A fundamental challenge in observational causal inference is the correct specification of a causal model. When identification assumptions are uncertain, analysts may seek to triangulate effect estimates across multiple candidate models that rely on different assumptions. Yet principled methods for quantitative triangulation remain underdeveloped. We develop two complementary frameworks to address this gap. The first concerns testing the causal null hypothesis and yields a procedure that is valid whenever at least one candidate model is correctly specified. The second focuses on effect estimation and produces a combined estimator that is consistent provided at least one model is both correct and empirically testable. Together, these frameworks formalize robustness under methodological pluralism without relying on model selection or agreement across models—an assumption often implicit in the triangulation literature. Throughout the talk, I will also weave in discussions on theoretical and practical challenges that arose (and some that still remain) in implementing such frameworks.
Co-authors on these works: Ted Westling, Youjin Lee, Junhui Yang, He Bai, Ina Ocelli