Output presented:
^bh;s^bh(empirical), and
Wald testP-values “Working” correlation matrix
(Ci) containingr^
Sample:
K¼168 infants,ni9, but
9 infants “exposed cases”:
(i.e.,D¼1 and
DIARRHEA¼1 for any
month)Two sections of the output are presented for
each model. The first contains the parameter
estimate for each coefficient (i.e., beta), its esti-
mated standard error (i.e., the square root of
the estimated variance), and aP-value for the
Wald test. Empirical standard errors rather
than model-based are used for all but the last
model. Recall that empirical variance estima-
tors are consistent estimators even if the corre-
lation structure is incorrectly specified (see
Chap. 14).The second section of output presented for
each model is the working correlation matrix
(Ci). The working correlation matrix contains
the estimates of the correlations, which depend
on the specified correlation structure. The
values of the correlation estimates are often
not of primary interest. However, the examina-
tion of the fitted correlation matrices serves to
illustrate key differences between the underly-
ing assumptions about the correlation struc-
ture for these models.There are 168 clusters (infants) represented in
the data. Only nine infants have a value of 1 for
boththe outcome and diarrhea variables at any
time during their 9 months of measurements.
The analysis, therefore, is strongly influenced
by the small number of infants who are classi-
fied as “exposed cases” during the study period.544 15. GEE Examples