Modified Studentized Residual in Nonlinear Regression

We discuss methods for detection of outlier in nonlinear regression. This issue is very essential in regression analysis because outlier can affect the covariate pattern and mislead regression interpretation. The existing literature on the identification of outlier in nonlinear regression is not as extensive as for linear regression. We propose a modification of studentized residual for testing outlier that overcomes many of the potential shortcomings of the ordinary studentized residual. The main idea of this article is to apply the linear approximation of nonlinear model function and consider the gradient as the design matrix. Also, we review leverage measures and explore the relationships between these leverage measures in nonlinear regression model. Example is given to illustrate the proposed methodology.