The estimating equations used for GEE models
have a similar structure to those score equations
but are generalized to accommodate multiple
responses from the same subject.
T F 10. If the correlation betweenXandYis zero, then
XandYare independent.Test True or False (Circle T or F)
T F 1. It is typically the regression coefficients, not the
correlation parameters, that are the parameters
of primary interest in a correlated analysis.
T F 2. If an exchangeable correlation structure is spe-
cified in a GEE model, then the correlation
between a subject’s first and second responses
is assumed equal to the correlation between the
subject’s first and third responses. However,
that correlation can be different for each sub-
ject.
T F 3. If a dichotomous response, coded Y¼0 and
Y¼1, follows a binomial distribution, then
the mean response is the probability thatY¼1.
T F 4. In a GLM, the mean response is modeled as
linear with respect to the regression para-
meters.
T F 5. In a GLM, a function of the mean response is
modeled as linear with respect to the regression
parameters. That function is called the link
function.
T F 6. To run a GEE model, the user specifies a work-
ing correlation structure which provides a
framework for the estimation of the correlation
parameters.
T F 7. The decision as to whether to use model-based
variance estimators or empirical variance esti-
mators can affect both the estimation of
the regression parameters and their standard
errors.
T F 8. If a consistent estimator is used for a model,
then the estimate should be correct even if
the number of clusters is small.
T F 9. The empirical variance estimator allows for
consistent estimation of the variance of the
response variable even if the correlation struc-
ture is misspecified.
T F 10. Quasi-likelihood estimates may be obtained
even if the distribution of the response variable
is unknown. What should be specified is a func-
tion relating the variance to the mean response.Answers to Practice Exercises 537