[R] an error to call 'gee' function in R

[Thomas Lumley]

From the gee() help page:

Data are assumed to be sorted so that observations
on a cluster are contiguous rows for all entities in the formula.

 	-thomas


On Sun, 7 Sep 2008, Qinglin Wu wrote:

> Dear List:
>
> I found an error when I called the 'gee' function.  I cannot solve and explain it.  There are no errors when I used the 'geeglm' function.  Both functions fit the gee model.  The project supervisor recommends me to use the 'gee' function.  But I cannot explain to him why this error happens.   Would you help me solve this problem?  I appreciate your help.
>
> In this project I will use the 'gee' or 'geeglm' and 'glmer' to fit the simulated multivariate count responses.  I generated the data like this:
>
> Set ¦Â0 = ¦Â1 = 1, ¦Ì0 = 3, and n = 50.
> For each 1 ¡Ü i ¡Ü n,
> Simulate xi from N (1, 1).
> Simulate zi0 and zit from
> zi0 follows i.d. Poisson (¦Ì0) ,
> zit | xi follows i.d. Poisson (¦Ìit) , 1 ¡Ü t ¡Ü 3,
> log (¦Ìit) = log(E (zit | xi)) = ¦Â0t + xi¦Â1t = 1+xi.
> Let yit = zi0 + zit, 1 ¡Ü t ¡Ü 3.
>
> So my data frame, let me call it 'simdata', the first 10 rows look like this:
>  id y.1 y.2 y.3           x
>   1   3   5   6 -0.06588626
>   2   6   7   6 -0.08265981
>   3   6   8  13  0.58307719
>   4  22  21  28  2.21099940
>   5   5  12   8  1.06299869
>   6   8  21  24  1.47615784
>   7  11   8   9  0.83748390
>   8  16  15  16  1.67011313
>   9   9   7   7 -0.14181264
>  10  31  37  40  2.56751453
>
>
> This is the longitudinal data.  I will change its shape to analyze it. 
> The changed 'newdata' looks like this:

>  id           x time  y
>   1 -0.06588626    1  3
>   2 -0.08265981    1  6
>   3  0.58307719    1  6
>   4  2.21099940    1 22
>   5  1.06299869    1  5
>   6  1.47615784    1  8
>   7  0.83748390    1 11
>   8  1.67011313    1 16
>   9 -0.14181264    1  9
>  10  2.56751453    1 31
> ...........................
>   1 -0.06588626    2  5
>   2 -0.08265981    2  7
>   3  0.58307719    2  8
>   4  2.21099940    2 21
>   5  1.06299869    2 12
>   6  1.47615784    2 21
>   7  0.83748390    2  8
>   8  1.67011313    2 15
>   9 -0.14181264    2  7
>  10  2.56751453    2 37
> ...........................
>   1 -0.06588626    3  6
>   2 -0.08265981    3  6
>   3  0.58307719    3 13
>   4  2.21099940    3 28
>   5  1.06299869    3  8
>   6  1.47615784    3 24
>   7  0.83748390    3  9
>   8  1.67011313    3 16
>   9 -0.14181264    3  7
>  10  2.56751453    3 40
> ...........................
>
>
> My data 'y' comes from x.  So their correlations are not independent.  What does the argument 'corstr' mean it defined in the function.  I tried all choices.  But the error was still there.  Here was the function I used in my programming:
>
> mfit1 <- gee(y~x,data=newdata,family=poisson(link="log"),id=id,corstr="exchangeable")
>
> GEE:  GENERALIZED LINEAR MODELS FOR DEPENDENT DATA
> gee S-function, version 4.13 modified 98/01/27 (1998)
>
> Model:
> Link:                      Logarithm
> Variance to Mean Relation: Poisson
> Correlation Structure:     Exchangeable
>
> Call:
> gee(formula = y ~ x, id = id, data = newdata, family = poisson(link = "log"),
>    corstr = "exchangeable")
>
> Number of observations :  150
>
> Maximum cluster size   :  1
>
> Coefficients:
> (Intercept)           x
>  1.5849653   0.7937203
>
> Estimated Scale Parameter:  1.162505
> Number of Iterations:  1
>
> Working Correlation[1:4,1:4]
> Error in print(x$working.correlation[1:4, 1:4], digits = digits) :
>  subscript out of bounds
>
>
>
> Is this kind of data not fit for the function 'gee'?  Because when I tested this two functions  by using the R data 'warpbreaks' they worked perfect although some returned objects were different. I used the following to do it:
> 1.  (summary(gee(breaks ~ tension, id=wool, data=warpbreaks, corstr="exchangeable"))
> 2.  summary(geeglm(breaks ~ tension, id=wool, data=warpbreaks, corstr="exchangeable"))).
>
> The first one is from the example of ?gee file.
>
> I will attach the part of my programming as .R file.  You can excute in R software.  Thanks a lot.
>
>
> I appreciate your help.
>
>
> Best regards.
>
> Sincerely,
> Cynthia Wu
>

Thomas Lumley			Assoc. Professor, Biostatistics
tlumley at u.washington.edu	University of Washington, Seattle