A formal goodness-of-fit test for spatial binary Markov random field models

Area of Science:

• Spatial statistics
• Environmental science
• Ecological modeling

Background:

• Markov random field (MRF) models are widely used for spatial binary data in environmental and ecological research.
• Assessing the fit of MRF models, especially their neighborhood specifications, is challenging for binary data.
• Existing diagnostic tools for MRF models are insufficient for practical applications.

Purpose of the Study:

• To develop a formal goodness-of-fit (GOF) test for diagnosing MRF models applied to spatial binary data.
• To address the specific challenge of assessing neighborhood structures within these models.
• To provide a reliable method for validating MRF model assumptions in spatial analyses.

Main Methods:

• Proposed a novel goodness-of-fit (GOF) test for spatial binary Markov random field models.
• The test statistic is based on a conditional Moran's I, utilizing fitted conditional probabilities.
• The method is designed to detect deviations in model form, including neighborhood misspecification.

Main Results:

• Numerical studies demonstrated the GOF test's effectiveness in detecting departures from null models.
• The test showed particular strength in identifying issues related to neighborhood specifications.
• The proposed test provides a practical solution for diagnosing MRF models for spatial binary data.

Conclusions:

• The developed GOF test is a valuable tool for validating MRF models in spatial binary data analysis.
• It offers improved diagnostic capabilities, especially for complex neighborhood structures.
• The test has practical implications for environmental and ecological modeling, enhancing the reliability of spatial analyses.