covariates in repeated measures GLM

[Larry Hawk]

I'm running into some trouble incorporating a covariate into a repeated
measures GLM.  The model is below.  The major issue is that when I
include "WITH fsiq" to covary full-scale IQ, I get tests of FSIQ with
all within-subjects effects.  My understanding is that when evaluating a
covariate that such effects can be important for testing heterogeneity
of regression coefficients, but that when I run the actual ANCOVA I do
not want to include those interactions.  Perhaps I am misunderstanding
(I certainly find covariates much easier to think about in a regression
procedure).  What am I missing?  Or is it the case that I simply need to
suppress the interactions and that there is a way to do that?

Thanks in advance!
Larry Hawk

TITLE 'PPI - all soas by med - med v no med'.
GLM
  ppi120pi ppi120pa ppi120loi ppi120loa ppi120hii ppi120hia ppi180pi
ppi180pa ppi180loi ppi180loa ppi180hii ppi180hia WITH fsiq
  /WSFACTOR = soa 2 med 3 special (1  1  1
                        -2  1  1
                          0 -1  1) attend 2
  /MEASURE = PPI
  /METHOD = SSTYPE(3)
  /EMMEANS = TABLES(med*attend) COMPARE (attend)
  /EMMEANS = TABLES(med*soa*attend) COMPARE (attend)
  /PRINT = DESCRIPTIVE TEST(MMATRIX)
  /WSDESIGN
  /DESIGN.

--

Larry W. Hawk, Jr., Ph.D.
Associate Professor of Psychology
Park Hall, Box 604110
The University of Buffalo, SUNY
Buffalo, NY  14260-4110
Phone:  [phone] x231
Fax:    [phone]
E-mail: [hidden email]

[Swank, Paul R]

It's only ANCOVA if you have a grouping variable (between susbjects) and
the homogeneity assumption is ofr the grouping variable by the covariate
interaction. If you had a grouping variable here, it would also have
interaction terms with all the within subject effects. It is possible to
drop the interactions but I would suggest not doing so.

Paul R. Swank, Ph.D. Professor
Director of Reseach
Children's Learning Institute
University of Texas Health Science Center-Houston


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Larry Hawk
Sent: Thursday, October 11, 2007 12:52 PM
To: [hidden email]
Subject: covariates in repeated measures GLM

I'm running into some trouble incorporating a covariate into a repeated
measures GLM.  The model is below.  The major issue is that when I
include "WITH fsiq" to covary full-scale IQ, I get tests of FSIQ with
all within-subjects effects.  My understanding is that when evaluating a
covariate that such effects can be important for testing heterogeneity
of regression coefficients, but that when I run the actual ANCOVA I do
not want to include those interactions.  Perhaps I am misunderstanding
(I certainly find covariates much easier to think about in a regression
procedure).  What am I missing?  Or is it the case that I simply need to
suppress the interactions and that there is a way to do that?

Thanks in advance!
Larry Hawk

TITLE 'PPI - all soas by med - med v no med'.
GLM
  ppi120pi ppi120pa ppi120loi ppi120loa ppi120hii ppi120hia ppi180pi
ppi180pa ppi180loi ppi180loa ppi180hii ppi180hia WITH fsiq
  /WSFACTOR = soa 2 med 3 special (1  1  1
                        -2  1  1
                          0 -1  1) attend 2
  /MEASURE = PPI
  /METHOD = SSTYPE(3)
  /EMMEANS = TABLES(med*attend) COMPARE (attend)
  /EMMEANS = TABLES(med*soa*attend) COMPARE (attend)
  /PRINT = DESCRIPTIVE TEST(MMATRIX)
  /WSDESIGN
  /DESIGN.

--

Larry W. Hawk, Jr., Ph.D.
Associate Professor of Psychology
Park Hall, Box 604110
The University of Buffalo, SUNY
Buffalo, NY  14260-4110
Phone:  [phone] x231
Fax:    [phone]
E-mail: [hidden email]

[Maguin, Eugene]

All,

Does anyone know if there is a listserv devoted to meta analysis? If not,
then does anyone have any ideas about how to find out if there is one? Since
I know there is a missing data listserv so I tried searching on 'missing
data listserv' and nothing useful showed up quickly. When I tried 'meta
analysis listserv' nothing useful showed up as well. Some will know that
stat transfer maintains a page listing a number of statistics related
listservs. I checked and nothing on there. Their list apparently came from
somebody who compiled a list of lists. Going to that list is a deadend
because the compiler's homepage has been taken down. If anyone has direct
knowledge of such as list or suggestions about how to search for listservs
addressing a topic, I'd love to hear back.

