topic Re: Multinomial Probit in SAS Programming

Multinomial Probit

Kyungjae — Tue, 04 Jan 2022 06:37:38 GMT

Hello,

I am using SAS 9.4 to estimate multinomial probit model. The dependent variable has the values 1, 2, 3, 4 and there are about 10 (binary and continuous) independent variables. However, I could not find the proper SAS Procedure for multinomial probit regression except Proc MDC. Is there any procedure for multinomial probit model?

Thanks

Re: Multinomial Probit

Ksharp — Tue, 04 Jan 2022 12:01:50 GMT

Search it at
https://support.sas.com/kb/30/333.html

Re: Multinomial Probit

SteveDenham — Tue, 04 Jan 2022 13:02:58 GMT

Look at the examples for PROC GENMOD - there is an example that fits a multinomial response with a cumulative logit distribution. Changing to a cumulative probit is done with the LINK= option in the MODEL statement. I would recommend running the example program with the cumulative probit to see how the program performs, and then adapting it to your particular design.

SteveDenham

Re: Multinomial Probit

Kyungjae — Tue, 04 Jan 2022 13:42:14 GMT

Thank you for your advice, but cumulative probit shows the result of Ordered Probit Model. I tried several procedures which involve cumulative probit as a link function including Proc Genmod and Proc Gee, and none of them showed the result of Multinomial Probit Model.

Re: Multinomial Probit

SteveDenham — Tue, 04 Jan 2022 14:33:53 GMT

Have you tried a generalized logit link? A generalized logit link is available in PROC GLIMMIX. My thought is that a generalized probit would give approximately the same results, so this may be an alternative.

If you stay with GENMOD, you will probably have to write your own link functions using the FWDLINK and INVLINK statements, to get the generalized types.

Since the probit distribution assumes an underlying normal distribution with "cutpoints", I am not sure that a generalized probit is particularly meaningful. The alternative might be to set up various binary classifications. That would probably require some (meaning, A LOT) of post processing programming.

SteveDenham