People predict the outcome that appears most representative of the evidence not according to probability.