MANCOVA with all predictors being continuous variables - how to follow-up?

#1
Dear Stats Experts,

I would be most grateful for any thoughts on the following:

I have a cross-sectional dataset with 7 outcome variables from 100 subjects. The outcome variables are of equal theoretical priority and are moderately correlated. I chose MANCOVA because I am interested in the effect of the predictors not only on the individual outcome variables but also on "a linear combination" of the outcome variables.

I have two predictor variables which are both continuous, so in SPSS MANOVA I enter these as covariates. Both give a significant main effect.

As I understand, there are two ways to follow up a significant MANOVA/MANCOVA:
1. Simply look at the tests of between-subject effects that follow the multivariate tests.
2. Normally, discriminant analysis, as I understand from Tabachnick and Fidell, is a different (better) way to follow up a significant MANOVA because it is more "in the spirit" of MANOVA - taking advantage of linear combinations of dependants. However, I have no grouping variable, so discriminant analysis is not possible.

Question: How can I follow up on a significant multivariate test to demonstrate the relative contribution of each dependent variable to the linear composite that give the significant multivariate effect if my predictors are continuous? Is MANCOVA the wrong approach without "fixed factors" or grouping variables?

Many thanks for you thoughts and/or advice. I hope my questions make sense.

Best wishes,
Anton
 

spunky

Doesn't actually exist
#2
I have two predictor variables which are both continuous, so in SPSS MANOVA I enter these as covariates. Both give a significant main effect.
you kind of lost me here... if your predictor variables are continuous then you're not doing MANOVA, you're running multivariate multiple regression... or what are you adding as factors in your MANOVA?
 
#3
Thanks for your reply.

... And my apologies because I did neglect to mention that there is indeed a dichotomous variable - a genetic factor (present/absent) - but there is no significant main effect. There is however a main effect with one of the continuous predictors. I would like to examine in more detail the relationship between this continuous predictor and the outcomes. However, I would like to do so not just by looking at the follow up between subject effects; I would like to see, if possible, "the linear combination" of dependent variables that is associated with the significant predictor.

I hope this makes sense! Is there a way to examine the above question? Here is SPSS log of what I did:

GLM outcome1 outcome2 outcome3 outcome4 outcome5 outcome6 outcome7 BY gene WITH covariate1 covariate2
/METHOD=SSTYPE(4)
/INTERCEPT=INCLUDE
/CRITERIA=ALPHA(.05)
/DESIGN=covariate1 covariate2 gene.


Many thanks for your thoughts.
 

spunky

Doesn't actually exist
#4
no prob, i just sort of assumed there must be a categorical predictor somewhere....

anyways, well, then i guess i have to re-iterate my position there. if you are dropping your categorical covariate and just adding the continuous one, then you're doing multivariate regression...

problem is it's been a long, looooong time since i used SPSS to do any serious data anlysis (i'm an R user, like the majority of the people here), but i do remember the GLM prop window was very flexible. what i would try first is to go to analyze->General Linear Model-> Multivariate and in the "Covariates" window, add your continuous variable and run it, without anything on the "Factors" box. i'm pretty sure that should work...i think... i dunno, as i said, i havent used it in a long, long time...