Abstract
A recent paper (R. C. Elston and J. E. Grizzle (1962) Biometrics 18, 148-59) presented a simple method, based on the mixed model of the analysis of variance, for obtaining confidence bands for estimated time-response curves. The bands obtained depend in part upon the zero point that is assumed for the time scale, which suggests that the model may be unrealistic. This note gives an approximate method of checking whether the model is appropriate, and then a generalization of the model that may often represent the true situation better. It is shown that under such a model the estimate of the mean response curve is the same as before, which is the same as that obtained when all observations are assumed to be completely independent. This more general model does not, however, lead to a simple method of obtaining confidence bands.

This publication has 1 reference indexed in Scilit: