The hazards of underspecified models: the case of symmetry in everyday predictions.

Should one be more confident when predicting the whole (or an event based on a larger sample) from the part (or an event based on a smaller sample) than when predicting the reverse? The relevant literature on judgment under uncertainty argues that such predictions are symmetrical but that, as an empirical matter, people often fail to appreciate this symmetry. The authors show that symmetry in prediction does not necessarily hold. In addition to an empirical study involving predictions about soccer games, they develop a theoretical model showing that, at least for the ranges of numerical values usually found in everyday judgment problems, symmetry in predictions is uncommon when 2 different sample sizes are involved. The complexity of the theoretical model used in this analysis raises questions about model specification in judgmental research.