Statistics and Data Science Seminar
Prof. George Karabatsos
UIC, Education Psychology
A Bayesian Nonparametric Causal Model For Observational Studies
Abstract: Often, causal inference is conducted on the basis of the randomized
experiment. However, in many settings, a randomized experiment is
infeasible, because treatments cannot be directly assigned to subjects.
Specifically, it may not be possible for the investigator to assign
treatments to subjects, because of ethical concerns, or because of
excessive expense in terms of time or money. In such settings, causal
inference needs to be undertaken in an observational study, where the
subjects received different treatments, but the investigator did not
assign the treatments, and therefore the treatment assignment
probabilities are unknown.
Typically, in the practice of causal inference from observational studies,
a parametric model is assumed for the joint population density of
potential outcomes and treatment assignments, and possibly this is
accompanied by the assumption of no hidden bias. However, both assumptions
are questionable for real data, the accuracy of causal inference is
compromised when the data violates either assumption, and the parametric
assumption precludes capturing a more general range of density shapes
(e.g., heavier tail behavior and possible multi-modalities in the joint
density). We introduce a flexible, Bayesian nonparametric causal model to
provide more accurate causal inferences. The model makes use of a
stick-breaking prior distribution, which has the flexibility to capture
any multi-modalities, skewness and heavier tail behavior in this joint
population density, while accounting for hidden bias. We prove the
asymptotic consistency of the posterior distribution of the model. Also,
we illustrate our Bayesian nonparametric causal model through the analysis
of small genetic data set, and a large data set of Chicago public schools.
Wednesday February 4, 2009 at 3:00 PM in SEO 636