Skip to Main Content
In designed experiments, we often encountered non-normal response variables. The data transformations (Transf) approached are frequently employed to deal with these problems. One has to realize that analyzing such data based on transformations posed many drawbacks. A better approach in dealing with these problems is by using the Generalized Linear Model (GLM). The problem becomes more complicated when there existed outlier in the data set. As an alternative, we may turn to robust (M- based) Generalized Linear Model (GLM) technique, which is less affected by outlier. In this paper we investigate the performance of the M-based GLM by doing the Monte Carlo simulation and its performance is compared to the Transf. and the GLM techniques. The empirical evidence shows that the M-based GLM is slightly better than the GLM and the Transf. approach in a well-behaved data. However, when contamination occurs in the data, its performance is remarkably robust with respect to outlier and non-normal responses.