Closed shabbychef closed 6 years ago
This was indeed an unsolved annoyance of the method. De-meaning the input is unfortunately not the way to go: the method would the compute a moment of the input, which is exactly what I want to avoid. However, you are right about forcing the average of the input to zero. The solution I will implement is to apply fit.tuv on c(x,-x).
In addition, a forthcoming new version allows one to bypass fit.tuv by inputing a value for nu if it is known by some other means.
While the
fit.tuv
procedure can, in theory, deal with an offset via themu
parameter, the fitting ofnu
should give the same results independently of the empirical mean. Subtracting the empirical mean makes the procedure more robust.