Bias correction of bounded location error in binary data.

Area of Science:

• Spatial statistics
• Biostatistics
• Geospatial analysis

Background:

• Spatially referenced binary data are crucial in epidemiology and ecology.
• Location error in recorded observations can lead to unavailable covariate values at true spatial locations.
• Existing regression models may be unreliable when faced with location error.

Purpose of the Study:

• To introduce and evaluate a change of support (COS) method for spatial binary regression models with location error.
• To demonstrate how COS can provide accurate coefficient estimates when true covariate values are unknown.
• To compare the performance of COS against naive approaches that ignore location error.

Main Methods:

• Application of change of support (COS) to binary regression models.
• Accommodation of both spatial and nonspatial covariates.
• Simulation experiments to compare COS with naive methods.
• Illustration using a real-world disease risk data set with observations aggregated within administrative units.

Main Results:

• Naive regression models that ignore location error yield unreliable results.
• The COS method effectively eliminates bias in coefficient estimates.
• COS preserves the interpretability of standard regression models like logistic and probit regression.
• The method demonstrates flexibility in handling spatial and nonspatial covariates.

Conclusions:

• Conventional spatial binary regression models are susceptible to bias from location error.
• Change of support (COS) offers a robust solution for analyzing spatial binary data with location uncertainty.
• COS enables accurate modeling of individual-level risk and maintains model interpretability, making it valuable for ecological and epidemiological studies.