Survival of dominated strategies under imitation dynamics

Panayotis Mertikopoulos, Yannick Viossat

Abstract

In evolutionary game theory, it is traditionally assumed that imitation-based dynamics inevitably lead to the extinction of strictly dominated strategies—choices that are suboptimal regardless of the opponent's behavior. This paper challenges that consensus by examining the micro-foundations of how individuals actually imitate. By modeling revision protocols that incorporate social biases such as sampling and conformity, the authors demonstrate that "irrational" strategies can persist indefinitely. Through the use of the Hypnodisk game and Lyapunov stability analysis, the research proves that when imitation protocols favor rare or very common strategies, dominated behaviors can find a permanent niche within a population.

Introduction: The Rationality of Evolution

In classical game theory, a strictly dominated strategy is one that a rational agent should never choose because a superior alternative always exists. Evolutionary Game Theory (EGT) translates this concept into a population context: if agents copy successful peers, the "unfit" strategies should eventually be outcompeted and vanish.

Historically, EGT has distinguished between two types of dynamics:

  1. Imitative Dynamics: Agents replicate the behavior of successful peers (e.g., Replicator Dynamics). These are generally thought to eliminate dominated strategies.
  2. Innovative Dynamics: Agents occasionally experiment with new strategies or "best-respond" to the current environment. These can allow dominated strategies to survive by providing an advantage to rare behaviors.

This paper argues that the perceived "rationality" of imitative dynamics is actually an artifact of how they are modeled. By looking closer at the mechanics of human imitation, the authors show that the sharp contrast between imitation and innovation is an illusion.

Methods: Deconstructing Revision Protocols

To investigate this, the authors move away from abstract population equations and focus on revision protocols—the specific rules an individual follows when deciding to change their strategy. They decompose the imitation process into two distinct steps:

  • Step 1 (Selection): A "revising agent" observes other individuals in the population to see which strategies are being played.
  • Step 2 (Adoption): The agent decides whether to switch to one of the observed strategies based on their relative payoffs.

The core innovation of this study is the introduction of non-monotone selection rules. In standard models, the probability of encountering a strategy is exactly proportional to its frequency. However, the authors consider more realistic social protocols:

  • Sampling: An agent meets a small group and identifies the unique strategies present. Mathematically, this disproportionately favors rare strategies because they are more likely to be "noticed" than their raw frequency suggests.
  • Conformity: An agent picks the most frequent strategy within a sampled group, which favors common strategies.

To test the survival of dominated strategies, the authors utilize the Hypnodisk game—a non-linear game characterized by cyclic population behavior—and introduce a "feeble twin" (a strategy that is a slightly worse version of an existing one). They then apply Lyapunov stability analysis and numerical simulations to track the long-term persistence of these dominated strategies.

Key Insights: The Rarity Bonus and Cycling

The research identifies a phenomenon termed the Rarity Bonus. In standard evolutionary models, a strategy's growth is tied strictly to its payoff. However, if an imitation protocol involves sampling multiple people, the probability of noticing a rare strategy is higher than its actual frequency in the population. This bias acts as a lifeline.

Crucially, dominated strategies do not survive by reaching a stable equilibrium. Instead, they survive through population cycling. When a dominated strategy becomes dangerously rare, the rarity bonus from the imitation protocol kicks in, pushing its growth rate above its more successful rivals. By the time the strategy becomes common enough for its poor payoff to become a liability, the cycle repeats, preventing extinction.

This suggests that imitation is not inherently "rational." The "monotone" property—where better payoffs always lead to higher growth—is a legacy of biological selection models. Human imitation, driven by social biases and sampling limitations, naturally creates niches where "unfit" strategies can endure.

Results and Implications

The paper's findings are summarized in several key results:

  • Theorem 1: The authors provide a formal proof that for any imitation protocol satisfying "Advantage to Rarity" or "Advantage to Frequency," there exist games where a strictly dominated strategy persists forever for a large set of initial conditions.
  • Significant Population Share: Simulations indicate that a dominated strategy can maintain a non-trivial presence, sometimes accounting for 10% to 20% of the population, even when the margin of domination is significant.
  • Non-Nash Persistence: In certain scenarios, the population converges to a "rest point" that is not a Nash Equilibrium, such as a stable 50/50 split between a strategy and its dominated twin.

Limitations and Future Directions

While the paper proves the possibility of survival, it focuses on games that exhibit natural cycling (like Rock-Paper-Scissors variants). Whether dominated strategies can survive in "tamer" games that typically converge to a single point remains an open question. Additionally, the study uses deterministic, continuous-time dynamics; in small, real-world populations, stochastic "noise" might eliminate rare strategies before the rarity bonus can take effect. Future research may explore the "domination margin"—the exact point at which a strategy becomes so poor that no amount of social bias can save it from extinction.

Summary updated

Abstract

The literature on evolutionary game theory suggests that pure strategies that are strictly dominated by other pure strategies always become extinct under imitative game dynamics, but they can survive under innovative dynamics. As we explain, this is because innovative dynamics favour rare strategies while standard imitative dynamics do not. However, as we also show, there are reasonable imitation protocols that favour rare or frequent strategies, thus allowing strictly dominated strategies to survive in large classes of imitation dynamics. Dominated strategies can persist at nontrivial frequencies even when the level of domination is not small.
BibTeX
@misc{mertikopoulos2022survival,
  author = {Panayotis Mertikopoulos and Yannick Viossat},
  title = {{Survival of dominated strategies under imitation dynamics}},
  year = {2022},
  eprint = {2209.08416},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2209.08416},
}
Figure