topic Log-GLM with response set to negative binomial distribution in SAS Data Science

Log-GLM with response set to negative binomial distribution

BMI_kid — Thu, 02 Nov 2023 15:44:46 GMT

What does the warning mean?

Re: Log-GLM with response set to negative binomial distribution

PaigeMiller — Thu, 02 Nov 2023 15:54:00 GMT

https://communities.sas.com/t5/Statistical-Procedures/How-do-I-solve-Negative-Binomial-convergence-problem/m-p/391228#M20409

Also I believe the algorithm could have found a local optimum rather than a global optimum.

Re: Log-GLM with response set to negative binomial distribution

BMI_kid — Mon, 06 Nov 2023 08:18:05 GMT

Does it mean the response does not fit the specified distribution or is it something else. I am not sure I understand 'Lack of convergence.'

Re: Log-GLM with response set to negative binomial distribution

SteveDenham — Wed, 20 Dec 2023 14:19:36 GMT

Lack of convergence means that the g'H{inverse}g/abs(f) is not less than the tolerance set, which seems to be 1e-4 in this case. There are a lot of issues that can cause this - model misspecification relative to the data at hand, getting stuck in a local extremum (minimum or maximum) or even running up against a default number of iterations. I can't really tell from that particular output table, so a bit more info would help - the code for the PROC you are using and the LOG file, if there are any ERRORs, WARNINGs or NOTEs that relate to this.

SteveDenham