Thanks, Gene Maguin

[Catherine Kubitschek]

I just went out to Google Groups ( http://groups.google.com/grphp?tab=wg )
and searched for "meta analysis" (in quotes).  I got way too many hits but
sci.stat.consult and sci.stat.edu show up on the first page.  You might
find something useful.

Catherine

At 10/11/2007 05:19 PM, Gene Maguin wrote:

>All,
>
>Does anyone know if there is a listserv devoted to meta analysis? If not,
>then does anyone have any ideas about how to find out if there is one? Since
>I know there is a missing data listserv so I tried searching on 'missing
>data listserv' and nothing useful showed up quickly. When I tried 'meta
>analysis listserv' nothing useful showed up as well. Some will know that
>stat transfer maintains a page listing a number of statistics related
>listservs. I checked and nothing on there. Their list apparently came from
>somebody who compiled a list of lists. Going to that list is a deadend
>because the compiler's homepage has been taken down. If anyone has direct
>knowledge of such as list or suggestions about how to search for listservs
>addressing a topic, I'd love to hear back.
>
>Thanks, Gene Maguin

[Larry Hawk]

Thank you for pointing out the grouping variable issue.  In a similar
GLM (different study), we again have repeated measures, but we also have
a grouping variable.  The groups do not differ on the covariate and the
relationship between the covariate and the DV does not vary across
groups.  So, it's classic ANCOVA.  Now, why does SPSS insist on putting
in the interactions of the covariate with all of the within-subjects
contrasts/factors?  And can I suppress them?  Or ignore them?

Swank, Paul R wrote:

>
>It's only ANCOVA if you have a grouping variable (between susbjects) and
>the homogeneity assumption is ofr the grouping variable by the covariate
>interaction. If you had a grouping variable here, it would also have
>interaction terms with all the within subject effects. It is possible to
>drop the interactions but I would suggest not doing so.
>
>Paul R. Swank, Ph.D. Professor
>Director of Reseach
>Children's Learning Institute
>University of Texas Health Science Center-Houston
>
>
>-----Original Message-----
>From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
>Larry Hawk
>Sent: Thursday, October 11, 2007 12:52 PM
>To: [hidden email]
>Subject: covariates in repeated measures GLM
>
>I'm running into some trouble incorporating a covariate into a repeated
>measures GLM.  The model is below.  The major issue is that when I
>include "WITH fsiq" to covary full-scale IQ, I get tests of FSIQ with
>all within-subjects effects.  My understanding is that when evaluating a
>covariate that such effects can be important for testing heterogeneity
>of regression coefficients, but that when I run the actual ANCOVA I do
>not want to include those interactions.  Perhaps I am misunderstanding
>(I certainly find covariates much easier to think about in a regression
>procedure).  What am I missing?  Or is it the case that I simply need to
>suppress the interactions and that there is a way to do that?
>
>Thanks in advance!
>Larry Hawk
>
>TITLE 'PPI - all soas by med - med v no med'.
>GLM
>  ppi120pi ppi120pa ppi120loi ppi120loa ppi120hii ppi120hia ppi180pi
>ppi180pa ppi180loi ppi180loa ppi180hii ppi180hia WITH fsiq
>  /WSFACTOR = soa 2 med 3 special (1  1  1
>                        -2  1  1
>                          0 -1  1) attend 2
>  /MEASURE = PPI
>  /METHOD = SSTYPE(3)
>  /EMMEANS = TABLES(med*attend) COMPARE (attend)
>  /EMMEANS = TABLES(med*soa*attend) COMPARE (attend)
>  /PRINT = DESCRIPTIVE TEST(MMATRIX)
>  /WSDESIGN
>  /DESIGN.
>
>--
>
>Larry W. Hawk, Jr., Ph.D.
>Associate Professor of Psychology
>Park Hall, Box 604110
>The University of Buffalo, SUNY
>Buffalo, NY  14260-4110
>Phone:  [phone] x231
>Fax:    [phone]
>E-mail: [hidden email]
>
>
>
>

--
--

Larry W. Hawk, Jr., Ph.D.
Associate Professor of Psychology
231 Park Hall, Box 604110
The University of Buffalo, SUNY
Buffalo, NY  14260-4110
Phone:  [phone] x231
Fax:    [phone]
E-mail: [hidden email]