Dados do Trabalho

Título

CENSORED REGRESSION MODELS WITH AUTOREGRESSIVE ERRORS

Data de titulação

04/11/2016

Instituição de titulação

Universidade Estadual de Campinas

RESUMO (abstract)

Time series data are frequently encountered in diverse fields, including environmental monitoring, medicine, economics and social science, and they are often autocorrelated rather than independent. An additional complication arises when time series measurements are observed with data irregularities, such as observations subjected to upper or lower detection limits, below and above which they are not quantifiable, and missing observations. Practitioners commonly disregard censored data cases or replace these observations with some function of the limit of detection, which often results in biased estimates. In this work, we study some aspects of estimation and local influence analysis in censored regression models with autoregressive errors of order p (hereafter, AR(p)-CR models). The estimates of maximum likelihood (ML) of the parameters are obtained using a stochastic approximation of the EM (SAEM) algorithm. This approach allows for easy and fast estimation of the parameters of autoregressive models when censoring is present. As a byproduct, the SAEM algorithm enables predictions of unobservable values of the response variable. The observed information matrix is derived analytically to account for standard errors. Furthermore, local influence diagnostic measures are derived for the AR(p)-CR model on the basis of the Q-function under three usual perturbation schemes. The finite sample performance of the methods is evaluated through the analysis of several simulation studies and its applications to two real datasets. The proposed algorithm and methods are implemented in the new R package ARCensReg.