All,
When performing e.g. ANOVA to compare k treatments, obviously one of the
assumptions is Normality. If I am comparing treatments with a low number of
reps e.g. 5, it would be unsurprising that some of the treatments display
non-normality. Whilst I aunderstand that non-normality is not the most
crucial criterion for ANOVA, would extreme non-normality (e.g. all values
the same, or one extreme outlier) make that treatment totally inappropriate
for analysis by parametric methods.
If this is not the case are there any 'rules of thumb' which would state
that if n of the k (e.g. 4 from 6) treatments display non normality then
parametric methods are unsuitable, or is it reasonable to test the whole
data block for normality and continue with ANOVA if normality is indicated.
Similarly are there any 'rules' which say that if n of the k treatments have
heterogeneous variances then ANOVA is definitely not suitable.
Any help gratefully received
Regards
Alan Sharpe
.
|