one way anova or nonparametric?

in the attached file you can see the results from a viability test after normalization. Each column represents a different kind of treatment and in the last one are the controls (without any treatment). Each group has 12 replicates. Im trying to find the significant differences of each group compare to control group, using graphapad prism program. Firstly im testing if the values come from a Gaussian distribution, choosing D΄Agostino and Pearson omnibus normality test (which is the recommented by the program). Then i can see that not all the groups passed the normality test (alpha=0.05) so i use a nonparametric test (Kruskal Wallis test) in order to further analyse the data. Is it correct; Does it matter if some groups assume Gaussian distribution and im using a nonparametric method; View attachment 5111
Thank you for your help


TS Contributor
Hi sofiail,

The Kruskal Wallis test is useful whether or not your data follows the assumptions of the linear model, so it won't matter if you have normally distributed responses for some groups. Either way, if the assumptions of ANOVA are met, this procedure will yield better results. Since normality of the response is not an assumption (the actual one is normality of residuals) I'd run an ANOVA first and verify it.

Btw, both ANOVA and Kruskal Wallis only test for overall differences (i.e. one treatment is indeed different to the rest/control). Yuu'll need further tests to compare them with your control group, such has Dunnett's test of Mann-Whitney's test after Kruskal-Wallis
Hi terzi,

thank you very much for your quick response! It was really very helpfull.
For further analysis i use Dunn's Multiple Comparison Test and made the comparison between control goup and the other groups.
Best Regards,