Problem Note 45223: The Estimates and standard errors in the SolutionR table in PROC GLIMMIX might be incorrect when the response variable has missing observations
 |  |  |  |  |
When the data contains missing values in the middle of the clustered observations for the response variable, the estimates and standard errors in the SolutionR table in the GLIMMIX procedure might be misplaced. That is, the solution estimates might be displayed for the incorrect levels of the random effect. For example, consider the following program:
data test;
input year date $ y;
datalines;
1995 07/24/95 26.2
1995 07/24/95 28.0
1995 07/24/95 27.0
1995 07/25/95 27.3
1995 07/25/95 29.0
1995 07/25/95 28.6
1995 07/25/95 27.4
1995 07/25/95 28.6
1995 07/25/95 28.5
1995 07/25/95 27.6
1995 07/25/95 27.7
1995 07/25/95 28.6
1995 07/25/95 28.5
1995 07/25/95 27.8
1995 07/25/95 27.9
1995 07/25/95 27.8
1995 07/25/95 28.7
1995 07/25/95 28.6
1995 07/25/95 30.2
1995 07/25/95 29.7
1995 07/25/95 28.5
1995 07/25/95 28.1
1995 07/25/95 28.1
1995 07/25/95 28.0
1995 07/26/95 27.5
1995 07/26/95 27.5
1995 07/26/95 27.5
1995 07/26/95 28.3
1995 07/26/95 28.0
1995 07/26/95 28.5
1995 07/26/95 27.8
1995 07/27/95 .
1995 07/28/95 28.4
1995 07/28/95 29.2
1995 07/28/95 30.1
1995 07/28/95 29.9
1996 07/22/96 22.6
1996 07/22/96 22.7
1996 07/22/96 21.7
1996 07/22/96 22.7
1996 07/22/96 22.8
1996 07/22/96 22.0
1996 07/22/96 23.1
1996 07/22/96 23.0
1996 07/22/96 23.2
1996 07/22/96 23.0
1996 07/23/96 22.9
1996 07/23/96 23.0
1996 07/23/96 23.1
1996 07/23/96 23.4
1996 07/24/96 23.5
1996 07/24/96 23.8
1996 07/25/96 .
1996 07/26/96 .
1996 07/27/96 .
1996 07/28/96 .
1996 07/29/96 .
1996 07/30/96 24.4
1996 07/30/96 24.3
1996 07/30/96 24.6
1996 07/30/96 24.7
1996 07/30/96 24.7
1996 07/30/96 24.7
1996 07/30/96 24.8
1996 07/30/96 24.7
;
proc glimmix data=test;
class year date;
model y = / solution cl ddfm=kr;
random int / subject=year s;
random int / subject=date(year) s;
run;
This program produces the following output:
Solution for Random Effects
Std Err
Effect Subject Estimate Pred DF t Value Pr > |t|
Intercept year 1995 2.2912 2.3370 1 0.98 0.5063
Intercept date(year) 07/24/95 1995 -0.9312 0.5133 8.462 -1.81 0.1052
Intercept date(year) 07/25/95 1995 0.2029 0.4681 6.372 0.43 0.6789
Intercept date(year) 07/26/95 1995 -0.2485 0.4839 7.122 -0.51 0.6232
Intercept date(year) 07/27/95 1995 1.1381 0.5008 7.906 2.27 0.0530
Intercept date(year) 07/28/95 1995 0 0.8682 5.018 0.00 1.0000
Intercept 0 0.8682 5.018 0.00 1.0000
Intercept 0 0.8682 5.018 0.00 1.0000
Intercept 0 0.8682 5.018 0.00 1.0000
Intercept 0 0.8682 5.018 0.00 1.0000
Intercept year 1996 -2.2912 2.3370 1 -0.98 0.5063
Intercept date(year) 07/22/96 1996 -0.8361 0.4812 6.964 -1.74 0.1261
Intercept date(year) 07/23/96 1996 -0.4080 0.5046 8.036 -0.81 0.4421
Intercept date(year) 07/24/96 1996 0.07918 0.5402 9.482 0.15 0.8865
Intercept date(year) 07/25/96 1996 1.0036 0.4852 7.152 2.07 0.0766
Intercept date(year) 07/26/96 1996 0 0.8682 5.018 0.00 1.0000
Intercept date(year) 07/27/96 1996 0 0.8682 5.018 0.00 1.0000
Intercept date(year) 07/28/96 1996 0 0.8682 5.018 0.00 1.0000
Intercept date(year) 07/29/96 1996 0 0.8682 5.018 0.00 1.0000
Intercept date(year) 07/30/96 1996 0 0.8682 5.018 0.00 1.0000
In this output, the results for 07/28/95 are mistakenly placed for 07/27/95, and the results for 07/30/96 are mistakenly placed for 07/25/96.
