glm
. It describes a negative binomial
(as negative.binomial
in the MASS package), but with a special link function, namely
eta[i] = log( mu[i] / sf[i] ), i.e., each count value is divided by its size factor before
the log is taken. This is used internally by fitNbinomGLMs
.
nbkd.sf(r, sf)
dnbinom
), i.e., the reciprocal of
the dispersion.