topic Re: Without normality or homogeneity of variances in Statistical Procedures

Without normality or homogeneity of variances

LPC22 — Tue, 09 Jul 2019 21:32:56 GMT

I have four treatments. The variables to study, which are weights and lengths, do not follow a normal distribution and the groups do not have the same variance so the question is: which kind of test should I use?

Re: Without normality or homogeneity of variances

ballardw — Tue, 09 Jul 2019 22:02:57 GMT

@LPC22 wrote:

I have four treatments. The variables to study, which are weights and lengths, do not follow a normal distribution and the groups do not have the same variance so the question is: which kind of test should I use?

It will help if you can tell us what your research question is. Are you testing for identical means or median? A known/suspected difference in a mean or median? Distribution of values?

Re: Without normality or homogeneity of variances

PGStats — Tue, 09 Jul 2019 22:08:52 GMT

Non parametric tests do not rely on the normality of data for validity. Look at the documentation for proc npar1way.

Re: Without normality or homogeneity of variances

LPC22 — Wed, 10 Jul 2019 16:47:37 GMT

I thought to use Kruskal-Wallis test but data for some variables are heteroscedastic.

Re: Without normality or homogeneity of variances

PGStats — Wed, 10 Jul 2019 21:32:17 GMT

Heteroscedasticity might not be a real problem here. Variables such as weights and lengths are typically not very skewed. ANOVA is quite robust to mild non-normality and might be perfectly fine. Heteroscedasticity will cost you some power in K-W analysis, but inferences will remain valid. Try both ANOVA and K-W. If you get similar p-values, you're fine. Otherwise, you should look more closely at your data for outliers, subgroups, etc.