Help using Base SAS procedures

Generalized least squares

Accepted Solution Solved
Reply
Frequent Contributor
Posts: 104
Accepted Solution

Generalized least squares

Hello,

I want to perform locational attainment model, which takes the form:

Yj  = a + b1X1ij+ b2X2ij + ..... + eij 

where Y is a neighbourhood characteristic (such proportion of Black) and the Xs are individual characteristics (such as Black or White among others). Since, I estimate aggregate-level outcomes as a function of individual characteristics, this will generate autocorrelation and underestimation of standard errors. To solve that problem, I thus need to estimate the parameters using the generalized least squares method. My question is: how using that method with SAS?

Thanks


Accepted Solutions
Solution
‎01-16-2015 01:17 PM
Respected Advisor
Posts: 2,655

Re: Generalized least squares

Your response variable most likely does not have normally distributed residuals, at least as you have described it, so GLIMMIX makes more sense than MIXED.  Additionally, it has now had features added so that survey data can more readily be analyzed.

AUTOREG is specialized for time-series data, and that is not what you have described.

Steve Denham

View solution in original post


All Replies
Respected Advisor
Posts: 2,655

Re: Generalized least squares

You may wish to look at PROC GLIMMIX.  In SAS/STAT13.2 documentation for PROC GLIMMIX, the last two examples (44.17 Linear Inference Based on Summary Data, and 44.18 Weighted Multilevel Model for Survey Data) look like they may give you a starting point for your analysis.

Steve Denham

Frequent Contributor
Posts: 104

Re: Generalized least squares

Thanks for you answer.

I just read that the PROC MIXED could also estimate GLS. I'm not very familiar with all those different procedures. For my purpose, is the PROC MIXED will give the same results as the PROC GLIMMIX?

Can I use the PROC AUTOREG for the same purpose too?

Solution
‎01-16-2015 01:17 PM
Respected Advisor
Posts: 2,655

Re: Generalized least squares

Your response variable most likely does not have normally distributed residuals, at least as you have described it, so GLIMMIX makes more sense than MIXED.  Additionally, it has now had features added so that survey data can more readily be analyzed.

AUTOREG is specialized for time-series data, and that is not what you have described.

Steve Denham

☑ This topic is SOLVED.

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

Discussion stats
  • 3 replies
  • 199 views
  • 3 likes
  • 2 in conversation