Empirical content as a criterion for evaluating models.

Abstract

Hypotheses derived from models can be tested in an empirical study: If the model reliably fails to predict behavior, it can be dismissed or modified. Models can also be evaluated before data are collected: More useful models have a high level of empirical content (Popper in Logik der Forschung, Mohr Siebeck, Tübingen, 1934), i.e., they make precise predictions (degree of precision) for many events (level of universality). I apply these criteria to reflect on some critical aspects of Kirsch's (Cognit Process, 2019. https://doi.org/10.1007/s10339-019-00904-3 ) unifying computational model of decision making.