Effect of Using Parametric Multiple Comparison Tests After Non-Parametric Tests on Type I Error Rate and Power of Test

GENC S., SOYSAL M. İ., YİĞİT S., MENDEŞ M.

Scholars Bulletin, vol.4, no.3, pp.247-257, 2018 (Scopus)

  • Publication Type: Article / Article
  • Volume: 4 Issue: 3
  • Publication Date: 2018
  • Journal Name: Scholars Bulletin
  • Journal Indexes: Scopus
  • Page Numbers: pp.247-257
  • Çanakkale Onsekiz Mart University Affiliated: Yes

Abstract

This study was conducted to investigate the effect of using parametric multiple comparison tests (e.g. Duncan test) after Kruskal-Wallis test on type I error rate and test power. Results of simulation study showed that Kruskal-Wallis tests are nonparametric alternative of one-way analysis of variance especially when normality assumption is not satisfied. Research showed that parametric multiple comparison tests can be used then the non-parametric tests. For this propose; Duncan Method (DUNNR) is used instead of Nonparametric Dunn's method (DUNN) after Kruskal-Wallis (H) test. Random numbers from N (0,1), Ki(3), Beta(5,3) distributions were generated; with the various sample sizes (n=3, 5, 7, 10, 15, 30), group numbers (k=3, 4, 5) and the different variance ratios ( Var =1:1:...:1, 1:1:...:4, 1:1:...:10) in this study, The aim of study was determination of the possibility of use parametric pair wise multiple comparison; by determination of type I error and power of tests instead of no-parametric pair wise multiple comparison tests can be used.