Title: Optimal tradeoffs in matched designs for observational studies
Authors: Samuel Pimentel - UC Berkeley (United States) [presenting]
Rachel Kelz - University of Pennsylvania (United States)
Abstract: An effective matched design for causal inference in observational data must achieve several goals, including balancing covariate distributions marginally, ensuring units within individual pairs have similar values on key covariates, and using a sufficiently large sample from the raw data. Yet optimizing one of these goals may force a less desirable result on another. We address such tradeoffs from a multi-objective optimization perspective by creating matched designs that are Pareto optimal with respect to two goals. We provide tools for generating representative subsets of Pareto optimal solution sets and articulate how they can be used to improve decision-making in observational study design. We illustrate the method in reanalysis of a large surgical outcomes study comparing outcomes of patients treated by US-trained surgeons and of patients treated by internationally-trained surgeons. Formulating a multi-objective version of the problem helps us evaluate the cost of balancing an important variable in terms of two other design goals, average closeness of matched pairs on a multivariate distance and size of the final matched sample.