PROXIMAL CAUSAL INFERENCE AND BEYOND:
MEDIATION AND IV PROXIMAL METHODS FORUNMEASURED CONFOUNDING
Abstract：
Unmeasured confounding is a central challenge in observational studies,
often undermining causal interpretation of exposure outcome associations.
While instrumental variable (IV) methods can address this problem, they rely
on strong and difficult to verify validity assumptions. Proximal causal
inference provides a more flexible alternative by leveraging negative
control variables variables that are not causally related to the primary
exposure or outcome but are associated with latent confounders to detect
and correct bias from unobserved confounding without additional data
collection. This talk will introduce the key concepts through intuitive examples,
illustrate how proximal causal inference can complement and, under weaker
assumptions, overcome limitations of traditional IV methods, and discuss
applications across medicine, public health, and the social sciences. In
addition, I will highlight our recent work extending proximal ideas to more
complex causal structures, including proximal mediation analysis for
decomposing effects in the presence of unmeasured mediator outcome
confounding, as well as IV proximal causal inference frameworks that
integrate instrumental variables with proximal identification strategies to
improve robustness when standard IV assumptions are questionable.