topic Re: SAS two-part model in Statistical Procedures

SAS two-part model

uzma03505621 — Mon, 01 Mar 2021 14:59:42 GMT

Hello everyone,

I am a new SAS user and I am using Medical Expenditure Panel Survey( MEPS) for my research. I have to use a Two-part model to study the impact of various independent variables on my dependent variable ( COST). Can someone please help me with the SAS SYNTAX for the two-part model.

Re: SAS two-part model

mkeintz — Mon, 01 Mar 2021 20:31:55 GMT

I am unfamiliar with the term "two-part model". Can you describe?

Re: SAS two-part model

uzma03505621 — Mon, 01 Mar 2021 20:43:39 GMT

It's a statistical method used when the expenditure data is not normal I.e. it can be rightly skewed or with excess zero expenditures. So the first part involves a binary logit model to differentiate positive and zero expenditures and the second part of the model involves a glm model with gamma distribution and link function.
I'm attaching the link to the article I used as my reference :
https://scholarworks.gvsu.edu/fsdg/556/

Re: SAS two-part model

StatDave — Mon, 01 Mar 2021 20:54:34 GMT

These are zero-inflation models which are in the class of finite mixture models as can be fit using PROC FMM. While there is some question of the appropriateness of doing this with a continuous distribution like gamma, the following code is an example with single predictor, G, and will provide a model fit and predicted values.

proc fmm data=a plots=density(nbins=9);
   class g;
   model y = g / dist=gamma;
   model   +  / dist=constant;
   probmodel g;
   output out=fmmpred pred;
   run;

Re: SAS two-part model

uzma03505621 — Mon, 01 Mar 2021 21:00:51 GMT

Can you share any source for me to interpret the output of this code?

Re: SAS two-part model

StatDave — Mon, 01 Mar 2021 21:07:51 GMT

...but perhaps a better, and more easily interpreted, model is the Tweedie model that can be fit in PROC GENMOD with the DIST=TWEEDIE option. The Tweedie distribution can model a nonnegative response with a mass at zero, so it is often used to model cost when there can be a set of zero values. The output is much like an ordinary gamma model since it isn't a "two part" model.

Re: SAS two-part model

Rick_SAS — Wed, 03 Mar 2021 21:58:02 GMT

The Getting Started examples in the documentation often explain how to interpret the results.

Re: SAS two-part model

Season — Tue, 18 Mar 2025 13:29:07 GMT

I am also interested in building this kind of model. Could you please explain why there it might be inappropriate to build such models for continuous distributions in more detail? Thank you!

Re: SAS two-part model

Ksharp — Wed, 19 Mar 2025 01:24:15 GMT

As @StatDave said it is a zero-inflation model generally suited to COUNT data ,not a continous data.

Check this brand-new session:

https://communities.sas.com/t5/SAS-Communities-Library/Making-Zero-Inflation-Count/ta-p/962019/jump-to/first-unread-message