Date: Sat, 12 Mar 2005 17:29:42 -0800
Reply-To: LUCINDA M TEAR <lucindatear@msn.com>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: LUCINDA M TEAR <lucindatear@msn.com>
Subject: Re: MANOVA with covariate
Content-Type: text/plain; charset="iso-8859-1"
I've mostly used covariates with the GLM procedure, but I imagine the
concepts are similar with the MANOVA routine. Unless you add an interaction
term between the covariate and a main effect, the model tests for a
constant, LINEAR relationship between the covariate and the dependent
variable across all the levels of any categorical factors. The slope of the
covariate can be positive or negative, and if the effect is not linear, then
you have not met the assumptions of the test and should be careful
interpreting your p values (to check, do a scatter plot of time til death
and self esteem and see if the relationship seems linear). It can be
helpful to check for interactions between the covariate and the main effects
(specified in the model in the same way that you would specify an
interaction of categorical main effects) to see if the slope of the
relationship between the covariate and the dependent variable is the same
for each level of the factor. For example, your results indicate that self
esteem increases (or decreases, depending on the slope of the covariate)
with time to death at the same rate at all levels of intervention. If you
add an interaction term to your model (between time to death and
intervention), you will find out if the slope of the relationship is
constant across all levels of intervention (the interaction will not be
significant) or if it differs depending on level (the interaction will be
significant). Adding an interaction makes the ANOVA an Analysis of
Covariance. With MANOVA, I imagine the significant covariate means that
there is significant linear relationship between time to death and self
esteem, pain intensity, and # of physical symptoms AND that the slope of the
relationship between time to death and each of those variables is the same.
Printing out the parameter estimates will show you what the slope
is...Plotting a scatter plot of time to death and each of your dependent
variables and coding the data by level of intervention would help you see if
the data for each level are intermixed with data for other levels or if the
data for each level cluster together and apart from other levels.
Hope that makes sense....
Lucinda
----- Original Message -----
From: "michael healy" <healym@earthlink.net>
Newsgroups: bit.listserv.spssx-l
To: <SPSSX-L@LISTSERV.UGA.EDU>
Sent: Friday, March 11, 2005 12:57 PM
Subject: Re: MANOVA with covariate
> that older people are more likely to die.
>
> -----Original Message-----
> From: Andrew Walsh <awalsh@uhnresearch.ca>
> Sent: Mar 11, 2005 11:41 AM
> To: SPSSX-L@LISTSERV.UGA.EDU
> Subject: MANOVA with covariate
>
> hello all,
> i am unsure of the proper interpretation of this analysis. i've run a
> MANOVA with a single continuos variable as a covariate (time to death).
> my factor has three levels (no intervention vs. late intervention vs.
> early intervention) and i have 3 dependent variable (self esteem, pain
> intensity, and # of physical symptoms). so my question is, what does a
> significant F indicate when it is associated with the covariate for the
> MANOVA?
> thanks very much,
> ~aw
>
| 
