|
Activity Number:
|
128
- New Methods and Diagnostics for Propensity Score Matching
|
|
Type:
|
Topic-Contributed
|
|
Date/Time:
|
Monday, August 9, 2021 : 1:30 PM to 3:20 PM
|
|
Sponsor:
|
Health Policy Statistics Section
|
|
Abstract #317616
|
|
|
Title:
|
Full Matching with More Precise Set Size Constraints
|
|
Author(s):
|
Mark Fredrickson* and Ben B. Hansen
|
|
Companies:
|
University of Michigan, Ann Arbor and University of Michigan
|
|
Keywords:
|
causal inference;
matching;
optimal full matching;
optimization
|
|
Abstract:
|
Given collections of treatment and control units, a matching partitions all units into sets containing at least one treatment and at least one control subject. When the units are augmented with a treatment-control distance matrix, optimal full matching returns a matching that minimizes the sum of within-group treatment-control distances among all possible partitions. Optimal full matching has the interesting additional qualification that all matches are pairs or have only one unit from either the treatment group or the control group. In practice, this often results in at least some sets with a very large number of treated units sharing a control, or vice versa. In this talk we introduce a new variant of full matching affording the statistician precise control over the extent to which treatment units share control units. We explain the implementation of this method in the open source Optmatch package for R and demonstrate the method on several data sets.
|
Authors who are presenting talks have a * after their name.
