Detailed
Outline
I. Overview(pages 492 – 493)
A. Focus: modeling outcomes with dichotomous
correlated responses.
B. Observations can be subgrouped into clusters.
i. Assumption: responses are correlated
within a cluster but independent between
clusters.
ii. An analysis that ignores the within-cluster
correlation may lead to incorrect inferences.
C. Primary analysis method examined is use of
generalized estimating equations (GEE) model.
II. An example (Infant Care Study)(pages 493 – 498)
A. Example is a comparison of GEE to
conventional logistic regression that ignores the
correlation structure.
B. Ignoring the correlation structure can affect
parameter estimates and their standard errors.
C. Interpretation of coefficients (i.e., calculation of
odds ratios and confidence intervals) is the
same as for standard logistic regression.
III. Data layout(page 499)
A. For repeated measures forKsubjects:
i. Theith subject hasnimeasurements
recorded.
ii. Thejth observation from theith subject
occurs at timetijwith the outcome
measured asYijand withpcovariates,
Xij1,Xij2,...,Xijp.
B. Subjects do not have to have the same number
of observations.
C. The time interval between measurements does
not have to be constant.
D. The covariates may be time-independent or
time-dependent for a given subject.
i. Time-dependent variable: values can vary
between time intervals within a cluster;
ii. Time-independent variables: values do not
vary between time intervals within a cluster.
IV. Covariance and correlation(pages 500 – 502)
A. Covariance ofXandY: the expected value of
the product ofXminus its mean andYminus
its mean:
covðX;YÞ¼E½ðXmxÞðYmyÞ:Detailed Outline 529