Univariate or Multivariate?

My data is on two equal groups of patients. The two groups have roughly equal proportions of males/females. A blood sample was obtained form each patient, and 10 of the blood components were measured either as a count, % of volume or a concentration.

The objective is to see whether the blood components differ between the two groups of patients, between sexes and the interaction of group and sex.

All components are not normally distributed; one group have more right skew for all the components. Each component has different shape in the two groups of patients.
I don’t want to go for transformation. So, the only choice left is non-parametric analysis.
My questions are:
1- Is it OK to analyze each blood component separately (10 uni-variate analyses)?
2- If so, do I have to adjust for multiple testing?
3- Else, if I have to use non-parametric mufti-variate ANOVA ,e.g. the Two-way NPMANOVA module, available in PAST package, after randomly dropping some subjects (about 10% of the data!) to make sex sub groups equal, what distance measure is appropriate for this kind of data (blood tests)? And how to make individual component comparisons in this case?

I would very much appreciate your suggestions
Last edited:


Point Mass at Zero
You seem to have quite a few variables in your dataset. So each of the 10 components are an endpoints of interest? Then I would assess the effect of all the available predictors on your response variables separately, this means 10 different models.
The distributions might be quite skewed. Then a generalized linear model (glm) with a gamma distributed error structure might be useful

It is okey to analyse them separately.

Edit: I first entered something in the the wrong thread. But almost the same comment can be relevant here too.
Last edited: