Differential privacy improves treatment effect estimates in sensitive data
Differentially Private Average Treatment Effect Estimation by Propensity Score Blocking
Cryptography and SecurityMachine Learning
Summary
Estimating how a treatment affects outcomes is important in many fields that handle private data, like medicine and social science. The authors present two new ways to calculate the average effect of a treatment, while protecting individual privacy. Their methods use techniques based on how likely someone is to receive the treatment, called propensity scores. One of their approaches greatly reduces errors compared to previous methods, making estimates safer and more accurate when dealing with sensitive information.
What this means in practice
- •For hospital data teams: Estimate patient treatment effects from sensitive medical records while ensuring strong privacy protections.
- •For policy analysts: Evaluate social program outcomes from confidential survey data with improved accuracy and privacy guarantees.
Authors
Duncan Stewardson, Grayson W. White, Adam Groce
Abstract
Average treatment effect (ATE) estimation in observational studies is a fundamental statistical tool used frequently in social science, medicine, and other fields. These fields often work with sensitive data where privacy protections are important, so a differentially private mechanism for ATE estimation is highly desirable. Here we present two propensity score-based algorithms for ATE estimation on observational data, one improving the inverse probability weighting (IPW) method used in prior work, and the other using blocking on the propensity score (BPS). Both show lower error and less bias than prior work, with the BPS-based algorithm frequently reducing error by 75% or more compared to prior work.