These objects are generated by function geeDesign.
The class
"geeDesign"
has
print
and
summary
methods.
Object of class
"geeDesign"
is a list with
three components:
"geeModel"
used in the
algorithm iterations. It contains the following components.
"family"
:
a character string indicating the family.
"link"
:
a character string indicating the link used in the family.
"x"
: numeric, the design matrix.
"y"
: numeric, the response vector.
"z"
: numeric, the random effect design vector.
"id"
: an integer vector of cluster ids.
"offset"
:
numeric vector for offset.
"n.trials"
:
an integer vector indicates the sample size of each observation in binomial
experiments. Only used for binomial family, otherwise use a vector of 1s.
"regressionParInit"
:
numeric, a vector of initial estimates of regression parameters.
"varStruct"
:
an objact of class
"geeVarStruct"
with five
components:
(1)
"type"
: a character string
indicating a variance type. When random effect is specified,
"type"
is specified as "mixedQuad";
(2)
"info"
:
an integer vector listing the information of variance design.
It contains an indicator for the number of additive terms, the number of
unknown scale parameters to be estimated, the number of unknow varaince
of the random effect varaibles, number of weights used in the variance,
and number of weights used in the variance of random effect variables;
(3)
"scale"
: a numeric vector containing
unknown scales and fixed scales for variance together with the variance
of random effect variables.
(4)
"data"
: a numeric vector
containing the weights, and the square of covariates of the random effect
variables .
"corStruct"
:
an object with eleven components
(1)
"type"
:
a vector of character strings, one per layer, containing the correlation
type for the given layer. This includes
"fixed"
for fixed design;
(2)
"nLayer"
:
an integer specifying the number of layers in correlation design;
(3)
"nParLayer"
:
a vector of integers indicating
the number of parameters associated with each layer of correlation structure;
(4)
"parLayerID"
: an integers vector of
length of the number of parameters used in the correlation design; ;
(5)
"parInit"
:
a numeric vector of initial values of correlation parameters.
This is only used if the estimation.flag for alpha is set to 0;
(6)
"parMap"
:
a matrix of integers indicating the correlation parameterization.
Each matrix entry is either -1, the diagonal, -9 ,not used, or
a positive integer, the parameter id for that entry;
(7)
"layerMap"
:
a matrix of integers indicating the correlation design.
The value in each matrix cell, if positive, is the layer id
associated with that cell. Otherwise no layer is associated with that cell;
(8)
"fixed"
:
a null matrix or a correlation matrix used in fixed correlation design or used
as a fixed correlation value for negative corrLayerMap entries;
(9)
"X"
:
a list of two components,
"names"
and
"map"
. The
"names"
contains a vector of character strings listing the names of variables in
"data"
.
The
"map"
is itself a list of length n.layer,
and each component, a vector
of positive integers, maps a layer.id to variables in
"data"
to indicate the
usage of variables in the structure of the layer;
(10)
"data"
: a list with components of
"record.names"
and other variables as a
column referring to time for "AR" or "ARcont", or for
resolving positions within a cluster for each block of an "unstruct" layer.
(11)
"flags"
: a list with components of
estInitAlpha"
indicating whether to estimate the
scale and initialize correlation parameters, respectively
(1=estimate values, 0=do not estimate values).
balanced"
indicating a balanced design or
longitudinal design respectively. (1=True,0=False), and
"cor.once"
.
A longitudinal design is such that the correlation
matrix can be calculated once, and subset as needed for each cluster.
"ksStruct"
:
"control"
:
A list of control parameters: algorithm, tolerance.reg, tolerance.cor,
maxit, trace. When trace=T, information in the iteration process is printed.
The object is mainly used to specify
the argument
design
in the gee.fit function.