Vaccination in a Large Population: Mean Field Equilibrium versus Social Optimum

Bruno Gaujal, Josu Doncel, Nicolas Gast

TL;DR

This paper analyzes vaccination strategies in large populations, comparing individual incentives (Nash Equilibrium) with societal well-being (Social Optimum). It finds that individual optimization leads to a less effective vaccination strategy than what's best for society, suggesting that subsidies are needed to align individual and collective interests.

What Problem Does This Paper Address?

The paper tackles the challenge of designing effective vaccination strategies in large populations. Specifically, it investigates the discrepancy between what individuals might choose to do regarding vaccination (driven by their own costs and benefits) and what would be optimal for the entire population's health and economic well-being. This is crucial because individual decisions, even if rational for the individual, might not lead to the best collective outcome, especially when dealing with contagious diseases where individual actions have externalities. The authors frame this as a comparison between a "Mean Field Equilibrium" (representing individual rational choices in a large population) and a "Social Optimum" (representing a centrally planned strategy for maximum societal benefit).

Methods

The authors employ a stochastic SIRV (Susceptible-Infected-Recovered-Vaccinated) model to represent virus propagation. Agents interact randomly, and susceptible individuals can choose to vaccinate. The core of their analysis involves:

  • Mean Field Limit: They analyze the system as the population size ($N$) approaches infinity. This allows them to treat the population's state as a continuous fluid governed by differential equations (Kolmogorov equations), simplifying the analysis from discrete agent interactions to continuous population dynamics.
  • Bellman Optimality Equations: To determine optimal strategies, they use dynamic programming principles, specifically Bellman equations, to calculate the expected future costs for an individual agent in different states (Susceptible, Infected).
  • Best Response Correspondence: They define a "best response" strategy for an individual agent given the vaccination strategy of the rest of the population.
  • Nash Equilibrium: They define a Symmetric Nash Equilibrium (SNE) where no individual agent can improve their outcome by unilaterally changing their strategy, assuming all other agents stick to the equilibrium strategy.
  • Social Optimum: They define a Social Optimum (SO) strategy that minimizes the total cost for the entire population.
  • Threshold Policies: A key finding is that both the Mean Field Equilibrium and the Social Optimum strategies take the form of "threshold policies." This means there's a critical time; before this time, agents vaccinate at a certain rate (either maximum or zero), and after this time, they switch to the other rate.

Datasets

Not discussed in the paper. The paper uses simulated parameters for its numerical evaluation.

Key Insights and Ideas

  • Individual vs. Collective Rationality: The most significant insight is the inherent conflict between individual incentives and collective well-being in vaccination decisions. Individuals, driven by their own vaccination costs and perceived risks, may choose to vaccinate less than what is socially optimal.
  • Threshold Strategies are Key: Both individual equilibrium and social optimum strategies simplify to threshold policies. This means decisions are not continuous but rather a switch from one extreme (vaccinate fully or not at all) to another at a specific point in time.
  • Social Optimum is More Aggressive: The socially optimal strategy always involves vaccinating for a longer period (a later "jump time" or earlier start time) than the mean-field equilibrium. This means society would benefit from more vaccination than individuals would voluntarily undertake.
  • Subsidies are Necessary: To align individual behavior with the social optimum, the paper strongly suggests that vaccination should be subsidized. This subsidy effectively lowers the individual cost of vaccination, encouraging people to vaccinate for longer periods, thus bridging the gap between individual and collective interests.

Key Results

  • The paper proves the existence of a unique Mean Field Equilibrium (MFE) and a Social Optimum (SO) strategy.
  • Both MFE and SO strategies are shown to be threshold strategies.
  • Crucially, the threshold for the social optimum strategy is always later than the threshold for the mean-field equilibrium (i.e., $\tau_{opt} > \tau_{MFE}$). This means individuals stop vaccinating earlier than is optimal for society.
  • Numerical simulations (Figure 2 and 3) demonstrate this gap. For specific parameters, the equilibrium strategy leads to a higher total cost (0.6824) compared to the social optimum (0.6818).
  • Figure 3 shows that the jump times for both strategies decrease as vaccination cost ($c_v$) increases, and the gap between them persists.
  • The paper quantifies the required subsidy. For instance, with $c_v = 0.8$, a subsidy of $g = 0.15$ is needed to make the equilibrium jump time match the optimal jump time.

Limitations and Future Work

  • Computational Complexity: While the mean-field approach simplifies analysis, computing the exact thresholds ($\tau_{MFE}$ and $\tau_{opt}$) in closed form as a function of all parameters is not feasible. However, the reduction to a single real number makes numerical computation tractable.
  • Model Simplifications: The SIRV model is a simplification of real-world epidemics. Future work could explore more complex disease dynamics, heterogeneous populations, or different interaction structures.
  • Mechanism Design: The paper touches upon mechanism design by suggesting subsidies. Further research could explore other policy interventions or more sophisticated subsidy structures.
  • Finite Populations: The analysis is based on the mean-field limit ($N \to \infty$). Extending these findings to finite populations, where individual decisions can have a more significant impact on the overall dynamics, would be a valuable direction.

Summary updated

Abstract

International audience
BibTeX
@inproceedings{gaujal2020vaccination,
  author = {Bruno Gaujal and Josu Doncel and Nicolas Gast},
  title = {{Vaccination in a Large Population: Mean Field Equilibrium versus Social Optimum}},
  booktitle = {NetGCooP},
  year = {2020},
  doi = {10.1007/978-3-030-87473-5\_6},
}
Figure