform algorithms) could substantially change observed selectivity without altering user preferences.

The paper concludes that understanding the dynamics of matching is crucial for distinguishing between genuine preference differences and structural artifacts of the market.