ii. The correlation is assumed to depend on the
interval of time between responses.
iii. AR1is a special case of the autoregressive
correlation structure:
a. Assumption of AR1: the correlation
between any two responses from the
same subject taken at timet 1 andt 2 is
rjt^1 t^2 j.
b. There is one correlation parameter, but
the order within a cluster is not arbitrary.
D. Stationarym-dependentcorrelation structure
i. Assumption: correlationskoccasions apart
are the same fork¼1, 2,...,m, whereas
correlations more thanmoccasions apart
are zero.
ii. In a stationarym-dependent structure,
there aremcorrelation parameters.
E. Unstructuredcorrelation structure
i. In general, fornresponses in a cluster, there
aren(n1)/2 correlation parameters.
ii. Yields a separate correlation parameter for
each pair (j, k, j 6 ¼k) of observations within a
cluster.
iii. The order of responses is not arbitrary.
F. Fixedcorrelation structure
i. The user specifies the values for the
correlation parameters.
ii. No correlation parameters are estimated.
IX. Empirical and model-based variance estimators
(pages 516–519)
A. If a GEE model is correctly specified (i.e., the
correct link function and correlation structure
are specified), the parameter estimates are
consistent and the distribution of the estimates
is asymptotically normal.
B. Even if the correlation structure is misspecified,
the parameter estimatesð^bÞare consistent.
C. Two types of variance estimators can be
obtained in GEE:
i. Model-based variance estimators.
a. Make use of the specified correlation
structure.
b. Are consistent only if the correlation
structure is correctly specified.532 14. Logistic Regression for Correlated Data: GEE