R help - May 2011 - questions about the output of "gee" and its summary

Dear R-helpers,

I am using the package "gee" to run a marginal model.

Here is the output.
In my simulated data, both x and z are time-varying, so I include their
interaction terms with time indicator (i.e. tind=0, if time 1, and 1 if time
2)
The data is simulated, so the true parameter of z both at time 1 and time 2
is 5, which is very close from the model output
for time 1, z = 5.0757760, and for time 2, z is 5.0757760-0.6379866 = ~5
> model=gee(y~x+z+x*tind+z*tind, family=gaussian(link =
"identity"), id=sid,
corstr="exchangeable")
Beginning Cgee S-function, @(#) geeformula.q 4.13 98/01/27
running glm to get initial regression estimate
(Intercept)                 x                  z            tind
x:tind          z:tind
  2.9342186   1.5002601   5.0757760   2.0846327   0.1869748  -0.6379866



However, when I use the summary command, the coefficients are changed.
Am I missing anything here ??
> summary(model)
 GEE:  GENERALIZED LINEAR MODELS FOR DEPENDENT DATA
 gee S-function, version 4.13 modified 98/01/27 (1998)

Model:
 Link:                      Identity
 Variance to Mean Relation: Gaussian
 Correlation Structure:     Exchangeable

Call:
gee(formula = y ~ x + z + x * tind + z * tind, id = sid, family gaussian(link =
"identity"),
    corstr = "exchangeable")

Summary of Residuals:
       Min         1Q     Median         3Q        Max
-5.9273676 -2.0072725 -0.7169515  2.3709969  8.2377283


Coefficients:
             Estimate  Naive S.E.    Naive z Robust S.E.   Robust z
(Intercept) 4.1450504 0.331866699  12.490106 0.264666616  15.661403
x           1.5155102 0.008479614 178.723972 0.006854627 221.093020
z           0.6463947 0.111148094   5.815617 0.100379444   6.439513
tind        1.5986872 0.163851622   9.756920 0.175947744   9.086148
x:tind      0.1434216 0.005288708  27.118450 0.005924767  24.207123
z:tind      4.2951055 0.168647198  25.467992 0.166776520  25.753658

Estimated Scale Parameter:  6.800334
Number of Iterations:  10



Any helps would be very much appreciated!

Thank you,
Carrie--

	[[alternative HTML version deleted]]

On Sun, May 8, 2011 at 6:36 PM, Carrie Li <carrieandstat at gmail.com>
wrote:
> Dear R-helpers,
>
> I am using the package "gee" to run a marginal model.
>
> Here is the output.
> In my simulated data, both x and z are time-varying, so I include their
> interaction terms with time indicator (i.e. tind=0, if time 1, and 1 if
time
> 2)
> The data is simulated, so the true parameter of z both at time 1 and time 2
> is 5, which is very close from the model output
> for time 1, z = 5.0757760, and for time 2, z is 5.0757760-0.6379866 = ~5
>
>> model=gee(y~x+z+x*tind+z*tind, family=gaussian(link =
"identity"), id=sid,
> corstr="exchangeable")
> Beginning Cgee S-function, @(#) geeformula.q 4.13 98/01/27
> running glm to get initial regression estimate
> (Intercept) ? ? ? ? ? ? ? ? x ? ? ? ? ? ? ? ? ?z ? ? ? ? ? ?tind
> x:tind ? ? ? ? ?z:tind
> ?2.9342186 ? 1.5002601 ? 5.0757760 ? 2.0846327 ? 0.1869748 ?-0.6379866
>
>
>
> However, when I use the summary command, the coefficients are changed.
> Am I missing anything here ??
Yes.

The printout above gives the starting values obtained by running
glm(), not the results of gee().  If you use summary() on the model
object, or just print the model object you will get the results of
gee().

   -thomas

-- 
Thomas Lumley
Professor of Biostatistics
University of Auckland