Variable Selection for the Spatial Autoregressive Model with Autoregressive Disturbances

Along with the rapid development of the geographic information system, high-dimensional spatial heterogeneous data has emerged bringing theoretical and computational challenges to statistical modeling and analysis. As a result, effective dimensionality reduction and spatial effect recognition has be...

Full description

Saved in:
Bibliographic Details
Published inMathematics (Basel) Vol. 9; no. 12; p. 1448
Main Authors Liu, Xuan, Chen, Jianbao
Format Journal Article
LanguageEnglish
Published Basel MDPI AG 01.06.2021
Subjects
Algorithms
Autoregressive models
Data analysis
Dependent variables
Disturbances
Econometrics
Error reduction
Feature selection
Food science
Geographic information systems
Heterogeneity
Iterative algorithms
Iterative methods
Methods
Normality
Parameter estimation
Parameter identification
penalized method
Reptiles & amphibians
SCAD
spatial
Spatial data
Statistical methods
Statistical models
variable selection
Variables
Online AccessGet full text

Cover

More Information
Summary:Along with the rapid development of the geographic information system, high-dimensional spatial heterogeneous data has emerged bringing theoretical and computational challenges to statistical modeling and analysis. As a result, effective dimensionality reduction and spatial effect recognition has become very important. This paper focuses on variable selection in the spatial autoregressive model with autoregressive disturbances (SARAR) which contains a more comprehensive spatial effect. The variable selection procedure is presented by using the so-called penalized quasi-likelihood approach. Under suitable regular conditions, we obtain the rate of convergence and the asymptotic normality of the estimators. The theoretical results ensure that the proposed method can effectively identify spatial effects of dependent variables, find spatial heterogeneity in error terms, reduce the dimension, and estimate unknown parameters simultaneously. Based on step-by-step transformation, a feasible iterative algorithm is developed to realize spatial effect identification, variable selection, and parameter estimation. In the setting of finite samples, Monte Carlo studies and real data analysis demonstrate that the proposed penalized method performs well and is consistent with the theoretical results.
ISSN:2227-7390
2227-7390
DOI:10.3390/math9121448