Homogenity in mulitivariate data

jw3

New Member
#1
Dear all,

I have the following problem.

There is a multivariate data set, around hundred samples and almost
five hundred variables. The samples belong to two groups, and several
variables show significant differences between the two groups. The
samples are biological and belong to one of the two sets, A and B; the
variables correspond to a (large) number of physiological parameters.
Naturally, the physiological parameters are not expected to be
independent.

Looking at single variables I have the strong impression that one data
set is more variable than the other. For example, frequently a
variable in set A shows a roughly normal or log-normal distribution,
while in set B it shows two sub-groups of samples, a sub-group similar
to A and another group of outliers. Correspondingly, mean and median
of a variable frequently differ in set B.

The hypothesis that I would like to test is that set B is less
homogenious than set A. I am not interested in the homogenity of
particular variables, only in the overall comparisons of the set.
Where should I head, which family of statistical methods could be
reasonably applied in such a case?

Kind regards,

January
 

bugman

Super Moderator
#2
PERMANOVA + offers a great routine called permdisp, which tests thst null hypothesis that all groups have the same variance. It is analagous to levenes test, but for multivariate data. the assumptions are minimal since is test through permutation, and I think this might be what you need.

Here is the reference. I think this canbe done in R with the package vegan, but I use in in the PERMANOVA + platform.

Anderson, M.J. (2006) Distance-based tests for homogeneity of
multivariate dispersions. Biometrics 62, 245--253