geeDesign
that represents the design of a generalized estimation equation model.
geeDesign(formula, cluster, variance, correlation="independent", random=NULL, ...)
cluster
argument for the
gee
function.
varDesign
object to specify a variance structure.
For constant variance, variance with weights, heterogeneous
within-cluster variances or overdispersion, enter a
varDesign
object.
See the
variance
argument
for the
gee
function for
simple and common options, and
the
varDesign
function for more
complicated options.
corDesign
object
to specify correlation structures.
Character string or list options are
the same as those described in the
correlation
argument for the
gee
function
for a single-layer correlation structure.
To specify a fixed correlation design or
multi-layer correlation structures in nested designs or unbalanced
block designs, enter a
corDesign
object.
For details see the
corDesign
help file.
ranDesign
specifying
the random effects component of a mixed model.
A mixed model is only permited for the
binomial
family
with the
logit
and
probit
links,
the
poisson
family with the
log
link, and the
gaussian
family with
the
identity
link.
See the documentation for
ranDesign
for details on specifying the random component of a mixed model.
gee
function.
For details see the
gee
help file.
"geeDesign"
is returned.
See
geeDesign.object
for details.
glm
function.
This
geeDesign
is an extension of
gee
function to allow
experienced users to specify more complicated GEE models, including
heterogeneous variance and correlation and mixed models.
All arguments allowed in
gee
are supported in
geeDesign
as well.
In order to use
geeDesign
, users
must to be familiar with
gee
,
varDesign
and
corDesign
.
The output of a
geeDesign
is intended mainly
to be used in a call to
gee.fit
.
Seizure.Subject <- recordDesign("Subject",data.frame(Seizure, offset=rep(log(c(8,2,2,2,2)),59))) gee.out <- geeDesign(y~group+offset(offset), cluster=cbind(clusterID,recordID),variance="glm.scale", data=Seizure.Subject, correlation="exchangeable", family=poisson,link=log,control=geeControl(trace = T), subset=Subject!=49,contrasts=list(group=contr.treatment)) ## Review the design before calling the fitting algorithm. print(gee.out) gee.fit(gee.out) Seizure1.Subject <- data.frame(Seizure.Subject,post=rep(c(0,1,1,1,1),59)) gee.out <- geeDesign(y~-1+group*post+offset(offset), cluster=cbind(clusterID,recordID),variance="glm.scale", family="poisson", link="log",data=Seizure1.Subject, correlation=list(type="AR",x.layer="Time"), subset=Subject!=49) gee.fit(gee.out) SpruceGrpd.Subject <- recordDesign("Subject",na.omit(SpruceGrpd)) gee.out <- geeDesign(y~Time+group, cluster=cbind(clusterID,recordID), variance=0.02, family=Gamma, link="power(1.5)", correlation=list(type="contAR", x.layer="Time"), data=SpruceGrpd.Subject,contrasts=list(group=contr.treatment)) gee.fit(gee.out)