Estimates of Uncertainty in Generalized Linear Regression Models with Spatially Misaligned Data

• Activity Number: 117
• Type: Topic Contributed
• Date/Time: Monday, August 1, 2011 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Kenneth Kyle Lopiano*+ and Linda J. Young and Carol A. Gotway
• Companies: University of Florida and University of Florida and Centers for Disease Control and Prevention
• Keywords: Spatial Misalignment ; Berkson Error ; Kriging ; Generalized Linear Models

Abstract

Researchers in the fields of climate change, environmental risk assessment, and public health often augment their data collection with existing data or work entirely with existing data from multiple sources. When the datasets have a spatial component, the datasets are often spatially misaligned. Spatial misalignment occurs when two or more variables are observed at different locations or aggregated over different geographical units. When the datasets are spatially misaligned, the data must be combined using a common set of geographical units before relationships can be assessed. When smoothing techniques, such as kriging, are used to align the disparate datasets, Berkson-type measurement error is induced. As a result, although regression estimates are unbiased, estimates of their uncertainty are biased. In this work, an iteratively reweighted generalized least squares approach is proposed that produces unbiased estimates of the regression parameter and its standard error when kriging is used to align datasets in point-to-point and point-to-areal misalignment problems. The statistical properties of the approach are presented and simulation studies are conducted.