Measures of Nonlinearity for Segmented Regression Models.

摘要

A simple physical model of residential energy consumption provides the framework for an exploration of segmented regression models fit by least squares. The energy model is a generalization of a linear, single change-point model such as that considered by Hinkley. Some simple geometric measures of nonlinearity and nondifferentiability are proposed. These measures are related to the construction of approximate confidence regions for the parameters of a general segmented model. In addition, the relation shown between these measures and those proposed by Bates and Watts may be useful in analyzing continuously differentiable models. (Author)