On theory testing in econometrics: Modeling with nonexperimental data

The paper has two main objectives. The first is to trace the roots of the traditional textbook approach to econometric modeling to two older statistical traditions, Fisherʹs experimental design paradigm, and Gaussʹs ‘theory of errors’, and argue that this approach is ill-suited for nonexperimental (observational) data because these older traditions are appropriate for modeling observable phenomena which can be viewed as generated by ‘nearly isolated’ systems. The second aim is to argue in favor of an alternative approach, which can be traced back to the Galton-Pearson biometric tradition, that takes account of the nonexperimental nature of economic data. It is argued that this approach often leads to more reliable empirical evidence and hence to a better evaluation of economic theories. The (weak) Efficient Market Hypothesis is used throughout to illustrate some of the arguments in the paper.