Procedure for binary DV equivalent to MANOVA

Tommy Xie

unread,

Jun 2, 2009, 11:05:14 PM6/2/09

to

Hi folks,

I found this group very helpful and would like to ask a question here.

Is there a statistical procedure that is similar to MANOVA but for
binary DVs? I am aware that I can run logistic regression on each DV,
but is there a way to examine them in one procedure?

I am using SPSS 16, by the way.

Thanks!

Ryan

unread,

Jun 3, 2009, 7:59:55 AM6/3/09

to

Depending on your research question, yes. Look into generalized linear
modeling (no random effects) or generalized linear mixed modeling
(random effects) . If you tell us a bit more about your model, we
might be able to help you code it up.

Ryan

Tommy Xie

unread,

Jun 3, 2009, 10:37:25 AM6/3/09

to

Thanks for replying! There are four DVs (W,X,Y,Z) coded in binary (0,
1), and two IVs (A,B), each with two levels. My research question is
"How do W, X, Y, Z vary by A and B?" I hope the information is
detailed enough. Have a good day!

Ryan

unread,

Jun 3, 2009, 11:22:11 PM6/3/09

to

> detailed enough. Have a good day!- Hide quoted text -
>
> - Show quoted text -

It isn't clear to me based on the limited info you provided that GEE
is the way to go, but assuming it is, take a look at a recent thread
on how to create code in the GENLIN procedure-->"Repeated Measures
with Within Subjects factor Logistic Regression in SPSS?" You should
be able to figure out how your data need to be structured (long
format) and how to write the code by reviewing this thread. Also,
consider taking a look at the help tutorial in SPSS.

Bruce Weaver

unread,

Jun 4, 2009, 8:38:54 AM6/4/09

to

Ryan, I've probably not used GENLIN much as you, but I think it is
limited to models with one dependent variable, is it not?

--
Bruce Weaver
bwe...@lakeheadu.ca
http://sites.google.com/a/lakeheadu.ca/bweaver/
"When all else fails, RTFM."

Ryan

unread,

Jun 4, 2009, 9:28:59 AM6/4/09

to

> bwea...@lakeheadu.cahttp://sites.google.com/a/lakeheadu.ca/bweaver/
> "When all else fails, RTFM."- Hide quoted text -

>
> - Show quoted text -

Hi, Bruce:

It is possible to parameterize a model in such a way that GENLIN or
even the MIXED procedure can handle multiple dependent variables.

First you need to add an indicator variable as a link to the dependent
variables. Then you’d code up the model to allow for separate
intercepts for each level of the indicator variable by excluding the
grand fixed intercept, and *only* including interactions when adding
the covariates of interest (i.e. Y = indicator indicator*X1
indicator*X2…Indicator*Xk).

Here are a couple of articles that explain the concept in mixed
modeling:

http://ssc.utexas.edu/docs/sashelp/sugi/23/Stats/p229.pdf

http://arxiv.org/ftp/arxiv/papers/0705/0705.0568.pdf

***Here's a more eloquent explanation by Dale in the SAS google group
(It's a must read!):

http://groups.google.com/group/comp.soft-sys.sas/browse_thread/thread/d2ee17fcec2a6d02/8e630553e60bbf6c?hl=en&lnk=gst&q=mixed+multivariate+#8e630553e60bbf6c

Best,

Ryan

Bruce Weaver

unread,

Jun 4, 2009, 6:24:06 PM6/4/09

to bwe...@lakeheadu.ca

> http://groups.google.com/group/comp.soft-sys.sas/browse_thread/thread...
>
> Best,
>
> Ryan

Thanks. I don't have time to look at these right now, but will save
the links for later.

--
Bruce Weaver
bwe...@lakeheadu.ca