Statistical problems in the analysis of comparative pharmaco-EEG trials.

Abstract

Assuming already condensed data from the EEG power spectra, problems arising with the statistical analysis of such data in comparative pharmaco-EEG trials are discussed. Since, in general, there is only a limited number of subjects available in a study to evaluate and compare the effects of a number of drugs by means of the EEG, the experimental design is usually of the cross-over type. With respect to this design, the requirements for the application and for the appropriate analysis are stated. Further, the problems are indicated which arise when the number of subjects is not a multiple of the number of drugs and/or when there are missing values. Finally, the topic of repeated null hypothesis testing and the inflation of the overall significance level is treated. Suggestions are made how to at least partly avoid the indicated problems at the present stage of methodological statistical research.