Proceedings of the Annual Meeting of the Cognitive Science Society

The Hot Stove Effect is a negativity bias resulting from theadaptive character of learning. The mechanism is that learn-ing algorithms that pursue alternatives with positive estimatedvalues, but avoid alternatives with negative estimated values,will correct errors of overestimation but fail to correct errorsof underestimation. Here we generalize the theory behind theHot Stove Effect to settings in which negative estimates do notnecessarily lead to avoidance but to a smaller sample size (i.e,a learner selects fewer of alternative B if B is believed to be in-ferior but does not entirely avoid B). We demonstrate formallythat the negativity bias remains in this set-up. We also showthat there is a negativity bias for Bayesian learners in the sensethat most such learners underestimate the expected value of analternative.