Outlier Detection in Balanced Replicated Linear Functional Relationship Model

Abstract

Identification of outlier in a dataset plays an important role because their existence will affect the parameter estimation. Based on the idea of COVRATIO statistic, we modified the procedure to accommodate for replicated linear functional relationship model (LFRM) in detecting the outlier. In this replicated model, we assumed the observations are equal and balanced in each group. The derivation of covariance matrices using Fisher Information Matrices is also given for balanced replicated LFRM. Subsequently, the cut-off points and the power of performance are obtained via a simulation study. Results from the simulation studies suggested that the proposed procedure works well in detecting outliers for balanced replicated LFRM and we illustrate this with a practical application to a real data set. The implication of the study suggests that with some modification to the procedures in COVRATIO, one could apply such a method to identify outliers when modelling balanced replicated LFRM which has not been explored before.