CensRegMod (version 1.0)

em.cens: Fits Univariate Censored Linear Regression Model With Normal or Student-t Errors

Description

Returns EM estimates for right censored regression model (under Normal or Student-t distribution) and calculates some diagnostic measures for detecting influential observations

Usage

em.cens(cc, x, y, nu="NULL", dist="Normal", diagnostic="FALSE", typediag="NULL")

Arguments

cc

Vector of censoring indicators. For each observation: 0 if non-censored, 1 if censored

x

Design matrix

y

Vector with the responde variable

nu

Initial value for the degree of freedon in case of Student-t model (greater than 2)

dist

Distribution to be used for the errors. "Normal", for normal or "T" for Student-t

diagnostic

TRUE or FALSE. Indicates if any diagnostic measure should or not be computed

typediag

If diagnostic=TRUE, typediag indicates which diagnostic measure should be computed. If typediag=1, computes generalized Cook distance (GD) and its decomposition into the generalized Cook distance for the parameter subsets: betas (GDbeta) and sigma2 (GDsigma2). For local influence with case-weight perturbation, set typediag=2. For local influence with scale perturbation, set typediag=3

Value

beta

EM estimatives for regression coefficients

sigma2

EM estimative for the error variance

nu

EM estimative for degree of freedom. Only returned when type="T"

logver

Value of the log-likelihood under the fitted model

SE

Standard error for EM estimators

measure

Vector with the diagnostic measure chosen in typediag. Only returned when diagnostic=TRUE

AIC

AIC model selection criteria

BIC

BIC model selection criteria

EDC

EDC model selection criteria

Details

Despite of this function has been built to deal with right censored response variables, one can easily adapt it to work with left censored responses: set -y and -x to obtain the left censored model fit and any diagnostic measure for it. The specification of the initial value for nu must be made carefully: if the data have many outliers, then you must choose a small value for nu (greater but near to 2), otherwise you can choose a greater value

References

Monique B. Massuia, Celso R. Cabral, Larissa A. Matos, Victor H. Lachos. "Influence Diagnostics for Student-t Censored Linear Regression Models"