Logistic Regression: A Self-learning Text, Third Edition (Statistics in the Health Sciences)

Estimated V Correlation Matrix for IDNO 1

Row Col1 Col2 Col3 Col4 Col5 Col6 Col7 Col8 Col9

1 1.0000 0.5370 0.2884 0.1549 0.08317 0.04467 0.02399 0.01288 0.006918
2 0.5370 1.0000 0.5370 0.2884 0.1549 0.08317 0.04467 0.02399 0.01288
3 0.2884 0.5370 1.0000 0.5370 0.2884 0.1549 0.08317 0.04467 0.02399
4 0.1549 0.2884 0.5370 1.0000 0.5370 0.2884 0.1549 0.08317 0.04467
5 0.08317 0.1549 0.2884 0.5370 1.0000 0.5370 0.2884 0.1549 0.08317
6 0.04467 0.08317 0.1549 0.2884 0.5370 1.0000 0.5370 0.2884 0.1549
7 0.02399 0.04467 0.08317 0.1549 0.2884 0.5370 1.0000 0.5370 0.2884
8 0.01288 0.02399 0.04467 0.08317 0.1549 0.2884 0.5370 1.0000 0.5370
9 0.006918 0.01288 0.02399 0.04467 0.08317 0.1549 0.2884 0.5370 1.0000

The GLIMMIX Procedure

Covariance Parameter Estimates

Cov Parm Subject Estimate Standard Error

AR(1) IDNO 0.5370 0.02514

Residual 0.9709 0.05150

Solutions for Fixed Effects

Effect Estimate Standard Error DF tValue Pr>jtj

Intercept 1.3969 1.1949 133 1.17 0.2445

BIRTHWGT 0.00049 0.000307 133 1.61 0.1095

GENDER 0.004201 0.5549 133 0.01 0.9940

DIARRHEA 0.2112 0.8648 1066 0.24 0.8071

Type III Tests of Fixed Effects

Effect Num DF Den DF FValue Pr>F

BIRTHWGT 1 133 2.60 0.1095

GENDER 1 133 0.00 0.9940

DIARRHEA 1 1066 0.06 0.8071

Estimates

Label Estimate

Standard

Error DF tValue Pr>jtj

Exponentiated

Estimate

log odds ratio

(DIARRHEA 1

vs 0)

0.2112 0.8648 1066 0.24 0.8071 1.2352

The model results from PROC GLIMMIX are close but not exactly the same as was
obtained from PROC GENMOD in the previous section. The odds ratio estimate for
DIARRHEA¼1 vs. DIARRHEA¼0 derived from the ESTIMATE statement is given
as 1.2352 (last row at the end of the output). The odds ratio was estimated at 1.2479
using PROC GENMOD.

The default optimization method for parameter estimation in PROC GLIMMIX is a
residual pseudo-likelihood technique that utilizes a workingcovariancestructure
provided by the user. The AR(1) covariance structure contains two parameters: a

630 Appendix: Computer Programs for Logistic Regression