Estimates of Parameters of a Censored Regression Sample

Winsorization and trimming using stratified and pooled procedures are proposed for linear regression analysis. The regression model assumes an equal number of observations at each value of an equally spaced independent variable. In the null case, the error variable is normally distributed. Efficiencies of estimates are obtained analytically for the stratified procedure and by Monte Carlo methods for the pooled procedure. Means and variances of estimates are obtained analytically for the untreated sample and by Monte Carlo methods for the Winsorized and trimmed samples for both location and scalar contamination. Winsorization and trimming show increased efficiency over the standard method.