Model 5:Standard logistic regres-
sion (naive model)
Variable Coefficient
Model-based
Std ErrWald
p-valueINTERCEPT 1.4362 0.6022 0.0171
BIRTHWGT 0.0005 0.0002 0.0051
GENDER 0.0453 0.2757 0.8694
DIARRHEA 0.7764 0.4538 0.0871
^b 3 for DIARRHEA samebuts^
band
WaldP-values differ.
Model 4 vs. Model 5
Parameter estimates same
s^bModel 4 >s^bModel 5
Other data: possible that
s^b(empirical)<s^b(model based)
Summary. Comparison of model
results for DIARRHEA
Correlation
structure
Odds
ratio 95% CI
1 AR(1) 1.25 (0.23, 6.68)
2 Exchangeable 1.91 (0.44, 8.37)
3 Fixed (user defined) 1.29 (0.26, 6.46)
4 Independent 2.17 (0.69, 6.85)
5 Independent (SLR) 2.17 (0.89, 5.29)
In particular, the coefficient estimate for
DIARRHEA is 0.7764 in both Model 4 and
Model 5; however, the standard error for DIAR-
RHEA is larger in Model 4 at 0.5857 compared
with 0.4538 for Model 5. Consequently, theP-
values for the Wald test also differ: 0.1849 for
Model 4 and 0.0871 for Model 5.The other parameters in both models exhibit
the same pattern, in that the coefficient esti-
mates are the same, but the standard errors are
larger for Model 4. In this example, the empiri-
cal standard errors are larger than their model-
based counterparts, but this does not always
occur. With other data, the reverse can occur.A summary of the results for each model for
the variable DIARRHEA is presented on the
left. Note that the choice of correlation struc-
ture affects both the odds ratio estimates and
the standard errors, which in turn affects the
width of the confidence intervals. The largest
odds ratio estimates are 2.17 from Model 4 and
Model 5, which use an independent correlation
structure. The 95% confidence intervals for all
of the models are quite wide, with the tightest
confidence interval (0.89, 5.29) occurring in
Model 5, which is a standard logistic regres-
sion. The confidence intervals for the odds
ratio for DIARRHEA include the null value of
1.0 for all five models.Presentation: II. Example 1: Infant Care Study 549