| Researchers in Education and Psychology often encounter skewed, sparse, or small data sets, called nonnormal, which are ordinal categorical in nature. A proper approach to analysis of these data kindled a debate which has concluded, for the most part, that nonparametric statistical tests outperform their parametric counterparts for many situations. Yet, researchers continue to use normal theory statistics to solve problems with nonnormal data. This paper discusses the virtues of nonparametric statistics for analysis of nonnormal data and applies four nonparametric tests appropriate for ordinal categorical data sets: the Wilcoxon Mann-Whitney U test, the Normal Scores test, the Savage Scores test, and the Permutation test. The properties of these tests are compared for various situations of sample size, alpha levels, and number of ordinal categories in outcome data. Twenty-nine journals from the fields of Education and Psychology were reviewed over a four-year period to collect real data sets.;In addition to identifying the most appropriate test statistic for each situation, this paper brings the researcher well into the 1990s through its advocacy of modern, user-friendly computer software programs which educators and psychologists will not find intimidating and which will provide quick, simple, and accurate analysis of their data. Specifically, the paper promotes the use of StatXact Turbo 2.0 (Gajjar, et al., 1992) for use with ordinal categorical data. This program produces exact permutations of p-values for each data set, providing the researcher with reliable results, an issue of paramount importance when prescribing treatment for the welfare of human subjects. |
