On a tobit–Birnbaum–Saunders model with an application to medical data

Mário F. Desousa, Helton Saulo, VICTOR ELISEO LEIVA SANCHEZ, Paulo Scalco

Research output: Contribution to journalArticlepeer-review

18 Scopus citations

Abstract

The tobit model allows a censored response variable to be described by covariates. Its applications cover different areas such as economics, engineering, environment and medicine. A strong assumption of the standard tobit model is that its errors follow a normal distribution. However, not all applications are well modeled by this distribution. Some efforts have relaxed the normality assumption by considering more flexible distributions. Nevertheless, the presence of asymmetry could not be well described by these flexible distributions. A real-world data application of measles vaccine in Haiti is explored, which confirms this asymmetry. We propose a tobit model with errors following a Birnbaum–Saunders (BS) distribution, which is asymmetrical and has shown to be a good alternative for describing medical data. Inference based on the maximum likelihood method and a type of residual are derived for the tobit–BS model. We perform global and local influence diagnostics to assess the sensitivity of the maximum likelihood estimators to atypical cases. A Monte Carlo simulation study is carried out to empirically evaluate the performance of these estimators. We conduct a data analysis for the mentioned application of measles vaccine based on the proposed model with the help of the R software. The results show the good performance of the tobit–BS model.

Original languageEnglish
Pages (from-to)932-955
Number of pages24
JournalJournal of Applied Statistics
Volume45
Issue number5
DOIs
StatePublished - 4 Apr 2018

Keywords

  • Birnbaum–Saunders distribution
  • diagnostics
  • likelihood-based methods
  • R software
  • residuals
  • tobit models

Fingerprint Dive into the research topics of 'On a tobit–Birnbaum–Saunders model with an application to medical data'. Together they form a unique fingerprint.

Cite this