topic Re: Proc mi with no intercept in Statistical Procedures

Proc mi with no intercept

MsGeritO — Thu, 28 Jan 2021 13:35:28 GMT

Hello!

I run imputation models for many subgroups of a population. Before I run my proc mi procedure I do a model selection for each subgroup. For some subgroups the optimal model would not contain an intercept.

I could not find any way of running proc mi with no intercept. The equivalent in proc reg or proc glm would be

proc reg data=have;
model dependent = &macrolistofindependent./ noint selection=adjrsq sse aic ;
run; quit;

Does someone know how to run proc mi with no intercept?

Thank you for any pointers!

Kind regards,

Gerit

Re: Proc mi with no intercept

SAS_Rob — Thu, 28 Jan 2021 14:54:52 GMT

Proc MI does not have anything analogous to the NOINT option.

In general the imputation model and the analyst model should be the same, but in practice that is not always the case. Schafer(1997) has shown a lot of work on this, but in general he found that if the analyst's model is reduced from the imputer's model (like in the case of Intercept and no-intercept models) then the imputer's model is still valid provided that the analyst's assumption is correct. The MI documentation has a full discussion of this

https://go.documentation.sas.com/?cdcId=pgmsascdc&cdcVersion=9.4_3.4&docsetId=statug&docsetTarget=statug_mi_details56.htm&locale=en

Re: Proc mi with no intercept

MsGeritO — Fri, 29 Jan 2021 08:57:24 GMT

Thank you, Rob, for the weblink. It was helpful indeed!