gms | German Medical Science

Background: Although p-values are still the dominant representation in 2- and k-samples significance tests, confidence intervals should be used instead, especially as compatibility intervals. Firstly, in regulatory toxicology, this recommendation is specific, simply because of the proof of hazard used in routine, although the proof of safety would be more appropriate. Secondly, although endpoint-specific noninferiority limits are hardly available, post-hoc interval inclusion in a (1-2alpha) interval can be used to avoid the 0.05 quasi standard. Thirdly and as the main focus of the presentation, the problem of compatibility is broken down into individual concepts.

Methods: The consequence of assuming homogeneous or heterogeneous variances in the usually unbalanced k-sample design is illustrated. Surprisingly, in the triad of bias, flase positive, false negative rates, there seems to be only the choice of less bad methods today.

The choice of parametric or non-parametric tests is discussed as a further aspect: once as maxT-test, on the other hand from the aspect most likely transformation models [1].

But also such details in the analysis of mortality adjusted tumor rates using poly-k trend tests. If the dose is to be modelled qualitatively (as a factor) or quantitatively (as a covariate), one should assume a-priori linear shapes or consider arbitrary (monotone) ones and should choose the tuning paprameter k to 3 or 6. By means of multiple marginal models and max(maxT)-test related solutions are proposed.

Results: These compatibility issues are illustrated by three real data examples with the CRAN packages multcomp, MCPAN, tukeytrend and tram.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.