Maximum Likelihood

Spatial Self-Confounding: Smoothness-related estimation bias in spatial regression models

1 min read · Thu, Oct 30 2025

Spatial regression Maximum Likelihood Estimation Gaussian random fields spatial statistics

Spatial regression models are widely used to capture the relationship between observations and covariates, employing Gaussian random fields to account for spatial variability not explained by the covariates. A new study by researchers David Bolin and Jonas Wallin addresses a critical yet often overlooked problem in these models: smoothness-related spatial self-confounding. The work examines how misspecified covariates, particularly when there are differences in smoothness between variables, can lead to severe and counter-intuitive biases in the estimation of regression parameters. These