Learning from Data: the Role of Error in Statistical Modeling and Inference ∗

A widely held view in economics is that the current untrustworthiness of empirical evidence and the inability to forecast economic phenomena is largely due to the fact that the economy is too complicated and the resulting data are too heterogeneous to be amenable to statistical modeling. The primary objective of this paper is to argue that whether economic data are amenable to statistical modeling is largely an empirical issue and the current untrustworthiness is mainly due to an inadequate implementation of statistical modeling and inference. The latter stems from two main sources: (i) commingling the statistical and substantive premises of inference, and (ii) neglecting the validation of the latter the secures the reliability of inference. It is argued that error statistics, ar efined/extended frequentist perspective, attains learning from data about phenomena of interest by employing reliable and effective procedures to establish trustworthy evidence.