A Sampling Study of Minimum Absolute Deviations Estimators

Abstract

Small sample properties of minimum absolute deviations estimators were studied via a simulation model. A four-parameter model of full rank was proposed and 50 random samples were drawn consistent with this model. Results indicated that, for this experiment, m a d estimators were about 20 per cent efficient compared to minimum variance estimators, using Aitken's concept of generalized efficiency. No significant biases were found in the experiment.