Reciprocal altruism(1) can become established among selfish, unrelated
individuals if they use responsive strategies such as 'tit-for-tat'(2
-4). This result raises the fundamental question: how altruistic shoul
d one be? The problem is difficult to solve using current 'prisoner's
dilemma' based models because they allow only the discrete choice of c
ooperating or defecting. In reality, however, cooperation is rarely al
l-or-nothing. Furthermore, if cooperative investment is variable, a ne
w and more subtle kind of cheating becomes possible: individuals may i
nvest slightly less than their partner. A concern is that this 'short-
changing' will erode cooperative ventures. Here we show that cooperati
on can thrive despite variable investment through the new strategy of
'raise-the-stakes'. This strategy offers a small amount on first meeti
ng and then, if matched, raises its investment, something that no stra
tegy in the discrete model can do. We show that such behaviour can rea
dily invade a population of non-altruists and cannot be effectively ex
ploited. The practice of 'testing the water' rather than making sudden
cooperative 'leaps of faith' powerfully reinforces the stability and
effectiveness of reciprocity.