Detailed description |
I have come across multiple instances of the model not return standard errors/inf standard errors. Excluding individual variables by trial and error fixes the problem, but there is no specific variables that triggers the NA/Inf SEs.
The model produces the hessian matrix. When trying to manually solve the hessian for the coefficients (sqrt(diag(solve(-model$hessian))) I get an error message: system is computationally singular. I have not been able to find a solution for this online.
If I reproduce the estimation with STATAs heckman command, Standard errors are produced, while coefficients are the same.
Unfortunately I can not provide a reproducible example, as my data lies on a protected server. I would appreciate any help to fix this as my only current workaround is to switch to STATA.
R version is 4.1.3, SampleSelection version is 1.2-12
|
|