For identity():
for deriv = 0, the identity of theta, i.e.,
theta when inverse = FALSE,
and if inverse = TRUE then theta.
For deriv = 1, then the function returns
dtheta / deta as a function of theta
if inverse = FALSE,
else if inverse = TRUE then it returns the reciprocal.
For nidentity(): the results are similar to identity()
except for a sign change in most cases.
Details
The identity link function $g(\theta)=\theta$
should be available to every parameter
estimated by the VGAM library. However, it usually results in
numerical problems because the estimates lie outside the permitted
range. Consequently, the result may contain
Inf, -Inf, NA or NaN.
The function nidentity is the negative-identity link function and
corresponds to $g(\theta)=-\theta$.
This is useful for some models, e.g., in the literature supporting the
egev function it seems that half of the authors use
$\xi=-k$ for the shape parameter and the other half use $k$
instead of $\xi$.
References
McCullagh, P. and Nelder, J. A. (1989)
Generalized Linear Models, 2nd ed. London: Chapman & Hall.