Dec 14, 2019
We consider the selection potential confounding variables at the stage of the design of a planned observational study. With such aim, given a tentative non-parametric graphical causal model the goal is to select the set of covariates that both suffices to control for confounding under the model and is satisfactory according to some pre-determined criterion. In this talk we focus on the criterion according to which an adjustment set is preferable to another if it yields a non-parametric estimator of some causal contrast of interest with smaller asymptotic variance. For studies aimed at assessing the effect of point exposure, static or dynamic, trx regimes we derive two graphical criteria: one to compare certain pairs of adjustment sets and a second to determine the optimal adjustment set. We show that these graphical rules coincide with rules derived in a recent article by Henckel et al, 2019, assuming linear causal graphical models and treatment effects estimated via ordinary least squares. For point exposure static regimes we also provide a rule for determining the optimal adjustment set among minimal adjustment sets. For studies aimed at assessing the effects of interventions at multiple time points, static or dynamic, we derive a graphical rule for comparing certain pairs of time dependent adjustment sets but we show that no global graphical rule is possible for determining optimal time dependent adjustment sets.
Neural Information Processing Systems (NeurIPS) is a multi-track machine learning and computational neuroscience conference that includes invited talks, demonstrations, symposia and oral and poster presentations of refereed papers. Following the conference, there are workshops which provide a less formal setting.
Total of 0 viewers voted for saving the presentation to eternal vault which is 0.0%
Presentations on similar topic, category or speaker