Closed SzymonNowakowski closed 3 months ago
On the other hand, grpreg
instructions on a bottom of page 15 clearly state that the loss
variable is equal to RSS in case of a gaussian family.
I leave this issue for your consideration as a possible inconsistency between different sources of information on your package.
There is
fit$loss
variable iffit
is a model returned bygrpreg(...)
call. Andfit$loss
is equal to RSS, but it should be equal to 1/2 of RSS, i.e. negative log-likelihood, according to Models vignette, Gaussian (linear regression) Section. See the reproducible example below:The correct value would be that of 1/2 of RSS.