I'm trying to simulate outcomes in a zero-inflated negative binomial model. My model gives an estimate of the dispersion parameter theta, as well as a standard error for log (theta).
My question is, what is the uncertainty distribution around theta? In other words, for my simulation, should I be using:
theta_sim <- rnorm ( 1, mean = theta_hat, sd = exp(SE.logtheta) )
or
theta_sim <- exp (rnorm ( 1, mean = log(theta_hat), sd = SE.logtheta ))
or should it be something else entirely?
