“Explaining Your Results” as Access-Biased Memory
Edward E. Leamer · Journal of the American Statistical Association · 1975
This article presents an inference-cum-memory model with uncertain parameter p and binomial data. The model is intended to represent the phenomenon especially common in the social sciences of explaining the results of a statistical calculation. The principal assumptions of the model are that experiments, not sufficient statistics, are stored in memory, and that the experiments are categorized by and accessed according to their implications for inference. The nonsample information thereby generated is weaker than it appears. In the context of this model, optimal memory interrogation involves search for previous experiments that slightly contradict the current evidence.