A Set of Nonparametric Tests for Experiments with Lattice-Ordered Means: Theory, Programs, and Applications

Abstract

In many factorial experiments where the factors have levels that are ordinal or quantitative, a researcher may predict that the mean response in certain treatments will be higher or lower than those in other treatments. One type of order that may be anticipated is called lattice order, where average response tends to increase (or decrease) as the levels of any one of the factors is increased, holding the others fixed. A Kendall-type statistic, which measures the degree of lattice order in the data, can also be used to carry out a test involving lattice-ordered means. In this article, tests for individual factors are developed to complement the overall test of lattice order, and the methods are then applied to relevant and current data. Programs in R and FORTRAN are included to carry out the tests.