Program

Confirmed speakers

Frederick Eberhardt (CalTech), Robin Evans (Oxford), Jared Murray (UT Austin), Stefan Wager (Stanford), Guillaume Pouliot (UChicago), Ilya Shpitser (JHU), Eli Ben-Michael (UCBerkeley), Steven Howard (UCBerkeley), Elie Wolfe (Perimeter), [Judea Pearl (UCLA)---unable to attend]

Abstracts not listed below are available on the Papers page

Tentative Schedule

8:55am - 9:00am Opening remarks.

9:00am - 9:40am: Jared Murray - Bayesian regression tree models for causal inference: regularization, confounding, and heterogeneous effects

9:40am-10:00am: Ilya Shpitser

10:00am-10:40am: Elias Bareinboim (on behalf of J Pearl)

COFFEE

11:00am-11:40am: Robin Evans - Causal Model Selection and Local Geometry

11:40am-12:00pm: Elie Wolfe - The Inflation Technique for Causal Inference with Latent Variables

12:00pm-12:20pm: Eli Ben-Michael - Matrix Constraints and Multi-Task Learning for Covariate Balance

LUNCH

2:00pm-2:40pm: Stefan Wager

2:40pm-3:00pm: Steven Howard - Uniform nonasymptotic confidence sequences for sequential treatment effect estimation

3:00pm-3:20pm: Guillaume Pouliot - Mixed Integer Linear Programming Formulation with Exact and Asymptotically Robust Inference for Instrumental Variables Quantile Regression

COFFEE

4:00pm-4:40pm: Frederick Eberhardt - Causal Inference in fMRI data

4:40pm-5:00pm (poster flash talks)

5:00pm-6:00pm (poster session)

6:00pm - 6:05 Closing remarks.

Abstracts

Jared Murray: Bayesian regression tree models for causal inference: regularization, confounding, and heterogeneous effects

We introduce a semi-parametric Bayesian regression model for estimating heterogeneous treatment effects from observational data. Standard nonlinear regression models, which may work quite well for prediction, can yield badly biased estimates of treatment effects when fit to data with strong confounding. Our Bayesian causal forests model avoids this problem by directly incorporating an estimate of the propensity function in the specification of the response model, implicitly inducing a covariate-dependent prior on the regression function. This new parametrization also allows treatment heterogeneity to be regularized separately from the prognostic effect of control variables, making it possible to informatively "shrink to homogeneity", in contrast to existing Bayesian non- and semi-parametric approaches.

Elias Bareinboim:

Stefan Wager:

Robin Evans: Causal Model Selection and Local Geometry

Model selection is a task of fundamental importance in statistics, and advances in high-dimensional model selection have been one of the major areas of progress over the past 20 years. Much of this progress has been due to penalized methods such as the lasso, and efficient methods for solving the relevant convex optimization problems that arise in exponential family models. However in some model classes, such as the directed graphical models common in Statistical Causality, correct model selection is provably hard. We give a geometric explanation for why standard convex penalized methods cannot be adapted to directed graphs, based on the local geometry of the different models at points of intersection.

These results also show that it is 'statistically' hard to learn these models, and that much larger samples will typically be needed for moderate effect sizes. This has serious implications for causal learning: we provide some relevant heuristics that give insights into the feasibility of model selection in various relevant model classes, including ancestral graph models and nested Markov models.

Guillaume Pouliot: Mixed Integer Linear Programming Formulation with Exact and Asymptotically Robust Inference for Instrumental Variables Quantile Regression

The instrumental variable quantile regression (IVQR) model provides a set of assumptions under which instruments allow for the identification of the regression coefficient on endogenous covariates in some quantile regression function. Although this approach to causal inference has entered common practice, computations involved in estimation and inference remain prohibitively challenging and often only deliver approximative answers. We deliver a mixed integer linear programming formulation for estimation of the causal effects in the IVQR model which is solved very rapidly with modern solvers and seamlessly accommodates multivariate endogenous variables. Furthermore, this formulation allows us to deliver inference for the causal estimate via inversion of a distribution free, asymptotically robust regression rankscores test, which is the typical method of inference for linear quantile regression.

Frederick Eberhardt: Causal Inference in fMRI data

Functional Magnetic Resonance Imaging (fMRI) is an indirect and aggregated form of measurement of neural activity in the brain. It provides a whole brain view and in the meantime is available in large standardized open-source datasets. I will report on our efforts to apply causal discovery algorithms to, among others, the Human Connectome Project's data, the challenges we have faced both in terms of the nature of the data and the computational features of the discovery algorithms and the different cases and settings we are exploring.

Stefan Wager: Augmented Minimax Linear Estimation

Many statistical estimands can expressed as continuous linear functionals of a conditional expectation function. This includes the average treatment effect under unconfoundedness and generalizations for continuous-valued and personalized treatments. In this paper, we discuss a general approach to estimating such quantities: we begin with a simple plug-in estimator based on an estimate of the conditional expectation function, and then correct the plug-in estimator by subtracting a minimax linear estimate of its error. We show that our method is semiparametrically efficient under weak conditions and observe promising performance on both real and simulated data.