One-sided tests in generalized linear dose–response models

This paper discusses the asymptotic null distribution of three asymptotically equivalent one-sided tests in generalized linear models with separate lines. Simple forms for the correlation coefficients, which appear in the weights of the asymptotic null distribution, that is a mixture of chi-squared distributions, are given. A particular structure is proposed for the experiments in order to simplify the correlations and consequently to allow the use of several approximations developed for the weights. It is showed that the hypothesis of synergism or antagonism between two compounds may be assessed, under parallelism, by one-sided tests. Finally, as illustration, the efficiency of a standard preparation is compared with the efficiency of four test preparations under a gamma log-linear model.