Loading Events
  • This event has passed.

Environmental Statistics Seminar – Boyu Ren

February 7, 2020 @ 12:00 pm - 1:00 pm

Nonparametric Bayesian estimation of causal exposure-response curves Causal exposure-response curves (ERCs) describe the causal effects of continuous exposures. Existing methods for the estimation of ERCs are mostly developed from a frequentist perspective and rely on computationally intensive resampling approaches for uncertainty evaluation. We propose a Gaussian process (GP) model to estimate ERCs, which achieves automatic uncertainty evaluation with high computational efficiency. The covariance function of the GP is specified to depend on exposure levels and generalized propensity scores (GPS), and we impose sparsity on the precision matrix via a nearest-neighbor assumption to make the model scalable to massive datasets. Our model effectively implements a GPS matching algorithm in a Bayesian framework and therefore endows the resulting estimates with causal interpretation. We examine the performance of the model with a simulation study and apply it to estimate the causal effect of long-term PM2.5 exposures on mortality from the nationwide Medicare data collected from 2000 to 2016.

Details

Date: February 7, 2020
Time: 12:00 pm - 1:00 pm
Calendars: General Event

Venue