Evolutionary stable strategy

An evolutionarily stable strategy is a pre-programmed behavioural policy that cannot be improved upon by an alternative when it is already common in a population. Its importance lies in what it does not require: a stable pattern need not be best for the group, or even especially beneficial to its members. It persists because no deviant strategy can successfully invade under the conditions that prevail. The concept turns evolutionary stability into a question about relative success within a population, not about collective purpose.

Definition and Evolutionary Logic

The idea is introduced through the problem of whether an animal should fight. A decision to escalate or withdraw carries costs as well as possible gains; even waiting may be advantageous if it allows a contestant to grow stronger before challenging a rival. Game theory gives a systematic form to this calculation. A strategy is not a consciously chosen plan but a behavioural policy that an animal acts as if it were following. Describing it as a set of instructions is a way for an observer to analyse behaviour, not a claim about the animal’s deliberations.

A strategy is evolutionarily stable when, if most of a population follows it, no alternative strategy can do better. The population’s existing behaviour therefore helps determine which strategies can succeed: the best response depends on what the majority is doing. Selection penalizes a deviant that performs worse in that setting. A major environmental change can temporarily unsettle the population, perhaps producing oscillation, but once an ESS is established, it resists invasion. This is a conditional definition of stability, not a guarantee that the state is optimal, permanent under every environmental condition, or consciously maintained.

Aggression, Mixtures, and Asymmetry

The simplest illustration divides contestants into Hawk and Dove. Hawks fight without restraint and retreat only when seriously injured; Doves threaten but do not cause harm. When two Hawks meet, a contest can end in serious injury or death. When Hawk meets Dove, Dove retreats. Two Doves posture until one backs down, without either being hurt. The model initially assumes that contestants cannot identify a rival’s strategy in advance or use memories of previous encounters. These simplifications make the consequences of different strategies easier to compare.

The resulting ESS is not an all-Hawk or all-Dove population. Under the example’s pay-offs, it is a mixture of 7/12 Hawks and 5/12 Doves. That stable ratio has an average pay-off of 6 1/4 per individual, whether the individual is Hawk or Dove. Yet a population consisting entirely of Doves would give each individual a pay-off of 15. Even a mixture of 1/6 Hawks and 5/6 Doves would produce a higher average pay-off, 16 2/3 per contest. The point of the comparison is that a collectively more rewarding arrangement does not necessarily persist. A Dove-only arrangement is vulnerable to a Hawk that exploits the restraint of the others; the ESS endures not because it is best for its participants, but because it is resistant to internal treachery.

The model can also be extended beyond fixed Hawk and Dove types. A Retaliator begins with conventional threats, responds Hawk-like if attacked, and behaves Dove-like when facing another Retaliator. A Bully acts aggressively until someone hits back, then retreats. A Prober-retaliator mostly behaves like a Retaliator but tests an opponent with brief escalation, continuing if unchallenged and returning to threats if resisted. In a computer simulation of five strategies, Retaliator alone emerges as evolutionarily stable; Prober-retaliator comes close. Dove can be invaded by Hawks and Bullies, Hawk by Doves and Bullies, and Bully by Hawks. Dove can nevertheless do as well as Retaliator in a Retaliator population, so its numbers might drift upward; if Doves become common, strategies that exploit them gain an advantage. The result points toward a predominance of Retaliators and Prober-retaliators, perhaps with a small Dove minority, while allowing that individuals need not each follow one pure strategy.

The apparent balance of aggression changes with the costs and benefits of the contest. A large resource, such as an elephant seal’s harem, makes victory valuable and may help explain why contests can be severe; where time is especially costly, as for a Great tit feeding nestlings, delay weighs differently. Contestants may also be unequal in size or fighting equipment, may have different amounts to lose from injury, or may arrive at different times. Even an arbitrary asymmetry such as arriving first can help settle a contest: the distinction between resident and intruder may support a stable convention despite having no general advantage in itself. The ESS framework thus accommodates outcomes that depend not only on behavioural type but also on the structure and stakes of the encounter.

From Individual Contests to Gene Pools

The concept is explicitly distinguished from group selection. A population can appear to return to a stable equilibrium after disturbance, but this does not mean that more successful groups have displaced less successful ones. A cooperative arrangement may raise everyone’s pay-off and still be unstable if an individual can gain by breaking it. In the Dove example, the all-Dove conspiracy is more rewarding for its members than the Hawk-Dove ESS, but a Hawk can exploit it; the arrangement is therefore liable to collapse from within.

Human pacts offer a qualified contrast because participants can use conscious foresight to recognize long-term benefits and follow agreed rules. Even then, a short-term gain can make defection tempting. Petrol price-fixing among garage owners illustrates this vulnerability: one owner may cut prices for an immediate advantage, prompting competitors to follow, before the owners’ long-term interest in a new pact reasserts itself. The example sharpens the distinction between deliberate agreement and evolutionary stability. In the account of wild animals, where genes rather than foresighted individuals shape behaviour, a group-benefiting conspiracy is not an adequate explanation for a persistent strategy.

The same logic is extended from visible contests among bodies to interactions among genes. A gene pool is described as an evolutionarily stable set of genes: a set that cannot be invaded by a new gene. A gene arriving through mutation, reassortment, or immigration is usually penalized by selection, restoring the existing set; occasionally, an invader spreads, producing a transitional instability that ends in a new stable set. There may be more than one stable point, with populations sometimes shifting between them. Evolution can consequently appear as a series of steps from one stable plateau to another rather than a steady climb. The appearance of population-level self-regulation is, in this interpretation, an effect of selection among genes judged against the gene pool in which they occur.

