Nonparametric step-down test procedures for finding minimum effective dose.

Nonparametric versions of normal theory step-down multiple-test procedures for inferring minimum effective dose (see Tamhane et al. (1)) were developed and studied by Monte Carlo simulation. Two types of step-down testing procedures were examined. For both procedures, pairwise, linear, or Helmert contrasts of mean ranks were studied. All nonparametric step-down procedures controlled familywise error rate under normal, double exponential, and exponential distributions. Specific recommendations based on power and bias comparisons of nonparametric methods among themselves, with Shirley's (2) test, and with parametric step-down multiple-test procedures are given.