Judy L. Klein
Probability theory is usually the starting point for pedagogical developments of econometrics. The ideal roots are traced to the mathematical treatment of games of chance, errors in measurement and combination of observations. The practices of merchants, bankers and captains of industry, however, had more influence on early economic statistics than did the logic of mathematical philosophers. The humbler origins of econometrics lie in the nineteenth-century rules of thumb on temporal comparisons of prices, sales and assets. Political economists adapted these tricks of the trade in their search for scientific means for investigating laws of motion.
Econometricians are not alone in glorifying the stimulus of probability theory at the expense of ignoring practical and cultural influences on the development of statistical method. As Laura Tilling (1973) and Theodore Porter (1986) have demonstrated in their respective histories of statistics, however, algorithms and interdisciplinary imagery applied to meet institutional exigencies usually preceded the theoretical applications of mathematics and probability to statistical studies in the natural and social sciences. It is likewise with econometrics. In fact it was not until Trygve Haavelmo’s work in 1944, as Mary Morgan (1987) has pointed out, that probability theory was introduced into econometrics. Long before then, the temporal and multivariate nature of economic theory and data forced nineteenth-century statisticians to look to business journals rather than Latin treatises for technique.
The legacy of statistical method from studies in games of chance, errors of measurement, social physics and eventually inheritance and natural selection was a concept of logical variation confronting cross-sectional data. Statistical population, frequency distribution and surfaces, and deviation from the mean involved a static comparison of differences. A goal of statistical investigations in social physics and natural selection had been to analyze change over time, temporal variation. The data used, however, were usually cross-sectional statistics, not time series. It was with the application of statistical method to meteorology and political economy that temporal variation was studied with temporal samples. This raised new problems and new concepts.