[R] drop1() seems to give unexpected results compare to anova()

[Thomas P C Chu]

I am not sure why my messages are not threaded together. Thank you to 
the author of this post:
https://stat.ethz.ch/pipermail/r-help/2008-August/169691.html
I have tried the suggestions, but I got the same results as in my 
original query:
https://stat.ethz.ch/pipermail/r-help/2008-August/169647.html

I have considered the issue of partial and sequential sum of squares. 
Given that the variable x4 (the red herring) entered the model last in 
the sequence, I thought partial and sequential SS ought to be 
numerically the same.

However, I later found out that using glm() instead of lm() gave the 
expected results. See:
https://stat.ethz.ch/pipermail/r-help/2008-August/169673.html
I have read that glm() uses iterative re-weighted least square (which I 
think is related to maximum likelihood) for fitting whereas lm() uses 
matrix decomposition. I do expect slightly different answers from these 
two functions, but not ones that were so far apart!

Now I can use glm() as a workaround, but I just want to make sure there 
are no bugs in drop1(). Hopefully more people can give their opinions 
whether there is a bug.

Thomas P C Chu
________________________________________________________________________
AOL Email goes Mobile! You can now read your AOL Emails whilst on the 
move. Sign up for a free AOL Email account with unlimited storage today.