A large-sample Kolmogorov-Smirnov test for normality of experimental error in a randomized block design

Abstract

SUMMARY The weak convergence of the sample distribution function based on the estimated residuals in a randomized block design is considered under the null hypothesis of normality of the experimental errors. This result serves as a basis for deriving the limiting distribution of the Kolmogorov-Smirnov statistic computed from the estimated residuals. Quantiles of Monte Carlo simulations of the limiting distributions for various numbers of treatments are given. These results are applicable also to the two-way mixed model (Scheffé, 1959).