A1098
Title: Combining experimental and observational data for identification and estimation of long-term causal effects
Authors: AmirEmad Ghassami - Boston University (United States) [presenting]
Ilya Shpitser - Johns Hopkins University (United States)
Eric Tchetgen Tchetgen - University of Pennsylvania (United States)
Abstract: The task of identifying and estimating the causal effect of a treatment variable is considered on a long-term outcome variable using data from an observational domain and an experimental domain. The observational domain is subject to unobserved confounding. Furthermore, subjects in the experiment are only followed for a short period of time; hence, long-term effects of treatment are unobserved, but short-term effects are observed. Therefore, data from neither domain alone suffices for causal inference about the causal and must be pooled in a principled way instead. Three approaches for data fusion are proposed. The first approach is based on assuming equi-confounding bias for the short-term and long-term outcomes. The second approach is based on a relaxed version of the equi-confounding bias assumption, where the existence of an observed confounder is assumed such that the short-term and long-term potential outcome variables have the same partial additive association with that confounder. The third approach is based on the existence of an extra variable in the system, which is a proxy of the latent confounder of the treatment-outcome relation. Influence function-based estimation strategies are proposed for each of the data fusion frameworks, and the robustness properties of the proposed estimators is studied.
