Fig. 4
Monkeys show the predicted tradeoff (model-independent analysis). a–d Changes in undermatching, variance of choice and harvesting efficiency over experimental days. e–g Monkeys show the bias-variance tradeoff. e Undermatching (1− the slope of matching behavior) and the variance of choice are significantly correlated negatively [permutation test: p < 10−8 for Monkey F and p < 10−8 for Monkey G]. This means that the monkeys trade off the bias (undermatching) against the variance of estimation, as predicted by our model. f By reducing variance of choice, monkeys improved their harvesting performance. The harvesting efficiency is significantly negatively correlated with the variance of choice [permutation test: p < 10−8 for Monkey F and p < 10−5 for Monkey G]. g Although increasing bias is harmful, benefits from reducing the variance of choice outweigh the costs [permutation test: p < 0.07 for Monkey F and p < 0.01 for Monkey G]. Note that in these figures squared root of variance is shown as variance
