Selection of a statistical procedure for experiments involving multiple dependent measures
Bechtel, Laura R.
Lane, David M.
Master of Arts
After the experimenter defines the statistical hypothesis and controls the type I error rate, the choice of a statistical procedure can be based on power. This study used Monte Carlo methods to investigate the relative power of univariate and multivariate tests of the levels, parallelism, and omnibus hypotheses. Four thousand runs of the Monte Carlo program were conducted for an experiment with given dimensions, population effects, and within groups variance-covariance matrix. As the experimental conditions were varied systematically, the effects on power and type I error rate were studied. Based on the results of 153 different Monte Carlo experiments, a set of guidelines was formulated for experimenters who employ multiple dependent measures. The guidelines were expressed in flowchart form as well as in text. The prudent experimenter who follows these guidelines will use multivariate profile analysis to test the parallelism hypothesis and combine multivariate profile analysis with between-groups analysis of variance to test the omnibus hypothesis.