Hello,

I am doing a parametric survival analysis with a loglogistic baseline hazard function. I will like to control for unobserved heterogeneity through a finite fixture model.

For this purpose, I am running the following code:


Code:
fmm 1, emopts(iterate(1000)) technique(bhhh 30 nr 30) difficult: streg $xlist , distribution($dist)



fmm 2, emopts(iterate(1000)) iterate(400) technique(bhhh 30 nr 30) difficult: streg $xlist , distribution($dist)


fmm 3, emopts(iterate(1000)) iterate(400) technique(bhhh 30 nr 30) difficult: streg  $xlist, distribution($dist)


fmm 4, emopts(iterate(1000)) iterate(400) technique(bhhh 30 nr 30) difficult: streg $xlist, distribution($dist)
I compare the 4 models with BIC, but I believe I have a problem in the output.
My loglikelihood value for the model with model with 3 groups is larger (less negative), than with 4 groups.

Furthermore, it seems like 3 and 4 does not converge. I have tried all kinds of techniques, and if the iterations are not limited, the code keeps running.

Is this possible, and if so, how? Please elaborate.

Sample size:
Observations = 15775
Covariates = 17