Search for: [Abstrakt = "Business data are often highly skewed to the right for two reasons\: occurrence of outliers and a large proportion of zeroes. If by chance several unusually large residuals should fall in the sample then applying estimator may grossly underestimate or overestimate the population totals. One technique to deal with this problem is to divide a sample into two parts basing on cutoff values. Observations outside preset cutoff values are modified to values closer to these coutoff values. This estimator is called the winsorized estimator. The affectivity of winsorized estimator depends on the choice of the cutoff values, and hence the methods used to estimate regression parameters used to calculate these cutoff values. In this paper we examine the problem of the choice of one of the robust regression techniques to determine which techniques resulted in the best performing winsorized estimator. Simulation study presented here shows that Sample Splitting Technique results in the largest percentage reduction in MSE."]
