Discussion of ‘Data mining reconsidered’

Abstract

The Hoover—Perez description of the LSE [London School of Economics] general-to-specific methodology of model selection is formalized and analysed using the theory of model selection. Numerical evidence is provided to justify the claim that simple and elegant information criteria (which are easy to implement in applications) work at least as well, if not better, than the complicated algorithm attributed to the LSE methodology.