Score contribution per author:
α: calibrated so average coauthorship-adjusted count equals average raw count
In this laboratory experiment, children and teenagers learn the composition of balls in an urn through sampling with replacement. We find significant aggregate departures from optimal Bayesian learning across all ages, but also important developmental trajectories. Two inference-based and two heuristic-based strategies capture the behavior of 65% to 90% of participants. Many of the youngest children (K to 2nd grade) base their decisions only on the last piece of information and use evolutionary heuristics (such as the “Win-Stay, Lose-Switch” strategy) to guide their choices. Older children and teenagers are gradually able to condition their decisions on all previous information but they often fall prey of the gambler’s fallacy. Only the oldest participants display optimal Bayesian reasoning. These results are modulated by task complexity, and Bayesian reasoning is evidenced earlier when inferences are simpler.