REF · CLASSIC MODEL

Replicator DynamicsREPLICATOR DYNAMICS

Taylor & Jonker, 1978 — writing "what catches on" as an equation

Referenced from: Topic 29, Evolutionary Games & the Origin of Cooperation

01The Question It Poses

Classical game theory assumes rational players: they know all the options, can compute the payoffs, and can reason about what the other side is thinking. That is a workable assumption for analysing a negotiation. It is an extravagant one for a colony of bacteria, the trees in a forest, or ten thousand retail traders who mostly copy whatever is working.

The latter is exactly what biologists ran into in the 1970s: animals cannot calculate, and yet the distribution of their behaviour looks suspiciously like the output of a calculation. Is there a mechanism that pushes a population towards the "computed" outcome without anyone in it understanding anything?

Replicator dynamics answers yes, and it needs only two ingredients: what does well gets copied more, and how well something does depends on what everyone else is doing. The copying can be reproduction, imitation, or capital flowing in; the mechanisms differ and the equation is the same.

02The Rules

  1. The population contains several strategies. Write the current share of strategy i as xi; all shares sum to 1.
  2. Each strategy has a payoff fi. The crucial part: fi is not a constant. It depends on the current distribution of shares — the same strategy is worth different amounts in different populations.
  3. Compute the population average payoff = the sum of each share times its own payoff.
  4. The rate of change of each share = xi × (fi). Above average and it grows, below average and it shrinks, exactly average and it holds.
  5. No mutation, no innovation: a strategy already at share 0 can never appear.

That leading xi factor in rule 4 is easy to skip past, and it sets the whole character of the model: however excellent a strategy is, if few are playing it right now it also spreads slowly. Good things do not roll out overnight — they have to survive the phase where the base is too small. The same factor explains rule 5: zero times anything is zero.

One more thing worth stating plainly: this equation describes no individual at all. It only tracks shares. Ask it whether a particular person will switch strategies and it has no answer, and no ambition to have one.

Bar height = the strategy's current payoff; bar width = its current share strategy A strategy B strategy C below average → share falls also below, but a small base above average → share rises rate of change = share × (own payoff − population average) So B is falling too, but more slowly than A: its base is small, and the product is small
The dashed line is the population average. Strategies above it gain share, those below lose it — and the line itself moves as the shares move, which is where all the difficulty comes from.

03What You See When It Runs

With two strategies you can draw everything on a single line segment: the horizontal axis is the share of strategy A, from 0 to 1, and the arrows show which way it travels. Different payoff structures produce only four possible pictures.

One equation, three payoff structures, three phase lines ① Dominance — A beats B at every mix all B all A ② Coexistence — whichever is common does worse (e.g. Hawk-Dove) push it away and it returns ③ Bistability — whichever is common does better (e.g. a coordination game) the watershed: cross it and you cannot come back
Filled dot = stable (nudge it and it returns), open dot = unstable (nudge it and it leaves). The fourth possibility is the neutral case, where both strategies always earn the same and every point on the line is a fixed point.

Add strategies and it stops being a line. Three strategies live inside a triangle (each corner is "everyone plays this one"), and the usual endings include converging to a corner, converging to a point on an edge, or circling forever. Rock-paper-scissors is the circling kind — it has no resting place, only a wheel that keeps turning. The three male morphs of the Californian side-blotched lizard (Uta stansburiana) run exactly such a cycle, with a period of roughly six years (Sinervo and Lively, 1996).

With four or more strategies, replicator dynamics can even become chaotic: trajectories never repeat and long-run behaviour is unpredictable. So "evolution tends towards stability" is not a general truth — it is a special case of certain payoff structures.

04What It Explains

The most direct application is why the proportion of a behaviour sits at some peculiar value. Why an animal population always contains a small band of cheats rather than all of them or none; why a low-price strategy and a high-service strategy coexist in an industry for decades; why some bad habit that hurts everybody cannot be swept away. That is what the first phase line is about.

It also explains why a head start is decisive in some settings and worthless in others. In the third phase line (bistability), which side of the watershed you start on determines everything: keyboard layouts, charging connectors, the early users of a social platform all have this structure. In the second (coexistence), the starting point is irrelevant — wherever you begin you return to the same mix, and spending money to win the opening is simply wasted.

There is a less obvious contribution too: it demotes rationality from an assumption to a result. Nobody in a replicator model computes anything, and yet the final mix frequently lands exactly on the point game theory derives from rational play. That means the reach of the Nash equilibrium is wider than its derivation suggests — you do not need the players to be clever, you only need a selection mechanism to exist.

What It Cannot Explain

Further Reading