question on glm() and lm.influence()

[impromptu@nifty.com]

S-Plus users
E-mail: s-news@wubios.wustl.edu

Hi!

   Let me make my question on glm() and lm.influence() more
specific.
   I carried out numerical experiments using simulation data
by qpois() (poisson distribution). And weighted poisson regression
by glm() was implemented to smooth histograms. 
The sum of trace of Hat matrix for each data set is
something like:
17.26824
17.26824
17.26824
17.26824
17.26824

The results by my original codes with the same data sets are:
17.40405
17.37464
17.39412
17.31815
17.36982

    It is natural that the values of number of freedom by 
my original code does not agree with those by lm() and 
lm.influence() because initial values and the number of
iteration in generalized linear regression are not the same.
But it is not natural that the values of number of freedom by
glm() and lm.influence() are identical in spite of
the variation of the simulation data.

   *****    Kunio Takezawa, Ph.D. (takezawa@affrc.go.jp)    *****
*****     <http://cse.inada.affrc.go.jp/takezawa/patent-e.html>    *****