completely random design, violate normality

#1
--------- DESIGN OF EXPERIMENTS -----
Hello
I have a difficulty in the following case:
  1. normality (0.0001)
  2. constant variance
  3. independence of errors
when I apply the Kolmogorov test corrected by Lilliefors (0.0001), then I reject H0 and this violates 1 test assumption.
I have now been told that I must apply the Kruskal-Wallis(non-parametric) test.
If I apply this test, does she replace Kolmogorov?
and the rest stays the same? or is there any other step I'm skipping?
because then there's ANOVA.
 

Attachments

Last edited:

Karabiner

TS Contributor
#4
normality (0.0001)
Please keep in mind that for oneway analysis of variance
not the unconditional dependent variable should be
normally distributed, but the residuals from the model.
I have now been told that I must apply the Kruskal-Wallis(non-parametric) test.
Who told you so, and why?
If I apply this test, does she replace Kolmogorov?
It does replace the usual oneway analysis of variance, if sample size
is small and/or crucial assumtions of the ANOVA are violated.

I does not use the original values, but the values transformed
into ranks (here, from 1 to 20).

With kind regards

Karabiner
 

Karabiner

TS Contributor
#6
I was just curious. Since normality of errors was analysed (not normality of the dependent variable), and sample size is small, the Kruskal-Wallis seems like a defensible alternative.

With kind regards

Karabiner