Eagle Dove Game

The Hawk–Dove game is a deliberately simple model of animal conflict that explains why selection need not favor either universal aggression or universal restraint. By assigning costs and benefits to contests, it makes the decisive point that the successful strategy depends on what others do: a stable population balance can persist not because it maximizes everyone’s welfare, but because no individual strategy can profitably invade it. The model therefore distinguishes evolutionary stability from collective advantage and provides a framework for interpreting aggression as conditional, costly behavior rather than an unvarying impulse.

The basic contest and its pay-offs

The model classifies hypothetical contestants as Hawks or Doves, names borrowed from human convention rather than the behavior of the birds themselves. Hawks fight without restraint and retreat only when seriously injured. Doves threaten in a conventional, non-injurious way. A Hawk facing a Dove wins as the Dove retreats; two Hawks continue until one is seriously injured or dead; two Doves posture for a long time until one tires or backs down. At first, neither contestant can identify the other’s strategy or use memories of earlier encounters to guide a decision.

To compare strategies, the model assigns 50 points for winning, 0 for losing, minus 100 for serious injury, and minus 10 for time wasted in a prolonged contest. These are not meant as literal measurements: they stand for consequences convertible into success in leaving genes in the population. The important calculation is thus not simply which contestant wins a particular fight, but which strategy earns the better average return under a given population composition.

In a population made entirely of Doves, each individual wins about half of its contests, earning 50 points for a win less the 10-point time cost, or 40; it loses the other half and pays the time cost, for a return of minus 10. The average is therefore 15 points per contest. That apparently comfortable equilibrium is vulnerable: a lone Hawk meets only Doves, wins every contest, and earns 50 points. Hawk genes can consequently spread while their advantage lasts.

But the advantage changes when Hawks become common. In an all-Hawk population, each Hawk wins about half of its fights and loses half, with the losing contestant seriously injured. The average pay-off becomes minus 25. A Dove among Hawks loses every contest but avoids injury, receiving 0; Doves then have the advantage and can spread. Neither pure strategy is evolutionarily stable by itself: each creates conditions in which a rare alternative can do better.

A stable mixture, not a social optimum

With the stated, deliberately arbitrary points, the population settles at a stable ratio of seven Hawks to five Doves: seven-twelfths Hawks and five-twelfths Doves. At that balance the average pay-offs of the two strategies are equal, so selection does not favor one over the other. If the Hawk share rises beyond its stable level, Doves gain an advantage and the balance tends back; small oscillations around the point need not become a continuing sweep from one extreme to the other.

This equilibrium may resemble group selection because the population has a stable composition, but the explanation is different. It does not depend on one group outperforming another. At the stable ratio, both Hawks and Doves average 6¼ points per contest. Every individual would do better in an all-Dove population, where the average is 15; a population with one-sixth Hawks and five-sixths Doves would do better still, averaging 16⅔. Yet a cooperative arrangement of Doves cannot hold against a mutant Hawk, which would exploit the restraint of the others and earn 50 points per contest. The Hawk–Dove game thus separates what would benefit everyone if collectively maintained from what can resist internal exploitation. An evolutionarily stable strategy is not necessarily the best outcome for its participants; it is stable because a deviant strategy cannot profitably break in.

The same vulnerability explains why conscious human agreements are not identical to evolved strategies. People can understand and obey a pact because they foresee its long-term benefit, but short-term gains from breaking it can still undermine cooperation. The game’s contrast is between a population-level outcome maintained by the incentives facing individual contestants and a beneficial arrangement that depends on participants continuing to restrain themselves.

From fixed types to conditional behavior

The seven-to-five result can represent a stable polymorphism: a lasting ratio of Hawk-associated to Dove-associated genes in the gene pool. Alternatively, every individual could use both behaviors, choosing Hawk with a probability of seven-twelfths and Dove with a probability of five-twelfths on each contest. For this mixed strategy to work, the choice must be unpredictable to an opponent. A fixed sequence of seven Hawk contests followed by five Dove contests would be exploitable, since rivals could choose Hawk whenever they knew the individual would play Dove.

The model can also be developed beyond two fixed strategies. It is explicitly an idealization, not a claim that real animals fall into only two behavioral types. One extension is Retaliator: it begins like a Dove, threatening rather than launching a savage attack, but retaliates if attacked. It behaves like a Hawk against a Hawk and like a Dove against a Dove. Bully begins like a Hawk but runs away as soon as someone hits back. Prober-retaliator resembles Retaliator but occasionally tests an opponent with a brief escalation, continuing if the opponent does not resist and reverting to conventional threat if the opponent does.

In a computer simulation involving Hawk, Dove, Retaliator, Bully, and Prober-retaliator, Retaliator emerges as the only evolutionarily stable strategy; Prober-retaliator comes close. Dove can be invaded by Hawks and Bullies, Hawk by Doves and Bullies, and Bully by Hawks. No alternative does better in a population of Retaliators, although Dove does equally well there. If Doves become appreciably more numerous, Prober-retaliators, along with Hawks and Bullies, gain an advantage against them. The model therefore allows a leading mix of Retaliators and Prober-retaliators, a small Dove minority, and possibly gentle fluctuations, rather than requiring each individual to play one unchanging strategy throughout life.

Biological interpretation and limits

These strategic calculations offer an account of the “gloved fist” of animal aggression: display and restraint can coexist with the capacity to escalate, and the balance depends on the pay-offs of the particular contest. An elephant seal competing for near-monopoly access to a large harem faces a very valuable prize, making winning worth more and serious injury more tolerable. For a Great tit feeding nestlings, by contrast, time may be the more pressing cost: it needs to catch an average of one prey every thirty seconds, so even daylight spent in a fight can matter more than the risk of injury.

The examples do not supply enough information to assign realistic numerical values to all these costs and benefits in nature. The points used in the model are arbitrary, and conclusions that depend only on a chosen set of numbers require caution. Its broader claims are more limited: selection tends to favor evolutionarily stable strategies; such stability is not the same as the collectively best arrangement; and intuitive judgments about the best behavior can mislead when an opponent’s response changes the pay-off.

A framework for contests and cooperation

The game sits within a wider argument about conflict that treats decisions to begin, escalate, or cool a fight as cost-benefit problems. The model does not imply that an animal consciously calculates outcomes. A strategy is a pre-programmed behavioral policy: describing it as a set of instructions is a way of representing how an animal behaves, not a claim that it deliberates in words. The pay-off framework instead asks how inherited tendencies fare across the encounters they generate.

The distinction between individual incentives and joint benefit also reaches beyond animal combat. Human agreements can produce gains for all parties, but remain exposed to short-term defection; civil disputes may likewise be reframed from apparent zero-sum contests into mutually beneficial cooperation. In the book’s account, the 1977 match between Bristol and Coventry changes character when news that Sunderland has lost makes a draw sufficient for both teams to avoid relegation: formerly competing sides then have a shared reason not to score. The point is not that the Hawk–Dove model directly explains every such situation, but that outcomes depend on the structure of incentives and on what participants know about the consequences of their moves.

That emphasis on rules and circumstances also marks the model’s limits. A calculation that assumes contestants cannot identify an opponent’s type, for example, differs from one in which conditional behavior can respond to an opponent’s actions. Adding Retaliator, Bully, or Prober-retaliator changes the contest because each strategy makes behavior contingent on what happens during the encounter. The model’s value lies not in presenting a final portrait of every animal fight, but in showing why aggression, restraint, and cooperation must be assessed in relation to opponents, costs, and the stability of the resulting behavior.