|
V6-GENMOD-D998
GEE standard errors inflated when single-observation clusters exist
|
When you use the REPEATED statement and some clusters contain only one
observation (e.g. only one observation per subject), then the
correlation matrix is underestimated. The GEE parameter estimates are
less statistically efficient resulting in larger standard errors and
more conservative tests of these parameters. While some efficiency is
lost, the parameter estimates and their standard errors are not
incorrect. Clusters are defined by the SUBJECT= effect on the REPEATED
statement. The GENMOD output displays the Minimum Cluster Size in the
GEE Model Information table. The severity of the underestimation
increases with the correlation and the proportion of clusters having
only one observation.
| Products: | STAT |
| Component: | GENMOD |
| Priority: | HIGH |
| Status: | Outstanding Problem |
| Date: | Tue, 27 Oct 1998 |
| System | Release Reported | Release Fixed |
| AIX/6000 | 6.12 TS020 | 6.12 TS055 |
| HP-UX Operating Systems | 6.12 TS020 | 6.12 TS055 |
| IBM OS/2 | 6.12 TS020 | 6.12 TS055 |
| Solaris | 6.12 TS020 | 6.12 TS055 |
| OpenVMS Alpha | 6.12 TS020 | 6.12 TS055 |
| Windows 3.11 | 6.12 TS020 | 6.12 TS055 |
| Windows NT | 6.12 TS020 | 6.12 TS055 |
| Windows 95 | 6.12 TS020 | 6.12 TS055 |
| Compaq Tru64 Unix | 6.12 TS040 | |
| Macintosh | 6.12 TS040 |
No Fixes Available