Comment: Matching Methods for Observational Studies Derived from Large Administrative Databases: Mark M. Fredrickson, Josh Errickson, and Ben B. Hansen

In the era of big data, finding a comparable control group for a set of treated units provides new opportunities and challenges. When controls vastly outnumber treated subjects, there will likely be many good potential matches for each treated subject. On the other hand, with larger data sets, increased computation time prevents applying existing methods to find the best possible match. Yu et al. propose a fast caliper solution that restricts the possible controls for each treated subject, making matching with large databases tractable. Their results on determining the narrowest caliper that is compatible with pair matching (without replacement) will be of great practical use. We take issue with the labeling of this caliper as "optimal." The label is accurate in a certain sense -- it does minimize an objective of caliper width, subject to the constraint that pair matching remain feasible while no treatment group member is discarded but these .ire quite different objectives and constraints from those otherwise targeted in the course of optimal matching.