The STATA has been running for one day

[Hm Saleh]

Hello,

[nbreg N_CHILDREN HOUSHOLD_INCOME FEMALE_SHARE_EARNING i.EDUCATION i.AGE i.MARITAL_STATUS i.RACE i.HISPANIC
nbreg N_CHILDREN HOUSHOLD_INCOME FEMALE_SHARE_EARNING i.EDUCATION i.AGE i.MARITAL_STATUS i.RACE i.HISPANIC, irr]

I use this code to do NBER regression. I use count data.

The STATA has been running for one day. Is it normal? I got this and it still continues.

"Iteration 2000: log-likelihood = -4761.2478 (not concave)"
I got this and it still continues.

- Could you please tell me if there are any technical steps I need to take to resolve this issue?

[Hm Saleh]

The run works if I drop any variable, but I don't think it is a good solution.

[Clyde Schechter]

If the value of the log-likelihood is not changing at the point where you got

Iteration 2000: log-likelihood = -4761.2478 (not concave)

then Stata is not able to converge to a solution here. If it is changing, then there might be some hope that with patience you will reach a solution. But, frankly, it isn't likely. It is rare that an estimation that has not converged in 2000 iterations ever does.

The fact that dropping any variable leads to convergence in a more reasonable number of iterations suggests that there may be some near perfect colinearity among your variables that makes the model unidentifiable. If that's the case, there will be no alternative to picking one of your variables and dropping it. But let's try to get a picture of what is actually going on first. Look over the output from the original regression and identify at which iteration the log-likelihood gets stuck. Then pick a number that is just slightly larger than that (call it # for purposes of discussion here) and then add the -iterate(#)- option to the original -nbreg- command and run it again. (Replace "#" by the actual number.) That will cause Stata to run the estimation that far and then display interim results. Those interim results are not valid estimates, but they will perhaps give us a clue as to what is causing the problem. So post back showing those interim results and perhaps we can identify what is going wrong. Be sure to surround them by code delimiters so they align nicely.