Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: RE: RE: non-parametric MANOVA


From   "Lachenbruch, Peter" <Peter.Lachenbruch@oregonstate.edu>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: RE: RE: non-parametric MANOVA
Date   Tue, 28 Oct 2008 08:23:31 -0700

I'm in semi-agreement with Steve Self's approach:  use a bootstrap on MANOVA.  However, I seem to recall a paper in the Annals of Statistics in the late 1970s by a statistician named Maronna who showed that the breakdown point (where contamination wiped you out) was at 1/k, the number of variables.  Thus, if you have more than 7% contamination (I assume that means non-normal) you have a problem.  But maybe I'm over-interpreting.  My initial reaction was to propose a bootstrap approach as Steve suggested.

Tony

Peter A. Lachenbruch
Department of Public Health
Oregon State University
Corvallis, OR 97330
Phone: 541-737-3832
FAX: 541-737-4001


-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Nick Cox
Sent: Tuesday, October 28, 2008 7:02 AM
To: statalist@hsphsun2.harvard.edu
Subject: st: RE: non-parametric MANOVA

I find some inconsistency in this request. After all, what would a non-parametric MANOVA look like except something like a MANOVA, except that your data have been transformed to ranks? If you are happy to reduce your data to ranks, why cavil at some other transformation, which typically would lose less information? 

Further, my visceral instinct is that MANOVA is more robust to non-normality than people fear. 

More positively, if this were my problem, I might do 

1. MANOVA on original data 
2. MANOVA on rank-transformed data 

If the conclusions were substantively similar, stop there. Otherwise, consider what specific transformations were advisable.  

Nick 
n.j.cox@durham.ac.uk 

Jochen Späth

I want to do a Manova (14 different dependent variables, 2 main factors) and am stuck with the problem that most of my fourteen variables are not normally distributed (and I do not want to transform them in order to get them normal since I have only remote access to the data which complicates things a lot). So, my question is: is there a way to do such a MANOVA in STATA using non-parametric techniques (the -kwallis- command allows only for one factor and one dependent variable as far as I know)?

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index