turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- SAS Proc Mixed random statement

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

09-07-2010 02:03 PM

Hi there,

I tried to do a mixed model with AR(1)+RE variance structure. I did it in two ways, as below. Here id is the subject id. There are two covariates: treatment (trt) and month (continuous).

proc mixed data=one;

class id;

model y= trt month /s;

repeated / type=ar(1) sub=id;

random id;

run;

proc mixed data=one;

class id;

model y= trt month /s;

random id / type=ar(1);

run;

However, they produced very different results. I checked the book "SAS for Mixed Models" Second Edition. It seems the first one is correct. But I wonder why the second one failed, though it converged with 3 covariance parameters: variance, AR(1) and residual. Does anyone know what is the actual covariance structure for the second model? Thanks a lot!

Lei

I tried to do a mixed model with AR(1)+RE variance structure. I did it in two ways, as below. Here id is the subject id. There are two covariates: treatment (trt) and month (continuous).

proc mixed data=one;

class id;

model y= trt month /s;

repeated / type=ar(1) sub=id;

random id;

run;

proc mixed data=one;

class id;

model y= trt month /s;

random id / type=ar(1);

run;

However, they produced very different results. I checked the book "SAS for Mixed Models" Second Edition. It seems the first one is correct. But I wonder why the second one failed, though it converged with 3 covariance parameters: variance, AR(1) and residual. Does anyone know what is the actual covariance structure for the second model? Thanks a lot!

Lei

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

09-07-2010 08:41 PM

The second model states (more or less) that, for ordered IDs, the person-specific mean values follow an AR(1) structure. That is, suppose that you had 5 individuals who had the following ID's and mean values:

__ ID __ __ mean __

243 18.21

129 18.96

141 19.44

227 26.33

313 22.71

Now, if we order this table by ID, we have

__ ID __ __ mean __

129 18.96

141 19.44

227 26.33

243 18.21

313 22.71

The ordered mean values are 18.96, 19.44, 26.33, 18.21, and 22.71. There is a variance to these mean values. That is what the first of the three covariance parameters is estimating. The AR(1) parameter is estimating a covariance structure in which individuals who have closer ID values would have stronger covariance of the mean values - an assumption that I doubt you want to make. The last of the three covariance parameters, the residual variance, is estimating the variability of person-specific values about their mean values.

243 18.21

129 18.96

141 19.44

227 26.33

313 22.71

Now, if we order this table by ID, we have

129 18.96

141 19.44

227 26.33

243 18.21

313 22.71

The ordered mean values are 18.96, 19.44, 26.33, 18.21, and 22.71. There is a variance to these mean values. That is what the first of the three covariance parameters is estimating. The AR(1) parameter is estimating a covariance structure in which individuals who have closer ID values would have stronger covariance of the mean values - an assumption that I doubt you want to make. The last of the three covariance parameters, the residual variance, is estimating the variability of person-specific values about their mean values.