拟泊松模型的 Akaike 准则 (AIC)

Question

I run quaisi-Poisson glm models in R and I want to compute AIC criterion. I run the code below but I am not sure if it is correct. Also, I find wierd the fact that the AIC decreases while degrees of freedom increase.

fqaic <- function(model) {
  loglik <- sum(dpois(model$y, model$fitted.values, log = TRUE))
  phi <- summary(model)$dispersion
  qaic <- -2*loglik + 2*summary(model)$df[3]*phi
  return(qaic)
}

I would be gratefull for any advice.

Answer 1

There are some notes here , indicating that among other things that there are existing methods for computing qAIC in the bbmle , AICcmodavg , and MuMIn packages. The correct formula for qAIC is

-2 * logLik/dispersion + 2 * df