The consequences of social determinants of health (SDOH) increasingly dominate public health discussions in the United States, as population health outcomes have not kept pace with those of other developed nations despite higher per-person spending for medical services.^1-3 An increased understanding of SDOH could be used to better connect patients with relevant social services in clinical contexts and could target vulnerable populations with health-improving social policies and programs while also addressing affordability.^4,5 Health policy frameworks that directly address the underlying social and behavioral determinants of health are now encouraged to promote improvements in population health outcomes and cost savings.⁴ However, more finely quantifying these determinants at scale remains a challenge.

Empirical work in regional science has seen a growing interest in causal inference, leveraging insights from econometrics, statistics, and related fields. This has resulted in several conceptual as well as empirical papers. However, the role of spatial effects, such as spatial dependence (SD) and spatial heterogeneity (SH), is less well understood in this context. Such spatial effects violate the so-called stable unit treatment value assumption advanced by Rubin as part of the foundational framework for empirical treatment effect analysis. In this article, we consider the role of spatial effects more closely. We provide a brief overview of a number of attempts to extend existing econometric treatment effect evaluation methods with an accounting for spatial aspects and outline and illustrate an alternative approach. Specifically, we propose a spatially explicit counterfactual framework that leverages spatial panel econometrics to account for both SD and SH in treatment choice, treatment variation, and treatment effects. We illustrate this framework with a replication of a well-known treatment effect analysis, that is, the evaluation effect of minimum legal drinking age laws on mortality for US states during the period 1970–1984, a classic textbook example of applied causal inference. We replicate the results available in the literature and compare these to a range of alternative specifications that incorporate spatial effects.