Controlling for spatial confounding and spatial interference in causal inference: Modeling insights and the spycause package
Causal inference is a rapidly growing field of statistics that applies logical reasoning to statistical inference to estimate causal relationships. Spatial data poses several problems in causal inference—namely, spatial confounding (spatial heterogeneity) and interference (spatial dependence)—that require different strategies when designing causal models. Given the blossoming literature on spatial causal inference, this research analyzes the usage of spatial causal models under a priori knowledge and a priori ignorance of the spatial structure of data. We synthesize existing research directions in noncausal spatial modeling and causal nonspatial modeling by assessing the performance of 28 spatial causal models across 16 spatial data scenarios. In doing so, we compare our results to principles of noncausal spatial modeling and investigate their implications for spatial causal modeling. In parallel, we build the spycause Python package of spatial causal models and data simulators to facilitate the widespread use of these models and to enable reproduction of this work. We find that the choice of modeled spatial structure matters immensely for identifying causal relationships in space. To conclude, we touch on emerging work in pursuit of more adaptive techniques that may result in more flexible structures for modeling spatial dependence.
This seminar series is co-organized by CHUD (Center for Housing & Urban Development), GeoSAT (Center for Geospatial Sciences, Applications and Technology), and TAMIDS-DAL (Design and Analytics Lab for Urban Artificial Intelligence @ Texas A&M Institute of Data Science).
Coming from a background of math and computer science, Tyler developed algorithms and statistical methods for spatial data science. His work aims to better represent and mathematically model spatial problems and to chip away at our understanding of core geographical concepts, such as process and scale. Through algorithm design, He also seeks to advance the principled use of (spatial) data science in social science. For more information about his research, go to https://www.tdhoffman.com/. In his spare time, He loves watching movies, playing piano, and biking. He is a recipient of the NSF Graduate Research Fellowship and currently a Ph.D student at Arizona State University in the Spatial Analysis Research Center (SPARC).
Time: 10:00-10:30 a.m. US Central Time (Thursday, May 11th, 2023)
Host: Jiaxin Du, Data Science Ambassador@TAMIDS, PhD candidate@LAUP, TAMU