geelm.control

Settings for controlling technical details of GEE fitting via geelm.

Estimation of generalized linear models with
correlated/clustered observations by use of generalized estimating
equations (GEE). See e.g. Halekoh and Højsgaard, (2005,
<doi:10.18637/jss.v015.i02>), for details. Several types of
clustering are supported, including exchangeable variance
structures, AR1 structures, M-dependent, user-specified variance
structures and more. The model fitting computations are performed
using modified code from the 'geeM' package, while the interface
and output objects have been written to resemble the 'geepack'
package. The package also contains additional tools for working
with and inspecting results from the 'geepack' package, e.g. a
'confint' method for 'geeglm' objects from 'geepack'.

Søren Højsgaard

geeasy

Solve Generalized Estimating Equations for Clustered Data

Anne Helby Petersen

Lee McDaniel

Claus Ekstrøm

geelm.control function

<dl><dt>init.beta</dt>
<dd>an optional vector with the initial values of beta. If not specified, then the intercept will be set to <code>InvLink(mean(response))</code>. <code>init.beta</code> must be specified if not using an intercept.</dd>
<dt>init.phi</dt>
<dd>an optional initial overdispersion parameter. If not supplied, initialized to 1.</dd>
<dt>tol</dt>
<dd>Tolerance for asserting convergence.</dd>
<dt>maxit</dt>
<dd>integer giving the maximal number of Fisher Scoring
iteration.</dd>
<dt>scale.fix</dt>
<dd>logical indicating if the scale should be fixed.</dd>
<dt>useP</dt>
<dd>If set to <code>FALSE</code>, do not use the n-p correction for
dispersion and correlation estimates. This can be
useful when the number of observations is small, as subtracting p may yield
correlations greater than 1.</dd>
<dt>std.err</dt>
<dd>Character string specifying which standard error
estimation method should be used. Supported options are
<code>san.se</code> (sandwich SE) and <code>naive</code>.</dd></dl>

Arguments

Control estimation of GEE models — geelm.control

<dl>

<dt>init.beta</dt>
<dd>an optional vector with the initial values of beta. If not specified, then the intercept will be set to <code>InvLink(mean(response))</code>. <code>init.beta</code> must be specified if not using an intercept.</dd>


<dt>init.phi</dt>
<dd>an optional initial overdispersion parameter. If not supplied, initialized to 1.</dd>


<dt>tol</dt>
<dd>Tolerance for asserting convergence.</dd>


<dt>maxit</dt>
<dd>integer giving the maximal number of Fisher Scoring
iteration.</dd>


<dt>scale.fix</dt>
<dd>logical indicating if the scale should be fixed.</dd>


<dt>useP</dt>
<dd>If set to <code>FALSE</code>, do not use the n-p correction for
dispersion and correlation estimates. This can be
useful when the number of observations is small, as subtracting p may yield
correlations greater than 1.</dd>


<dt>std.err</dt>
<dd>Character string specifying which standard error
estimation method should be used. Supported options are
<code>san.se</code> (sandwich SE) and <code>naive</code>.</dd>

</dl>

geelm.control: Control estimation of GEE models

Description

Usage

Value

Arguments