To circumvent the problem and get the correct ordering of the random effect estimates, remove the observations with the missing values for the response by using a WHERE statement or clause similar to the following in PROC GLIMMIX:
where y ne .;
Click the Hot Fix tab in this note to access the hot fix for this issue.
Operating System and Release Information
| Product Family | Product | System | Product Release | SAS Release | ||
| Reported | Fixed* | Reported | Fixed* | |||
| SAS System | SAS/STAT | z/OS | 9.22 | 9.2 TS2M3 | ||
| Microsoft® Windows® for 64-Bit Itanium-based Systems | 9.22 | 9.2 TS2M3 | ||||
| Microsoft Windows Server 2003 Datacenter 64-bit Edition | 9.22 | 9.2 TS2M3 | ||||
| Microsoft Windows Server 2003 Enterprise 64-bit Edition | 9.22 | 9.2 TS2M3 | ||||
| Microsoft Windows XP 64-bit Edition | 9.22 | 9.2 TS2M3 | ||||
| Microsoft® Windows® for x64 | 9.22 | 9.2 TS2M3 | ||||
| Microsoft Windows Server 2003 Datacenter Edition | 9.22 | 9.2 TS2M3 | ||||
| Microsoft Windows Server 2003 Enterprise Edition | 9.22 | 9.2 TS2M3 | ||||
| Microsoft Windows Server 2003 Standard Edition | 9.22 | 9.2 TS2M3 | ||||
| Microsoft Windows Server 2003 for x64 | 9.22 | 9.2 TS2M3 | ||||
| Microsoft Windows Server 2008 | 9.22 | 9.2 TS2M3 | ||||
| Microsoft Windows Server 2008 for x64 | 9.22 | 9.2 TS2M3 | ||||
| Microsoft Windows XP Professional | 9.22 | 9.2 TS2M3 | ||||
| Windows 7 Enterprise 32 bit | 9.22 | 9.2 TS2M3 | ||||
| Windows 7 Enterprise x64 | 9.22 | 9.2 TS2M3 | ||||
| Windows 7 Home Premium 32 bit | 9.22 | 9.2 TS2M3 | ||||
| Windows 7 Home Premium x64 | 9.22 | 9.2 TS2M3 | ||||
| Windows 7 Professional 32 bit | 9.22 | 9.2 TS2M3 | ||||
| Windows 7 Professional x64 | 9.22 | 9.2 TS2M3 | ||||
| Windows 7 Ultimate 32 bit | 9.22 | 9.2 TS2M3 | ||||
| Windows 7 Ultimate x64 | 9.22 | 9.2 TS2M3 | ||||
| Windows Vista | 9.22 | 9.2 TS2M3 | ||||
| Windows Vista for x64 | 9.22 | 9.2 TS2M3 | ||||
| 64-bit Enabled AIX | 9.22 | 9.2 TS2M3 | ||||
| 64-bit Enabled HP-UX | 9.22 | 9.2 TS2M3 | ||||
| 64-bit Enabled Solaris | 9.22 | 9.2 TS2M3 | ||||
| HP-UX IPF | 9.22 | 9.2 TS2M3 | ||||
| Linux | 9.22 | 9.2 TS2M3 | ||||
| Linux for x64 | 9.22 | 9.2 TS2M3 | ||||
| OpenVMS on HP Integrity | 9.22 | 9.2 TS2M3 | ||||
| Solaris for x64 | 9.22 | 9.2 TS2M3 | ||||
A fix for this issue for SAS/STAT 9.3_M1 is available at:
https://tshf.sas.com/techsup/download/hotfix/HF2/H03.html#45223A fix for this issue for SAS/STAT 9.22 is available at:
http://ftp.sas.com/techsup/download/hotfix/HF2/B80.html#45223| Type: | Problem Note |
| Priority: | high |
| Topic: | Analytics ==> Mixed Models SAS Reference ==> Procedures ==> GLIMMIX |
| Date Modified: | 2012-01-04 14:18:53 |
| Date Created: | 2011-12-23 10:51:59 |