I am trying to set up a parametric bootstrap using the mean and covariance matrix from my model fitted with nlm() which converged with code 1. However, when creating new parameters using
rmvnorm(1,mean=mod$estimate,sigma=solve(mod$hessian))
I get the following warning message:
Warning message:
In rmvnorm(1, mod$estimate, solve(mod$hessian)) :
sigma is numerically not positive semidefinite
The bootstrap completes without error and the results seem reasonable. My question is how reliable the results are?
