Heckman selection model

[donghyeon_ko]

I asked the question about missing data at this post:

Missing data handling

for(j in 1:N) { // see http://www.stata.com/manuals14/rheckman.pdf#rheckmanMethodsandformulas if (R[j] == 1) { real err = (y[j] - x[j] * beta) / sigma; target += normal_lcdf((z[j] * gamma + err * rho) / sqrt(1 - square(rho)) | 0, 1) - 0.5 * square(err) - log(sqrt(2 * pi()) * sigma); } else target += -normal_lcdf(-z[j] * gamma | 0, 1); }

which was the part of this post.

So, I tried to build the model in the link you sent and I think I did it correctly, but the result is off for some reason.

I tried real<lower=-1,upper=1> rho; but this doesn’t make any difference either.

Do you seen why it’s giving off result?