[R] local linear and local constant kernel regression with np

[Philipp Kunze]

Hi there,

I ran into a weird problem using the np-package doing some local linear
kernel regression. Whenever I run the function "npregbw(...)" with the
option "regtype=ll" (local linear modelling), my optimal bandwidth is
supposed to be 1278946. This is kind of funny, because my regressor data
(189 data points) only runs from about 3.4 to about 5.9. So, what I get
as a result is a nice and straight line, the same one I would get if I
would run a normal linear regression. Whenever I set the option
"regtype=lc" (i.e. local constant modelling), the optimal bandwidth is
calculated as 0.795 - which sounds right to me. 

Any ideas / similar experiences?

Thanks!
Philipp