topic Re: Probit model with continuous dependent variable in Statistical Procedures

Probit model with continuous dependent variable

AhmetHakan — Thu, 12 Aug 2021 09:43:12 GMT

Can the probit model with continuous dependent variable be estimated in SAS?

I have tried proc genmod, proc logistic, proc probit, etc. However, all of them require to have a binary dependent variable.

Re: Probit model with continuous dependent variable

Rick_SAS — Thu, 12 Aug 2021 10:53:50 GMT

By definition, a probit model has a binary response variable. Can you say more about your variables and the model you want?

Re: Probit model with continuous dependent variable

StatDave — Thu, 12 Aug 2021 13:23:14 GMT

See this note. Read the introductory text that will help you decide if you just need to use the events/trials response syntax for aggregated binary data, or use a model described further below for truly continuous proportions.

Re: Probit model with continuous dependent variable

SteveDenham — Fri, 13 Aug 2021 14:40:12 GMT

Have you tried PROC GLIMMIX with a beta distribution and a probit link? So long as your proportions fall in the open interval (0, 1), this would seem like a valid approach.

SteveDenham

Re: Probit model with continuous dependent variable

AhmetHakan — Wed, 18 Aug 2021 14:51:56 GMT

Many thanks for your comment, Rick. The variable is the between 0 and 1. I want to estimate the probit model with fixed effects and double-clustered standard error.

Re: Probit model with continuous dependent variable

AhmetHakan — Wed, 18 Aug 2021 14:52:30 GMT

Hi Steve,

It is great! It indeed works! Is it possible to use fixed effect in PROC GLIMMIX?

Re: Probit model with continuous dependent variable

SteveDenham — Wed, 18 Aug 2021 15:02:23 GMT

Fixed effects are included in the MODEL statement, random effects in the RANDOM statement. So you can add fixed effects to your analysis without difficulty.

SteveDenham

Re: Probit model with continuous dependent variable

AhmetHakan — Wed, 18 Aug 2021 15:05:54 GMT

Great, thanks!
I know I did ask many questions. Sorry for this. What about clustered standard error? Can we use double-clustered standard error?

Re: Probit model with continuous dependent variable

SteveDenham — Wed, 18 Aug 2021 15:09:09 GMT

Clustered and double-clustered standard errors sound to me like something you could get using appropriate random effects. Could you give a more concrete example of what the data may look like as far as these variables?

SteveDenham

Re: Probit model with continuous dependent variable

AhmetHakan — Wed, 18 Aug 2021 15:13:29 GMT

In the data, I have two ids: year and company ticker.
For each year/company, I have a key dependent variable (lets say Y), which is between 0 and 1. And for each year/company, I have various explanatory variables (lets say X1, X2, etc) like stock trading volume, etc.

Re: Probit model with continuous dependent variable

SteveDenham — Thu, 19 Aug 2021 12:00:24 GMT

So level-1 cluster would be year within company, and level-2 would be company for what you want?

SteveDenham