scispace - formally typeset
Search or ask a question
Book ChapterDOI

Optimal Nonbipartite Matching

01 Mar 2023-pp 227-238
TL;DR: In more complex real-world scenarios, multiple treatment groups (more than two) are not uncommon, either in the form of different dose levels or as several unordered intervention arms as discussed by the authors .
Abstract: Matching is a powerful design tool to remove measured confounding in both observational studies and experiments. Conventional matching designs focus on two-group setup, namely, treated and control groups, which is also known as bipartite matching. In more complex real-world scenarios, multiple treatment groups (more than two) are not uncommon, either in the form of different dose levels or as several unordered intervention arms. This chapter reviews the methodology for creating matched pairs in the presence of multiple groups, or in situations without clear grouping, which is referred to as nonbipartite matching. Such a design may be used to match with doses of treatment, or with multiple control groups, or with various time points in a longitudinal study, or as an aid to strengthen the instrumental variable analysis.