Advancing Methods for Spatial Analysis in Local Modeling

Arizona State University, Scottsdale AZ

Investigators

Abstract

This project develops statistical methods for spatial data. It is common for data to exhibit variation across spatial contexts, including outcomes such as crime rates, voting preferences in elections, and disease prevalence. Conventionally, researchers have used statistical models that implicitly assume that the processes generating these outcomes are uniform across locations. Yet, the processes that lead to this variation may vary in different spatial contexts, and statistical approaches are needed that account for this variation. In this study, the researchers develop statistical methods that address correlations between spatially proximate outcomes. As a complement to the methods that they develop, the researchers are developing open source software to make these approaches freely available to other scholars. In addition to the methodological advances, the award supports the involvement of two graduate students, who benefit from the training in scientific research. The award supports a researcher with a disability, which contributes to goals of broadening participation in science. In this study, the researchers examine and develop methods for the statistical modeling of data at multiple scales. Specifically, the researchers advance multiscale geographically weighted regression (MGWR) methods, which allows the effects of predictor variables to vary based on their spatial context. These approaches address known challenges to statistical inferences, specifically that regression models may indicate biased and contrary effects if the models do not adequately account for the scale and structure in the dataset. In this project, the researchers examine the extent to which these problems can be addressed with the use of MGWR methods. Additional aims include new methods for adjusting inferences when multiple hypotheses are examined and the implementation of diagnostic tools for examining the robustness of model assumptions. To advance the methods, the researchers apply the modeling approach to empirical datasets with spatial structure, such as mortality during the COVID-19 pandemic, teen pregnancy rates, and voting trends in recent elections. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.