logistf (version 1.10)

logistpl.control: Control parameters for logistf

Description

Sets parameters for Newton-Raphson iteration in Firth's penalized-likelihood logistic regression

Usage

logistpl.control(maxit=100, maxhs=5, maxstep=5, lconv=0.00001, xconv=0.00001, ortho=FALSE, pr=FALSE)

Arguments

maxit

the maximum number of iterations

maxhs

the maximum number of step-halvings in one iteration. The increment of the beta vector within one iteration is divided by 2 if the new beta leads to a lower log likelihood than the target value.

maxstep

specifies the maximum step size in the beta vector within one iteration.

lconv

specifies the convergence criterion for the log likelihood. For convergence, the loglikelihood may not be greater than target value + lconv

xconv

specifies the convergence criterion for the parameter estimates. For convergence, the change in any parameter estimate may not be greater than xconv

ortho

orthogonalizes the variable for which confidence intervals should be computed with respect to the other variables in the model. This may sometimes speed up or enable convergence. Default=FALSE

pr

replaces the matrix of 'other' variables in the model (those for which the current call of logistpl does not compute a confidence limit) by their principal components (of the same rank). This may sometimes speed up or enable conv

Value

• maxit
  the maximum number of iterations
• maxhs
  the maximum number of step-halvings in one iteration. The increment of the beta vector within one iteration is divided by 2 if the new beta leads to a decrease in log likelihood.
• maxstep
  specifies the maximum step size in the beta vector within one iteration.
• lconv
  specifies the convergence criterion for the log likelihood.
• xconv
  specifies the convergence criterion for the parameter estimates.
• ortho
  orthogonalizes the variable for which confidence intervals should be computed with respect to the other variables in the model. This may sometimes speed up or enable convergence. Default=FALSE
• pr
  replaces the matrix of 'other' variables in the model (those for which the current call of logistpl does not compute a confidence limit) by their principal components (of the same rank). This may sometimes speed up or enable convergence. Default=FALSE

Details

logistpl.control() is used by logistf (if pl==TRUE) and logistfplot, which both call the internal function logistpl, to set control parameters to default values. Different values can be specified, e. g., by logistf(...,plcontrol=logistpl.control(maxstep=1)).

Examples

data(sexagg)
fit1<-logistf(case ~ age+oc+vic+vicl+vis+dia, data=sexagg, weights=COUNT) # use default
fit1$pl.iter
fit2<-logistf(case ~ age+oc+vic+vicl+vis+dia, data=sexagg, weights=COUNT, plcontrol=logistpl.control(maxstep=1)) #change maxstep
fit2$pl.iter