Estimation of Multivariate Linear Regression Parameters Using Robust Methods (Comparison Study)

Abstract

In this paper In this Thesis has been use several robust methods – less sensitivity with outliers values – to estimate parameters matrix. These were compared by using criterion mean square random error by relying on simulation to get the data. Data mimic the reality and on which the comparison of methods is built. It was generated two kinds of variables: non-contaminated natural variables, contaminated natural variables from two sides. It was reached through the comparison between the results of the three methods. That data in the case of non-polluting match the results of the three robust methods. As in the case of polluting data, the suggested method - which has been assumed a constant cutoff value suggested for it - often give best results of two way methods: Estimate of minimum covariance determinate (MCD) & (MCD) re-Weight.