Master's thesis at University of Basrah discusses the multivariate robust linear regression: simulation and application
A master's thesis at University of Basrah entitled "Multivariate robust linear regression: simulation and application".
The student Thamer Warda's thesis aimed to present a robust version of the Wilks statistic for testing hypotheses in multivariate linear regression, to address the sensitivity of the classical Wilks statistic to extreme values.
The proposed method included the MM estimator and the reweighted minimum variance (MCD) estimator, with the use of Hempel and Hooper weight functions to find the weights of observations for the sample. Monte Carlo simulation was used to compare the performance of the classical and robust Wilks statistic.
The results showed that the robust statistic outperforms the classical Wilks statistic in the case of contaminated data, while the robust methods and the classical method provide similar performance in the case of normal distribution.
Department of Media and Government Communication