MANOVA of variances?

#1
I'm trying to figure out how to determine the effect of three (2 binary, 1 categorical) independent variables on the variability of a number of dependent variables.

My hypothesis is that subjects under certain treatments will be more variable for some or all of the dependent variables.

I know that if I wanted to test which treatments affect the mean of a number of dependent variables, I could use MANOVA, so I was wondering if I could substitute variances of each trait instead of means and test my hypothesis that way.

Thanks for any help!
 

JohnM

TS Contributor
#2
Theoretically, yes, you can use variance as the response rather than the means. If you use means to assess significance, you'll be evaluating the differences between the means - with variances, it's a bit different - you'll be evaluating the ratios of the variances, not the differences.

But - be careful. MANOVA should be used only when it makes sense, from your underlying theory, to combine the multiple dependent variables into one composite dependent variable (a linear combination, in other words). Just because you have >1 dependent variable does not necessarily mean you should be using MANOVA.

Otherwise, just use separate ANOVAs for each dependent variable.