A Bayesian Framework for Accumulating Geographic Knowledge Through Replication

Keywords: Geographic Information Science, Replicability, Meta-Analysis, Bayesian Analysis, Geographic Theory
Session Type: Virtual Paper Abstract
Day: Monday
Session Start / End Time: 2/28/2022 11:20 AM (Eastern Time (US & Canada)) - 2/28/2022 12:40 PM (Eastern Time (US & Canada))
Room: Virtual 59

Abstract

A principal motivation for conducting geographic research is to improve our understanding of spatial phenomena. With increased knowledge of phenomena, it is assumed we will be better-positioned to create interventions that can predictably change real world outcomes. The approach geographers commonly take to generating knowledge and translating that knowledge into interventions is to design, execute, and publish standalone analyses that make claims about the phenomena under investigation based primarily on the evidence of that study alone. We believe this approach often bypasses a formal assessment of the reliability of claims and gives insufficient weight to the evidence already available in the literature. Building on recent advances in meta-science, we introduce into the geographical sciences a formal Bayesian framework that can be used to accumulate evidence from replications performed across space. Unlike existing frameworks, which rely on assumptions that may not hold when examining geographic phenomena, our proposed framework makes explicit the ways in which spatial heterogeneity, spatial dependence, and scale complicate our ability to accumulate knowledge. In addition to outlining our framework, we present a series of examples that demonstrate how this approach to knowledge generation might be deployed in practice. Ultimately, we argue that geographers should take a prospective view of replication and design series of studies that use geographic variation to accelerate knowledge accumulation. We balance this view with a recognition of the confounding effects of place, and call for the further development of mixed methodologies capable of complementing our approach.