The Case against “Indirect” Statistical Inference

Abstract

How to infer from uncontrolled data the numerical values of the parameters of an econometric model? In interwar econometrics, this problem of statistical induction was tackled in various ways. Wassily Leontief’s inter-industry method offers a case study of a nonprobabilistic approach to this problem of the statistical determination of a model’s parameters. Leontief was an outspoken critic of the probabilistic approach to econometrics and presented his interindustry device as an alternative. Focusing on his book published in 1941, The Structure of American Economy, 1929–1929, which gathered a decade of works since 1932, I show how it offered an original answer to the inferential problem of the parameters’ statistical determination. Between what Jan Tinbergen called, in the same period, the “historical method,” based on regression techniques applied to time series, and the “structural method,” relying on direct measures from cross-sectional data and controlled experiments, Leontief engaged in the latter in an ambitious project of “expansion of the field of direct observation.” This laid the statistical foundations of what he called the “direct induction” approach to empirical modeling.