See references. Burnham and Anderson (2002). This
correction to the AIC is derived under some assumption of
normality, so it's validity may depend on there being large
counts for each capture pattern so that the multivariate
normal approximation to the multinomial is good.The AIC methods for VGAM purport to have the AICc option
for the vglm, but this did not seem to exist for the
vgam as of October 2013.