topic Re: Convergence criteria for Proc logistics, gemond, glimmix in Statistical Procedures

Convergence criteria for Proc logistics, gemond, glimmix

HappySASUE — Tue, 28 Apr 2020 23:59:48 GMT

Hi,

I try to compare convergence rate using different model procedure: logistic, genmod, and glimmix. However each procedure has difference convergence criterion. I listed some example below. Are these criterion similar to each other? Can anyone give me advice which one I should use ? Thanks so much in advance !

Proc genmod: CICON, CONVH, CONVER, default is 1e-4, INITI

Proc logistic: GCONV : default is 1e-8

proc glimmix: pconv, default is 1e-4

Re: Convergence criteria for Proc logistics, gemond, glimmix

Rick_SAS — Wed, 29 Apr 2020 15:09:48 GMT

You probably don't need to worry about the different criteria. Unless you are having convergence problems, you don't need to change the convergence parameters from their default values. If you are having convergence problems, the reason is often that the model does not fit the data. If so, changing the convergence parameters might not lead to convergence.

From a practical point of view, if the standard errors for the parameter estimates are much much larger than the convergence criterion, then the criterion doesn't play an important role in the point estimates and inferential statistics.

Re: Convergence criteria for Proc logistics, gemond, glimmix

HappySASUE — Wed, 29 Apr 2020 15:41:19 GMT

That makes sense. Thank you !