On Bootstrapping M-Estimated Residual Processes in Multiple Linear-Regression Models

It is shown, under fairly general conditions, that Efron's bootstrap procedure captures the limit distribution of weighted empirical processes based on M-estimated residuals in multiple linear regression models. As an application, we construct bootstrap confidence bands for the error distribution function F. The main result can also be used to design distribution-free goodness-of-fit tests for F without any recourse to the split-sample estimation of the regression parameters.