Recognizing and Overcoming Obstacles to Causal inference Arising from Spatial Structures

1041 

Invited Paper Session 

Brian Gilbert (1), Elizabeth Ogburn (2), Brian Gilbert (1), Youjin Lee (1), Brian Reich (3), Georgia Papadogeorgou (1), Mauricio Garcia Tec (4)

(1) N/A, N/A, (2) Johns Hopkins University, N/A, (3) North Carolina State University, N/A, (4) University of Texas At Austin, N/A

Brian Gilbert  
N/A

Elizabeth Ogburn  
Johns Hopkins University

Brian Gilbert  
N/A

Youjin Lee  
N/A

Brian Reich  
North Carolina State University

Georgia Papadogeorgou  
N/A

Mauricio Garcia Tec  
University of Texas At Austin

As ecological and political factors are critical to questions of contemporary public policy, the spatial aspects of statistical data cannot be ignored. Causal inference, which explicitly or not is invariably the goal of policy analysis, has traditionally relied on assumptions that are often violated in practice and faces special challenges when it comes to spatial data. In particular, the effects of the exposures of interest may "spill over" into adjacent areas, there may be unmeasured confounding variables that are functions of spatial location, and the nature of spatial sampling may not be compatible with the usual "independent and identically distributed" model of units. However, by incorporating spatial structure into statistical models, these issues may be mitigated; we are beginning to bridge the gap in the contemporary literature between causal inference and spatial statistics, which presents an exciting opportunity both for methodological advances and for more accurate inference on the part of applied statisticians.

Youjin Lee, Assistant Professor of Biostatistics at Brown University, will present "Interference, cross-border shopping, and substitution effects" which provides a methodology to evaluate the effects of policies enacted in one administrative unit which may also affect bordering units. This will be followed by Mauricio Tec, a postdoctoral fellow at Harvard University, presenting "SpaCE: The Spatial Confounding (Benchmarking) Environment" which provides suite of semi-synthetic datasets to analyze methods for controlling for unmeasured spatially-varying confounding variables. Next, Georgia Papadogeorgou, Assistant Professor of Statistics at the University of Florida will discuss "Spatial causal inference in the presence of unmeasured confounding and interference" which explains how interference and unmeasured confounding (the topics of the two previous talks) look similar in observational data and how substantive scientific assumptions may distinguish between them. The final talk will be "Resolving informative spatial sampling and causal inference" by Brian Reich, Distinguished Professor of Statistics at North Carolina State University, which will describe how nonrandom spatial sampling can be analyzed in a manner compatible causal inference. Betsy Ogburn, Associate Professor of Biostatistics at Johns Hopkins University, will serve as discussant. An expert in causal inference for dependent data, she will analyze the issues presented with reference to the fundamentals of causal modeling and put forth connections to analogous issues in the study of social networks.

This session will serve multiple aims of JSM 2024. All speakers will reference issues relevant to informing policy; for example, Dr. Lee's analysis of interference effects will involve a discussion of beverage taxation in Pennsylvania. More broadly, the methodology of causal inference for complex dependence structures is critical to most policy analysis, especially at the national level when intervention and confounding factors vary by state. The session will hold appeal for statisticians of diverse interests, both methodological and applied, whether working in public health, ecology, political science, economics, or sociology. Finally, the roster of speakers will contribute to representation and equity at the conference, with three women and three early-career researchers.

Biometrics Section ³

ENAR ²

Section on Statistics and the Environment ¹