This gene-level extension does not treat genes as independent agents constructing organisms alone. Genes cooperate and interact during development, and the effect of any one gene depends on other genes and the external environment. A gene associated with a feature such as leg length does not build a leg by itself; it may, other things being equal, shift the outcome relative to an alternative allele. The explanation distinguishes a gene’s possible long-term persistence through generations from its interdependent role in producing a body. The visible stability of a well-integrated organism is linked to an evolutionarily stable set of genes, while the interactions that maintain that integration mostly occur within bodies and developing cells.

Applications and Limits of the Model

The strategy framework is also applied to sexual behaviour, where the book considers two female strategies, Coy and Fast, and two male strategies, Faithful and Philanderer. These are modelled as unconscious behavioural programs, not as deliberate choices. If Philanderers become common, Coy females avoid the cost of producing offspring with a partner who will not remain; in the model, their pay-off is zero against a population of Philanderers, compared with an average of -5 for Fast females. As Coy females become common, Philanderers lose the advantage they had with Fast females, while a Faithful male can gain a pay-off of +2 by mating with a Coy female. The changing success of each strategy depends on the frequency of the strategies in the other sex.

For the example’s specified pay-offs, the stable state consists of 5/6 Coy females and 5/8 Faithful males. These proportions need not describe two distinct kinds of individual: each individual could instead spend the corresponding fraction of time following each policy. A deviation by one sex changes the strategic environment faced by the other, which can in turn penalize the original deviation. The model shows how a pattern that resembles coordinated restraint can arise without a conspiracy of females; under these assumptions, Coy behaviour can benefit the genes associated with it. The particular fractions depend on the arbitrary pay-offs chosen, and the argument presents them as model results rather than universal measurements.

Computer simulations of Sucker, Cheat, and Grudger strategies show that there can be more than one ESS. Grudgers groom strangers and individuals who have groomed them, but refuse future grooming to an individual who has cheated them. When Grudgers are rare among Cheats, they spend effort grooming others without receiving much in return; after they reach a critical proportion, encounters among Grudgers become frequent enough for them to outperform Cheats. The simulation identifies both a predominantly Grudger population and a predominantly Cheat population as evolutionarily stable: neither Sucker nor the alternative strategy can invade either state. Which stable state is more likely to be reached depends on the assumed pay-offs and the size of its zone of attraction. A Cheat-dominated population might be more likely to go extinct than a Grudger population, but that consequence does not alter its ESS status within the model.

The concept also changes how the results of the Prisoner’s Dilemma tournaments are interpreted. In the second tournament, participants knew that nice and forgiving strategies had done well in the first, so some submitted similarly cooperative strategies while others tried to exploit the expected abundance of such opponents. Tit for Tat, submitted by Anatol Rapoport, won again, scoring 96 per cent of the benchmark score; all but one of the top fifteen strategies were nice, and all but one of the bottom fifteen were nasty. Yet Tit for Two Tats, which would have won the first tournament if entered, did not win the second, because more subtle strategies could exploit its exceptional forgiveness. A strategy’s tournament performance depended on which opponents happened to be submitted.

That dependence distinguishes success in a round-robin tournament from success under natural selection. In a round-robin, each strategy is tested against the same submitted set, and its score measures how well it competes against that particular field. Tit for Tat was robust across the submitted strategies, but the field itself was arbitrary: in the first tournament, about half the entries were nice, and six of fourteen were nasty. A different balance could have changed the ranking. The ESS test asks instead whether a strategy performs well when it is already numerous—whether it can prosper in a population dominated by copies of itself. In Darwinian terms, winnings are offspring rather than points, so persistence in a self-dominated population is the relevant kind of robustness. Robert Axelrod’s later simulation used the same 63 strategies from Round 2 as an evolutionary succession: they began in equal representation, received offspring in proportion to their winnings, and changed the strategic climate as some became common and others went extinct. The ESS concept therefore supplies a standard less dependent on the accidental composition of a tournament field.

Explanatory Reach

Across these applications, the ESS is a criterion for persistence, not a moral endorsement or a claim that evolution plans ahead. It explains why a locally advantageous deviation may fail when most individuals behave differently, why an arrangement beneficial to everyone can be fragile, and why several stable states may exist under the same broad model. Its conclusions remain conditional on the strategies, pay-offs, and environmental assumptions used to define each case; changing those conditions can change the stable outcome.

The framework also belongs within a wider account of natural selection that separates an adaptation’s plausible function from adequate evidence for its history. Illustrations of altruism can clarify what is meant by selfish and altruistic behaviour, but examples alone are not serious evidence for a generalization. Similarly, an account of animal aggregation as a population census may sound plausible while remaining weakly supported if the same observations admit other explanations. An ESS can show how a pattern could resist invasion under specified conditions; that logical result is not, by itself, proof that the pattern evolved in nature by that route.

This distinction matters when the analysis reaches beyond ordinary behaviour. An ancestral Caddisfly house can be treated as an adaptation if hereditary differences in house properties were available for selection, even though the relevant genes act through development and behaviour. A fluke’s influence on a snail’s shell is proposed as another case in which interests diverge: the snail’s genes benefit from reproduction, while fluke genes need not leave through the snail’s offspring. The broader claim is that selection can favour genetic effects extending beyond the body that carries the genes. These examples do not redefine the ESS, but they show why evolutionary explanations must identify whose transmission and reproductive prospects count, and must distinguish a proposed mechanism from a tested result.

The concept’s enduring analytical force comes from refusing to confuse stability with harmony. A population may settle into a pattern that is costly, unequal, or vulnerable to extinction; it may also maintain a mixture rather than converge on one uniform type. What matters is whether an alternative can do better against the strategy already present. In moving from contests between animals to interactions within bodies and among genes, the ESS makes that question portable while keeping its answer dependent on the particulars of each case.