Date: Thu, 8 Apr 1999 20:24:14 -0400
Reply-To: "Baker, Harley E,, DMDCWEST" <BAKERHE@OSD.PENTAGON.MIL>
Sender: "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From: "Baker, Harley E,, DMDCWEST" <BAKERHE@OSD.PENTAGON.MIL>
Subject: Re: covariance analysis
The answer is that you cannot use ANCOVA/MANCOVA (based on your description
of the intended analysis.) This is because ANCOVA/MANCOVA require (as a
matter of the mathematics of the analysis) that the same covariates be used
across all levels of the independent variables.
To understand why, ANCOVA/MANCOVA perform their work by (1) adjusting the
between group sums of squares, and (2) set the groups "equal" on the
covariate. By these two adjustments to the sums of squares, ANCOVA/MANCOVA
equate the groups (statistically) on the covariates and in so doing "remove"
their influence on the outcome (dvs). To use such a design, the proposed
covariates should meet a fairly stringent set of requirements. Going over
them is beyond the scope of this response. I am not advocating any
particular reference, but one I have found useful is:
Tabachnick, B. G., & Fidell, L. S. (1996). Using multivariate statistics
(3rd ed.). New York: Harper-Collins.
You would probably find the chapter on ANCOVA most helpful. This is a
cookbook approach. A more rigorous and explanatory approach can be found in
Maxwell, S. E., & Delaney, H. D. (1990). Designing experiments and analyzing
data: A model comparison approach. Belmont, CA: Wadsworth.
Again, the ANCOVA chapters would be helpful. Hope this helps.
> -----Original Message-----
> From: Peter Caputi [SMTP:Peter_Caputi@UOW.EDU.AU]
> Sent: Thursday, April 08, 1999 4:37 PM
> To: SPSSX-L@UGA.CC.UGA.EDU
> Subject: covariance analysis
> Please excuse me if the solution to this problem is trivial.
> I have the followig design 2 within subjects variables, each with three
> levels and one between group variable. I want to use varying
> covariates in my analysis, but I want to use two covariates for the
> first and third level of my outer most within subjects var, and only one
> for the second level of that variable? That is, let A =(A1, A2,A3) and
> B =(B1,B2, B3) be my within Ss vars, then I would liek to use two
> different covars for A1 and A3, and a third covariate for A2.
> What would the spss sytnax look like? Can I do this problem in GLM or do
> I need to go back to the MANOVA command.
> many thanks, Peter