Conservative estimates of the variances of regression parameter estimators for classes of error model

SUMMARY Robust estimates of precision are obtained for generalized least-squares regression parameter estimators. They are robust in the sense of being conservative estimates of the variances of the parameter estimators when the error model is imprecisely specified. Serially-structured errors are one possible class. Substantial amounts of computing are required in each application. A numerical example is given in which errors are independent but do not necessarily have equal variances and results are compared with those obtained using a weighted jackknife estimator.