|
Activity Number:
|
108
- Can't Shake That Feeling That You're Missing Something?
|
|
Type:
|
Invited
|
|
Date/Time:
|
Monday, August 3, 2020 : 1:00 PM to 2:50 PM
|
|
Sponsor:
|
Health Policy Statistics Section
|
|
Abstract #309449
|
|
|
Title:
|
Efficient Sensitivity Analysis for Propensity Score Weighting
|
|
Author(s):
|
Lane Burgette* and Beth Ann Griffin and Joseph Pane and Daniel F McCaffrey
|
|
Companies:
|
RAND Corporation and RAND and RAND and ETS
|
|
Keywords:
|
causal inference;
propensity score ;
weighting;
sensitivity analysis;
confounding
|
|
Abstract:
|
In propensity score analyses based on weighting, unobserved confounders can bias treatment effect estimates because their exclusion distorts the estimated propensity score weights. We derive expressions that allow us to simulate the distortions to propensity score weights that would have occurred if an unobserved confounder – with a given in-sample correlation with the outcome and effect size with respect to the treatment indicator – had been excluded from the propensity score model. We demonstrate how this approach can be used to efficiently perform sensitivity analyses for weights that arise from a nonparametric propensity score estimation approach. In particular, we propose a graphical method of assessing sensitivity of point estimates and p-values to unobserved confounders with a range of associations with the treatment indicator and the outcome of interest.
|
Authors who are presenting talks have a * after their name.
