Spatial Censored Regression Models in R: The CensSpatial package

TL;DR

CensSpatial is an R package that facilitates analysis of spatial censored data using multiple estimation algorithms, diagnostics, and predictions, demonstrated through environmental data examples.

Contribution

The paper introduces the CensSpatial package, offering new tools for efficient estimation, prediction, and diagnostics of spatial censored data in R.

Findings

SAEM algorithm enables fast parameter estimation

Diagnostic tools utilize Hessian matrix analysis

Package demonstrated with environmental data examples

Abstract

CensSpatial is an R package for analyzing spatial censored data through linear models. It offers a set of tools for simulating, estimating, making predictions, and performing local influence diagnostics for outlier detection. The package provides four algorithms for estimation and prediction. One of them is based on the stochastic approximation of the EM (SAEM) algorithm, which allows easy and fast estimation of the parameters of linear spatial models when censoring is present. The package provides worthy measures to perform diagnostic analysis using the Hessian matrix of the completed log-likelihood function. This work is divided into two parts. The first part discusses and illustrates the utilities that the package offers for estimating and predicting spatial censored data. The second one describes the valuable tools to perform diagnostic analysis. Several examples in spatial environmental data are also provided.