On sensitivity of inverse response plot estimation and the benefits of a robust estimation approach
by Luke Prendergast
Abstract: Inverse response plots are a useful tool in determining a suitable transformation function that may be used to linearize the response given a linear combination of the predictors. Under some mild conditions it is possible to seek such transformations by plotting the ordinary least squares fits versus the response values. A common approach is to then use nonlinear least squares to estimate a transformation function which depends on an unknown parameter that needs to be estimated. In this talk we provide insight into this approach by considering sensitivity of the estimation via the influence function. For example, sometimes the results from such an analysis can be confusing in the sense that poor inverse response plots can still sometimes result in reasonable estimates of the transformation parameter and the influence function provides insight into why this can be the case. We will also show how this influence function leads to a useful influence diagnostic and also introduce a simple robustified process that can vastly improve estimation of both the inverse response plot and the transformation function parameter. This talk is based on joint work with Professor Simon Sheather, Head of the Department of Statistics at Texas A&M University.
For More Information: contact: Johanna Ziegel. email: firstname.lastname@example.org