Collectively stable strategy

A collectively stable strategy is a strategy, or mixture of strategies, that can persist as a cooperating collective even when it fails the stricter test of an evolutionarily stable strategy. The distinction matters because cooperation may be resilient against direct invasion by hostile behavior yet still depend on a fragile balance among strategies; a population can instead settle into persistent defection, with its outcome shaped by frequency thresholds and chance rather than any collective intention.

From Tit for Tat to collective stability

In the repeated Prisoner’s Dilemma, Tit for Tat appears robust because it answers cooperation with cooperation and can resist invasion by nasty strategies once it is common. But it is not technically an evolutionarily stable strategy. That stricter criterion requires that a common strategy cannot be invaded by a rare mutant strategy. Other nice strategies, including Always Cooperate, behave just like Tit for Tat when they meet a population of nice players: all cooperate, so none gains a selective advantage over the others. Such alternatives can therefore drift into the population without being noticed.

The apparent equivalence among nice strategies has a consequence. Always Cooperate is vulnerable to a nasty strategy such as Always Defect, which can repeatedly obtain the high Temptation score against it. Nasty strategies can thus keep excessively forgiving behavior in check. Tit for Tat’s practical resilience does not come from being invulnerable in the strict sense; it comes from its capacity to respond to defection, unlike unconditional cooperation.

Mixtures, forgiveness, and invasion

The account therefore treats a mixture of basically nice but retaliatory strategies as roughly equivalent to an evolutionarily stable strategy in practice, while allowing that such a mixture may include a small amount of nastiness. Robert Boyd and Jeffrey Lorberbaum examined one possible combination: Tit for Two Tats together with Suspicious Tit for Tat. The latter defects on its first move, making it technically nasty, but then behaves like Tit for Tat.

In a population dominated entirely by Tit for Tat, Suspicious Tit for Tat does poorly: its opening defection provokes an unbroken sequence of mutual retaliation. Against Tit for Two Tats, however, the more forgiving strategy interrupts that cycle. Both players can finish with at least the benchmark score for mutual cooperation, while Suspicious Tit for Tat receives a bonus from its initial defection. Boyd and Lorberbaum showed that the two strategies could prosper in each other’s company and, together, invade a population of Tit for Tat. The account presents this as one possible mixture, not the only one: other combinations of slightly nasty strategies with nice, very forgiving strategies may also invade. The example shows why cooperation is not secured simply by labeling a strategy nice; what matters is how strategies interact.

Collective stability and competing endpoints

Robert Axelrod recognized that Tit for Tat was not a strict evolutionarily stable strategy and coined the term “collectively stable strategy” for it. More than one strategy can be collectively stable at once, and which one comes to dominate is partly a matter of luck. Always Defect is also stable: in a population already dominated by it, no other strategy does better. The system can therefore be bistable, with one stable point at Always Defect and another at Tit for Tat or a mixture of mostly nice, retaliatory strategies. Once either point becomes dominant, it tends to remain so.

This is a conditional stability, not a promise that cooperation will always prevail. The proportion of Tit for Tat needed to outperform Always Defect depends on the payoffs set for the particular game. There is a critical frequency—a knife-edge separating the conditions under which selection favors more Tit for Tat from those under which it favors more Always Defect. The position of a population relative to that threshold matters: when it lies on the defection side, the few cooperative players may not meet one another often enough to benefit mutually, and selection pushes the population further toward defection.

A population might cross the threshold through random drift, after which selection could carry it toward the cooperative stable point. But populations have no shared will or intention and cannot deliberately leap the threshold. The contrast is central to the concept: collectively beneficial cooperation may be stable under some population conditions, yet neither its benefits nor its persistence guarantee that it will arise. Collective stability describes a possible equilibrium among strategies, not a conscious agreement or a general evolutionary preference for the good of the group.