crq.fit {crq}R Documentation

Fitting of censored quantile regression models

Description

Function controlling the fitting of censored quantile regression models

Usage

 crq.fit(x, y, cen, weights, method, mw = 20 , ginit = 0.001, gstep = NULL) 

Arguments

x the design matrix
y the response vector typically in log(event time) (AFT) form
cen censoring indicator: usually 0 for censored, 1 for uncensored. No provision for interval censoring (yet).
weights weights to be used for fitting see crq.
method method to be used: either "grid" or "pivot".
mw Maximal number of allowed calls to weighted rq (to handle possible degeneracy). Only relevant for method = "pivot".
gstep spacing of points of evaluation only relevant for method = "grid", defaults to min(0.01,1/(2*length(y)^{.7})).
ginit initial point of evaluation only relevant for method = "grid", defaults to 0.001.

Details

See {crq}

Value

sol A matrix with (p+2) rows: the first row is the taus (in (0,1)) corresponding to the fitted model parameters. The next p rows are the fitted parameters betahat (tau), and the final row contains entries predicting the conditional quantiles of the response at the mean of the design: xbar ' betahat (tau).
Isplit Indices of split censored observations.
tausplit Corresponding tau values.
status Status of censored observations: 0: not censored 1: split (crossed) censored 2: deleted censored as below tau = 0 solution 3: above last (maximal tau) solution

Author(s)

Stephen Portnoy & Tereza Neocleous

See Also

crq


[Package crq version 0.3 Index]