Addressing uncertainty in census estimates

Data from the American Community Survey (ACS) provides a wealth of information useful for learning more about social determinants of health and their spatial distributions within a defined region. Although available data includes quantified indicators of uncertainty in aggregated location-specific estimates for a range of variables, this uncertainty is often ignored, the consequences of which may include estimate bias and reduced statistical power. Fortunately, the measurement error literature provides a range of useful tools for handling such error. We propose and demonstrate a new application of existing, well-supported measurement error models to spatial regression models. We show that the existing solution of ignoring the measurement error inherent in these data precludes precise effect estimation and that straightforward modifications to traditional estimators can be made to correct for this error. We intend for this work to establish the basic principles of error correction in spatial data and a new method for applying corrected regression estimators to such data.