Finite Population Dynamics Resolve the Central Paradox of the Inspection Game

TL;DR

The paper resolves a central paradox of the Inspection Game by embedding it in a finite-population evolutionary framework subject to demographic noise. It derives deterministic replicator dynamics for asymmetric populations and analyzes stochastic fixation in finite populations, showing that high absolute penalties can suppress crime by biasing fixation toward criminal extinction, while a surprising second regime with near-equal penalties also suppresses crime. In the asymptotic limit of extreme population-size asymmetry (few inspectors), the long-run outcome becomes independent of citizen payoffs and hinges on the initial crime frequency relative to the deterrence threshold, linking initial conditions to policy effectiveness. Collectively, the work demonstrates that deterrence effectiveness emerges from the interplay between deterministic dynamics, demographic noise, and initial conditions, rather than static equilibria alone, with important implications for crafting robust crime-prevention policies.

Abstract

The Inspection Game is the canonical model for the strategic conflict between law enforcement (inspectors) and citizens (potential criminals). Its classical Mixed-Strategy Nash Equilibrium (MSNE) is afflicted by a paradox: the equilibrium crime rate is independent of both the penalty size ($p$) and the crime gain ($g$), undermining the efficacy of deterrence policy. We re-examine this challenge using evolutionary game theory, focusing on the long-term fixation probabilities of strategies in finite, asymmetric population sizes subject to demographic noise. The deterministic limit of our model exhibits stable limit cycles around the MSNE, which coincides with the neutral fixed point of the equilibrium analysis. Crucially, in finite populations, demographic noise drives the system away from this cycle and toward absorbing states. Our results demonstrate that high absolute penalties $p$ are highly effective at suppressing crime by influencing the geometry of the deterministic dynamics, which in turn biases the fixation probability toward the criminal extinction absorbing state, thereby restoring the intuitive role of $p$. Furthermore, we reveal a U-shaped policy landscape where both high penalties and light penalties (where $p \approx g$) are successful suppressors, maximizing criminal risk at intermediate penalty levels. Most critically, we analyze the realistic asymptotic limit of extreme population sizes asymmetry, where inspectors are exceedingly rare. In this limit, the system's dynamic outcome is entirely decoupled from the citizen payoff parameters $p$ and $g$, and is instead determined by the initial frequency of crime relative to the deterrence threshold (the ratio of inspection cost to reward for catching a criminal). This highlights that effective crime suppression requires managing the interaction between deterministic dynamics, demographic noise, and initial conditions.

Figures (5)

• Figure 1: Phase plane trajectories showing frequency of criminals $x$ and frequency of inspectors who inspect $y$. The constant of motion $H$ defining each trajectory is set by the initial condition $x_0=y_0 = 0.5$. The figure displays the influence of the population ratio $\alpha$ on the orbit geometry for two different inspection thresholds $g/p=0.1$ (left panel) and $g/p=0.8$ (right panel). In each panel, three trajectories for $\alpha = 0.1$, $0.5$, and $1$ are shown. The other parameters are fixed at $g=4$, $r=4$ and $k=1$. Trajectories are counterclockwise and centered at the neutral fixed point $(x^*, y^*) =( k/r, g/p)$, indicated as a filled circle.
• Figure 2: (Left panel) The minimum criminality incidence $x_{min}$, the maximum criminality incidence $x_{max}$, and the neutral fixed point $x^*$. (Right panel) The minimum inspector frequency $y_{min}$, the maximum inspector frequency $y_{max}$, and the neutral fixed point $y^*$. Both panels are shown as a function of the inspection threshold $g/p$. The initial condition is $x_0=y_0 = 0.5$. The other parameters are fixed at $\alpha=0.1$, $g=4$, $r=4$, and $k=1$.
• Figure 3: (Left panel) Phase plane trajectory showing criminal frequency ($x$) and inspectors who inspect frequency ($y$) for the population size ratio $\alpha =0.001$ and $k/r=0.2$. Trajectory is counterclockwise and centered at the neutral fixed point $(0.2, 0.4)$, indicated as a filled circle. (Right panel) The minimum ($x_{min}$) and the maximum ($x_{max}$) criminality incidence, and the neutral fixed point ($x^*$) as a function of $k/r$ for $\alpha =0.001$ The initial condition is $x_0=y_0= 0.5$. The other parameters are fixed at $p=10$, $g=4$, and $k=1$.
• Figure 4: Probability that criminality is extinct $\rho_0$ as a function of the inspection threshold $g/p$ for a citizen population size of $N=1000$ and an inspector population size of $M=100$ ($\alpha=0.1$). (Left panel) Varies the penalty parameter $p$ for different fixed values of the crime gain $g=5,10,15,20$. (Right panel) Varies the crime gain $g$ for different fixed values of the penalty $p=10,50,100, 200$. The initial condition is $x_0=y_0 = 0.5$. The other parameters are fixed at $r=4$ and $k=1$.
• Figure 5: Probability that criminality is extinct $\rho_0$ as a function of the population size ratio $\alpha$ for a citizen population size of $N=1000$ and different proportion of criminals in the initial population ($x_0=0.1,0.3,0.5,0.8$). (Left panel) Penalty parameter $p=10$. (Right panel) Penalty parameter $p=100$. The initial condition for the inspector population is $y_0 = 0.5$. The other parameters are fixed at $g=4$, $r=4$ and $k=1$.