logbin (version 1.2)

logbin.control: Auxiliary for Controlling logbin Fitting

Description

Auxiliary function for logbin fitting. Typically only used internally by nplbin, but may be used to construct a control argument to that function.

Usage

logbin.control(bound.tol = 1e-06, epsilon = 1e-08, maxit = 10000, trace = 0)

Arguments

bound.tol

positive tolerance specifying the interior of the parameter space. If the fitted model is more than bound.tol away from the boundary of the parameter space then it is assumed to be in the interior. This can allow the computational method to terminate early if an interior maximum is found. No early termination is attempted if bound.tol = Inf.

epsilon

positive convergence tolerance $\epsilon$; the estimates are considered to have converged when $\sqrt{ \sum (\theta_{old} - \theta_{new})^2} / \sqrt {\sum \theta_{old}^2} < \epsilon$, where $\theta$ is the vector of parameter estimates. See conv.test.

maxit

integer giving the maximum number of EM algorithm iterations for a given parameterisation.

trace

number indicating level of output that should be produced. >= 1 gives output for each parameterisation, >= 2 gives output at each iteration.

Value

A list with components named as the arguments.

Details

This is used similarly to glm.control. The control argument of logbin is by default passed to the control argument of nplbin.

When trace is greater than zero, calls to cat produce the output. Hence, options(digits = *) can be used to increase the precision.

See Also

glm.control, the equivalent function for glm fitting.

nplbin, the function used to fit logbin models.

Examples

## Variation on example(glm.control) :

evts <- c(18,17,15,20,10,20,25,13,12)
obs <- rep(30,9)
outcome <- gl(3,1,9)
treatment <- gl(3,3)
oo <- options(digits = 12)
logbin.D93X <- logbin(cbind(evts,obs-evts) ~ outcome + treatment, trace = 2, epsilon = 1e-2)
options(oo)
coef(logbin.D93X)