Date:         Sat, 14 Oct 2000 17:23:42 -0700
Reply-To:     Numchai Supparerkchaisakul <numchai1@YAHOO.COM>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Numchai Supparerkchaisakul <numchai1@YAHOO.COM>
Subject:      How to specify the error term in GLM for calculating mixed model
Content-Type: text/plain; charset=us-ascii

Dear All SPSS Users,

I am a doctoral student in behavioral science and I
have a problem to ask you about GLM when calculating
Mixed Model. According to the syntax reference, one of
the new key
features unavailable in MANOVA is "Automatic handling
of random effects in random-effects models and mixed
models, including
generation of expected mean squares and automatic
assignment of proper error terms." However, when I
analyzed the mixed model,
the error term of the random factor is not correct.
For example, when I write the command like this:

GLM y by A, B
  /Random=B.

This is the mixed model of which fixed factor is A
random factor is B. The appropriate error terms for
fixed factor and random factor should be MS(A*B)
and MS(error) respectively, but the printout of this
command for testing significance of factor B is not
correct. The appropriate error term for random factor
B should be MS(error) but GLM used MS(A*B) instead.
The result shown in the printout should be random
model, both factors are random factors, because it
used interaction term as an error term
for both factor. Therefore, the result of this
analysis is wrong because the error term of factor B
is inappropriate. In the MANOVA command, I can specify
the error term for each subject manually, but
in GLM I can not find that way. In MANOVA, I can write
the command for analyze the mixed model like this:

MANOVA y by a(1,4) b(1,4)
 /error=1
 /design= a, b vs within, a by b = 1.

Although in GLM it can use the subcommand TEST to
specify the error term to test against the effects,
but I do not know how to specify the within sums of
squares (the WITHIN in ERROR subcommand in MANOVA) in
the TEST subcommand . In conclusion, I would like to
ask two main question:

1. Is it true that GLM uses the wrong error term for
fixed factor when analyzing the mixed model, or I
myself made some mistakes? If I am wrong, how could I
write the correct syntax?
2. How can I write the GLM command to analyze this
mixed model that the error term of fixed factor is
MS(error)? How can I specify the within SS in the TEST
subcommand?

In my opinion, I think that MANOVA is more flexible
than GLM, why SPSS support us to use GLM rather than
MANOVA by taking the MANOVA command out of the menu?

I am looking forwards to hearing your precious and
invaluable answer soon and thank you very much in
advance.

Sincerely Yours,

Numchai Supparerkchaisakul.


__________________________________________________
Do You Yahoo!?
Yahoo! Messenger - Talk while you surf!  It's FREE.
http://im.yahoo.com/