Date:         Mon, 24 May 1999 10:27:20 -0500
Reply-To:     "Bauer, John" <bauer@SPSS.COM>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From:         "Bauer, John" <bauer@SPSS.COM>
Subject:      Re: new to glm, a couple of ?'s

Craig,
 
The sum of squares due to the intercept is calculated in the same way as the
other sums of squares.  Assuming you're working with the default, Type III
sums of squares:  First calculate the sum of squares due to the full model;
then fit a reduced model, omitting the effect in question (in this case the
intercept).  The difference in the two model sums of squares is attributed
to the effect in question (e.g. the intercept).
 
Like MANOVA, GLM is also using listwise deletion when there are missing
values.  The fractional degrees of freedom are a consequence of the random
effect in the design.  They are computed using Satterthwaite's
approximation.
 
The references below are only for Sattertwaite's approximation; perhaps
someone else can suggest a 'good' reference on the use of SPSS GLM.
 
 
John Bauer, Ph.D.
SPSS Support Statistician
 
REFERENCES
SPSS 7.5 Statistical Algorithms, pp. 260-262
 
Satterthwaite, F. E. 1946.  An approximate distribution of estimates of
variance components.  Biometrics Bulletin, 2: 110-114.
 
-----Original Message-----
From: craig enders [mailto:cenders@UNLINFO.UNL.EDU]
Sent: Monday, May 24, 1999 8:22 AM
To: SPSSX-L@UGA.CC.UGA.EDU
Subject: new to glm, a couple of ?'s
 
 
I am trying to convert from using manova to glm and have a couple
of questions about the glm output/calculations.  First, a simple one.
In running a simple one-way anova in glm you get a sum of squares
value for the intercept term.  How is that being calculated?  Second,
I was running a mixed design that had a couple of missing data points.
In manova, these points were excluded and I was left with only cases
that had complete data in all cells.  However, in running the same
analysis under glm I got slightly different results -- the df were
not integers. Unfortunately, I cannot remember if it was the between
or within factor that was affected (I think it was the within).  What
is the difference in the calculations under glm when there are missing
observations?  Also, if anyone could provide me a good reference
material on glm I would appreciate it.
Thanks in advance,
Craig Enders
 
--
Craig Enders
cenders@unlinfo.unl.edu