I read some journal that stated when the parametric data is not homogenous (tested with Kolmogorov-Smirnov or some other tests) we must use non-parametric test (Mann-Whitney or Kruskal-Wallis). What the reason? if i insist use parametric test (anova or t-test) , what's happens?
Hello Palagan,
Rank-based, non-parametric alternatives to the usual parametric tests comparing independent groups also require homogeneity of variance (e.g., comparing Mann-Whitney-Wilcoxon test to independent t; comparing Kruskal-Wallis one-way anova by ranks to one-way anova). The Kolmogorov-Smirnov goodness of fit test flags data sets for (mis)fit to a target distribution, so this is more likely to be identifying problems with normality than anything else.
What options are available when your data do not meet normality or homogeneity assumptions (or both)?
1. Exact/permutation tests.
2. Resampling/bootstrap tests.
Here are some starting links you may find helpful:
https://www.sciencedirect.com/topics/mathematics/permutation-test
http://compdiag.molgen.mpg.de/ngfn/docs/2004/jun/resampling.pdf
Good luck with your work.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA