Using absolute deviations to compute lines of best fit

Abstract

There is some renewed interest among agricultural economists and others in the old technique of using minimized absolute deviations in computing lines of best fit. H. B. Jones and J. C. Thompson, in a recent article in Agricultural Economics Research, discussed philosophical and practical questions raised when fitted lines using squared and unsquared deviations are compared. However, economic literature (including the Jones-Thompson paper) as well as elementary statistics and econometric literature seems to contain no concise description of the straight-forward and simple procedure for calculating straight lines of best fit which minimize the sum of absolute deviations from the empirical observations to the fitted line. In some work at the University of Minnesota, we find that lines fitted by minimizing the sum of absolute deviations can be useful in certain circumstances.