EVALUATIONS USING NONPARAMETRIC STATISTICS

In this paper the resilience of nonparametric statistics to an extension of the sample set is investigated. It is shown that the addition of fourth and fifth samples does influence one's decision about a difference in scores between the original samples. This could result in strange recommendations like: The experiment pointed out that no significant differences were found between the experimental and control group. However, we recommend you to repeat the experiment, because the differences over two experimental and two control groups with exactly identical results will probably be significant. The cause of the problem mentioned is sought in the ranking of scores and the statistics used. The paper concludes with a number of recommendations.