Dear Splus users:
Could anyone help me with two issues related with the local scoring method
in GAM (described in Hastie and Tibshirani (1990))?
1. In the local scoring method (on Page 141 of Hastie and Tibshirani's
book), eta = alpha + sum(fi). Does alpha change in the loop of the local
scoring procedure? I have read several papers related with this method.
Some papers fix the 'alpha'. Some papers put an index for 'alpha'
indicating loop. If it changes, how do we get the estimate of 'alpha' from
the backfitting procedure? what did SAS and Splus do about 'alpha' when
they implement this procedure?
2. I just want to make sure that the weighted backfitting (called from the
local scoring procedure) means the backfitting method with the weights
used in each smoother.
Any suggestion would be appreciated. Thanks!
Jin Zhu
|