Heteroscedasticity Generated By Errors In Predictors

The heteroscedasticity or changing variance observed in "raw" data may be the result of randomness or uncertainty in the predictor variables. As an example we consider "Charpy Test" experiments widely used to characterize the ductile-brittle toughness of steels such as those used for nuclear pressure vessels. While this type of experiment is of interest in itself, our main objective is to show that the use of a proper statistical technique may help to avoid the use of more complicated physical models to explain the heteroscedasticity of the observations. We also extend the existing method of regression analysis with errors in controllable variables to the case when the variances of the response and the controllable variables are both unknown. Key words: Predictors subject to errors, regression analysis, errors in controllable variables, iterated estimators, logistic response model. - vii - 1. Introduction This short study was essentially motivated by results reported by Downing, Ha..