A Note on the Minimization of the Trimmed Sum of Absolute Deviations as an Integer Program

Keywords: resistant estimation techniques, outlier detection, linear programming

Abstract: The trimmed sum of absolute deviations criterion (Tableman, 1994) is useful for diminishing the effects of large random error terms on estimated regression coefficients, as well as aiding in the detection of outliers. Our interest lies in the ability to write the minimization problem as an integer program. Thus, the requirements to perform a least trimmed absolute deviations regression analysis are met for any practitioner with access to a computer package capable of solving integer programming problems.