Am 31.03.2011 15:46, schrieb Anna Lee:
Dear List!
I have a unverse (basic population) which is not normally distributed.
Now from this universe I take some subsets. Each subset is normally
distributed within itself. I now want to compare the subsets and see
if they differ significantly. So what is my assumption - normal
distributed data and therefore comparison of means, or non normal
distributed data and therefore comparison of medians?
If you want to do an ANOVA, the assumption is normality /within/ the
groups. That is, Y given X=x is normal distributed. Also, you want the
same variance within each group (group = your subset = factor value).