GLM Multivariate on Numerous Dependent Variables ~ Isolate Different Types of Tests?

#1
So here is a question that is bothering me, and I cannot seem to find the answer on google or site searches

Let's say I have a database with two groups of rats, treatment and vehicle.

I have numerous dependent variables associated with each rat, for a long lengthwise database.

Some of the dependent variables are for their behavior, and some are for their anatomy, and some are for MRI measures such as diffusion in different brain areas.

So far I have just run a GLM Multivariate in SPSS to correct for the whole bunch, and the results are tolerable. My question is should I be running the GLM on the whole database or doing separate GLMs on three separate clusters of all behavioral measures, all anatomical measures, all diffusion measures, etc.?
 

bugman

Super Moderator
#2
Re: GLM Multivariate on Numerous Dependent Variables ~ Isolate Different Types of Tes

Look into MANOVA.

Cant really offer any more help unless you supply more ionformation about your data set.

Plot your data. Look at the relationships between variables. Are there any that are redundant to your study. Can some be used as a surrogate for another (i.e. weight vs. height?).
 
#3
Re: GLM Multivariate on Numerous Dependent Variables ~ Isolate Different Types of Tes

I've looked at MANOVA. I've plotted my data, thanks.

I'm not sure what extra information you need.

There are two groups: treated and vehicle.
There are a large number of tests on each animal.

So


Group v1 v2 v3
T
T
T
V
V
V


...etc.

So it's just 1 Factor and a ton of dependent variables.

Let's skip the original question since that is not getting anywhere. I was attempting to do some sort of multiple correction in SPSS (GLM Multivariate) but with this model, it does not allow you to. It gives you a corrected model, but I gather this is just for correlation between outcome variables. The only corrections available (Tukey, Dunnett, etc.) seem to be for post-hoc comparisons which are pairwise, so If you select one it tells you there are fewer than 3 groups so it won't run.

Does anyone know how to do some sort of correction for multiple dependent variables other than bonferroni that I can do on data set up this way?
 
#4
Re: GLM Multivariate on Numerous Dependent Variables ~ Isolate Different Types of Tes

btw chances are good I am just going to go back to excel and manually do a Benjamini-Hoch FDR correction but I really wish there were an easier way.