Fieller's method for mixed models

摘要

In complete, balanced dose-response trials with independent observations, the dose-ratio between different treatment groups can be estimated and exact confidence limits can be found by the Fieller method. We have used the same simple approach on the general parallel line model and found Fieller-type confidence limits derived from approximate distributions. A simulation study show that in situations similar to the dose-linearity trial, the approximate distributions seem to fit reasonably for samples, as small as 5 subjects per group. Furthermore confidence regions based on the approximate distribution results in far more sound conclusions than regions obtained by the delta method, relying on asymptotic results, when the dose-ratio is truly a non-linear function of the parameter estimates.