Learning SAS? Welcome to the exclusive online community for all SAS learners.

SAS EG 4.3 linear regression question - robust SE?

Accepted Solution Solved
Reply
Occasional Contributor
Posts: 11
Accepted Solution

SAS EG 4.3 linear regression question - robust SE?

I'm working with observational PUMS data.  I have cut it down to Age, Sex, education, etc.  and I am running a linear regression through SAS with Wages as the dependent variable.  I would like to change the Standard Errors to make them robust in order to decrease the implications of clustering (household data).

If you tell me to write some code your going to have to tell me where.  I am much more used to eviews, jmp, SPSS, excel if there is a way that's similar to them.  I'm at the end of my rope!

.......

After looking at other people's requests for help, it's almost 100% that the people helping say they haven't given enough information...  so here's more

I'm using data from the 2011 census data. My sample size is very large.  My goal is to find potential wage differences in different populations.

I have created my columns of data.

     - I click on the filtered data I want to use

     - Select analyze -> regression -> linear regression

     - I put wages in the dependent variable spot

     - I put age, age squared, and several dummy variables as the explanatory variables  (race, sex, education, etc)

I have looked through "statistics," "plots," and predictions and have not found what I'm looking for.  I've looked all over online and can't find anything.

hopefully this helps.


Accepted Solutions
Solution
‎04-22-2014 06:52 PM
SAS Employee
Posts: 89

Re: SAS EG 4.3 linear regression question - robust SE?

Tglass55,

I am guess you are using "robust" to mean Huber-White SE's. Try this out.

proc surveyreg data = pums;
  cluster household;
  model wage = female age education etc;
run;


Good luck-Ken

View solution in original post


All Replies
SAS Employee
Posts: 35

Re: SAS EG 4.3 linear regression question - robust SE?

In general, clustered data can be handled with a Mixed model, using the mixed model task. If the data are not measured at the individual level and it isn’t possible to fit a mixed model, then I’d need to know more about the model and what you mean by robust. Sometimes, robust means “robust to misspecification of the correlation structure.” This can be done with a mixed model, and adding the empirical sandwich estimator for the standard errors. You’ll run the Mixed Models task and then add EMPIRICAL to the PROC MIXED statement in code. Sometimes, robust means, “robust to outlying/influential observations”. This should be handled with robust regression. If the goal is to fit a robust regression, then code will be required, using PROC ROBUSTREG. There are lots of examples in the documentation, but there is not a task in EG for this. Finally, I notice that you are modeling wages as the DV. This is almost certainly not normally distributed; a Gamma distribution often works well here. This model can be fit with the generalized linear models task (which uses PROC GENMOD). If the observations are correlated, then you can use a working correlation matrix in the estimation with the REPEATED statement. You would have to use code for this. Alternatively, you can fit similar models in the GLIMMIX procedure as well.

The above response from Cat Truxillo, a statistical trainer in SAS Education

Solution
‎04-22-2014 06:52 PM
SAS Employee
Posts: 89

Re: SAS EG 4.3 linear regression question - robust SE?

Tglass55,

I am guess you are using "robust" to mean Huber-White SE's. Try this out.

proc surveyreg data = pums;
  cluster household;
  model wage = female age education etc;
run;


Good luck-Ken

New Contributor
Posts: 3

Re: SAS EG 4.3 linear regression question - robust SE?

How do find and apply Robust to the standard mean error in Enterprise Guide with out programing int into the code?

Community Manager
Posts: 552

Re: SAS EG 4.3 linear regression question - robust SE?

Hi @guybee, if you'll post on the SAS Enterprise Guide board, more experts who can help you will see it. Some more detail in your inquiry will also help you get a good answer fast; here's a how-to article on that.

☑ This topic is SOLVED.

Need further help from the community? Please ask a new question.

Discussion stats
  • 4 replies
  • 1090 views
  • 2 likes
  • 5 in conversation