ABSTRACT
‘How do we learn from data about phenomena of interest?’ is the key question underlying empirical modelling in any discipline. The discussion that follows focuses primarily on economics, but similar problems and issues arise in all applied fields. ‘Learning from data’ has a long and ambivalent history in economics that can be traced back to the seventeenth century. Despite the fact that statistics and economics share common roots going back to ‘political arithmetic’ in the mid-seventeenth century, the relationship between them has been one of mutual distrust and uneasy allegiance. 2 They developed into separate disciplines after the demise of political arithmetic in the early eighteenth century, due mainly to:
the inability of political arithmeticians to distinguish between real regularities and artefacts in observed data, combined with
the irresistible proclivity to misuse data analysis in an attempt to make a case for one’s favourite policies.
These excesses led political arithmeticians to exorbitant speculations and unfounded claims based on misusing the information in the data to such an extent that it eventually discredited these data-based methods beyond salvation. 3