A comparison of errors in variables methods for use in regression models with spatially misaligned data

Area of Science:

• Spatial statistics
• Geostatistics
• Statistical modeling

Background:

• Spatial data often involves variables measured at different locations or scales.
• This spatial misalignment complicates inference on variable associations.

Purpose of the Study:

• To compare methods for calculating accurate standard errors in regression when predictor variables are spatially misaligned with response variables.
• To evaluate these methods in both point-referenced and change-of-support settings.

Main Methods:

• Two simulation studies were conducted to compare standard error estimation methods.
• Methods were extended and tested for the change-of-support problem (point-to-area data).

Main Results:

• Regression parameter estimates are unbiased when using predicted predictor values (e.g., via kriging).
• Naive standard errors for these parameters are underestimated, leading to potentially incorrect statistical inference.

Conclusions:

• Accurate standard error estimation is crucial for valid statistical inference in spatially misaligned data.
• The study provides a comparison of methods for addressing this challenge in geostatistics.