Prediction and Categorical Control in Regression

A primary application of regression analysis is prediction. In this paper, we consider the definition of the domain of the model in which prediction is valid. This is important because prediction made outside the domain may be unacceptably different from the true responses. We provide a criterion that can be used to decide whether prediction is valid at a certain point. The criterion is based on the existence of an unbiased estimate of the distribution function associated to the "future" observation. In addition, in the context of regression analysis, the categorical control problem that is quite different from the numerical control problem in the setting of the target is considered. Categorical control may be compared to interval prediction, whereas numerical control is compared to point prediction. Our derivation is based on the Scheffé-type simultaneous tolerance interval at two distinct points.