aucm (version 2016.1-2)

rlogit: rlogit

Description

Robust logistic regression estimator of Bianco and Yohai

Usage

rlogit (formula, dat, const=0.5, kmax=1e3, maxhalf=10, verbose=FALSE)
"coef"(object,...)
"trainauc"(fit, training.data=NULL, ...)
"predict"(object, newdata, ...)
"ratio"(fit)
logistic.f(eta,h,loss=TRUE)

Arguments

formula

a formula specifying the model to be fit.

dat

a data frame containing the outcome and covariates in the model

const

tuning constant used in the computation of the estimator, defaults to 0.5

kmax

maximum number of iterations before convergence, defaults to 1000

maxhalf

max number of step-halving ,defaults to 10

verbose

logical

object

an object of class 'rlogit'

fit

an object that inherits from class 'auc' such as 'rauc' or 'sauc'

newdata

data at which to predict

training.data

data frame used to compute auc based on a fit obtained by a call to rauc, sauc or sauc.dca

eta,h

logistic.f computes for loss = FALSE expit(eta/h) or expit(-eta/h) for loss = TRUE

loss

a boolean. if TRUE (default) logistic loss is assumed.

...

arguments passed to or from methods

Value

A list with the follwoing components:

convergence

logical, was convergence achieved

objective

value of the objective function at the minimum

coef

estimates for the parameters

sterror

standard errors of the parameters (if convergence is TRUE)

Details

This program computes the estimator of Bianco and Yohai (1996) in logistic regression. By default, an intercept term is included and p parameters are estimated. The outcome is coded as a 0/1 binomial variable.

If initwml == TRUE, a weighted ML estimator is computed with weights derived from the MCD estimator computed on the explanatory variables. If initwml == FALSE, a classical ML fit is perfomed. When the explanatory variables contain binary observations, it is recommended to set initwml to FALSE or to modify the code of the algorithm to compute the weights only on the continuous variables.

References

Implementing the Bianco and Yohai estimator for Logistic Regression Croux, C., and Haesbroeck, G. (2003) Computational Statistics and Data Analysis, 44, 273-295

Examples

set.seed(1)
x0 <- matrix(rnorm(100,1))
y  <- as.numeric(runif(100)>0.5)        # numeric(runif(100)>0.5)
dat=data.frame(y=y, x=x0)
rlogit(y~x, dat)