normal vs gamma distributed frailty term in discrete time duration models

[Michael Keller]

Hi,

I am running a duration model with discrete time data. When testing for unobserved heterogeneity I find that assuming a normal distributed heterogeneity term as well as
a gamma distributed heterogeneity term both assumptions seem to be significant and result in a better model than the model without unobserved heterogeneity.

My question is now how to chose between normal or gamma distributed heterogeneity term?
Is there a test statistic or something I could use?

thanks,
Michael