Repeated-measures bioassay with correlated errors and heterogeneous variances: a Monte Carlo study

The relative potency of two drugs may be estimated from a small experiment in which all K doses of each drug are given in order of increasing dose level to each of n aminals at one testing session. Fiducial limits are often estimated from Fieller's theorem, with the (2K - 1)(n - 1)-degree-of-freedom error term obtained from a two-way analysis of variance. Monte Carlo results indicate that this method can yield limits which are far too narrow when there is serial correlation between successive doses. The jackknife confidence limits are found to behave well for the models investigated.