Abstract:The conditional autoregressive (CAR) model, simultaneous autoregressive (SAR) model, and its variants have become the predominant strategies for modeling regional or areal-referenced spatial data. The overwhelming wide-use of the CAR/SAR model motivates the need for new classes of models for areal-referenced data. Thus, we develop a novel class of Markov random fields based on truncating the full-conditional distribution. We define this truncation in two ways leading to versions of what we call the truncated autoregressive (TAR) model. First, we truncate the full conditional distribution so that a response at one location is close to the average of its neighbors. This strategy establishes relationships between TAR and CAR. Second, we truncate on the joint distribution of the data process in a similar way. This specification leads to connection between TAR and SAR model. Our Bayesian implementation does not use Markov chain Monte Carlo (MCMC) for Bayesian computation, and generates samples directly from the posterior distribution. Moreover, TAR does not have a range parameter that arises in the CAR/SAR models, which can be difficult to learn. We present the results of the proposed truncated autoregressive model on several simulated datasets and on a dataset of average property prices.

What problem does this paper attempt to address?

The main problems that this paper attempts to solve are the limitations of existing Conditional Autoregressive (CAR) and Simultaneous Autoregressive (SAR) models when dealing with regional reference spatial data. Specifically: 1. **Range Parameter Problem**: CAR and SAR models need to introduce a range parameter to ensure the positive definiteness of the spatial covariance matrix, but this parameter is difficult to interpret spatially and hard to learn. 2. **Flexibility of Dependence Structure**: Existing CAR and SAR models have limited flexibility in modeling dependence relationships and usually require a known adjacency matrix, which limits their application scope. 3. **Computational Complexity**: Traditional models such as CAR and SAR usually rely on the Markov Chain Monte Carlo (MCMC) method when performing Bayesian inference, which increases computational complexity and the difficulty of parameter tuning. To solve these problems, the author proposes a new class of models based on truncated Markov Random Fields (MRFs) - the Truncated Autoregressive (TAR) model. The TAR model is defined by truncating the full - conditional distribution or the joint distribution, avoiding the need to introduce a range parameter and not requiring the use of MCMC for Bayesian calculations, and can directly generate samples from the posterior distribution. In addition, the TAR model adds a "precision noise" term to the precision matrix, thereby ensuring the strict positive definiteness of the matrix and making the model more stable and easier to implement. Specifically, the paper proposes two versions of the TAR model: - **Conditional Truncated Autoregressive Model (TAR C)**: By truncating the full - conditional distribution at each location, the response value is made close to the average of its neighbors. - **Simultaneous Truncated Autoregressive Model (TAR S)**: By truncating the joint distribution of all locations in a similar way. These two models not only avoid the problem of the range parameter but also show an interesting connection with the classical CAR and SAR models and provide a more flexible modeling framework. The effectiveness and superiority of the TAR model are verified through experimental results on simulated data sets and actual data sets (such as Glasgow housing price data).

Markov Random Fields with Proximity Constraints for Spatial Data

Markov chain random fields, spatial Bayesian networks, and optimal neighborhoods for simulation of categorical fields

Conditional autoregressive models fused with random forests to improve small-area spatial prediction

A bayesian hierarchical spatial model for dental caries assessment using non-gaussian markov random fields.

Inferring the contiguity matrix for spatial autoregressive analysis with applications to house price prediction

A Semiparametric Bayesian Approach to Heterogeneous Spatial Autoregressive Models

Bayesian estimation for heterogeneous spatial autoregressive models with variance modelling

Spatial Gaussian Markov Random Fields: Modelling, Applications and Efficient Computations

A Constrained Spatial Autoregressive Model for Interval-valued data

A multivariate Gaussian random field prior against spatial confounding

Renal involvement in children with lymphoma: comparison of CT with sonography.

Spatial Autoregression And Related Spatio-Temporal Models

Bayesian Hierarchical Spatial Regression Models for Spatial Data in the Presence of Missing Covariates with Applications

On the relationship between conditional (CAR) and simultaneous (SAR) autoregressive models

Privacy-Protected Spatial Autoregressive Model

Nonstationary Spatial Gaussian Markov Random Fields

A random forests-based hedonic price model accounting for spatial autocorrelation

Highly Multivariate High-dimensionality Spatial Stochastic Processes-A Mixed Conditional Approach

Spatial extension of generalized autoregressive conditional heteroskedasticity models

Partially Specified Spatial Autoregressive Model with Artificial Neural Network

Dynamic Spatial Autoregressive Models with Autoregressive and Heteroskedastic Disturbances