Primary Submission Category: Matching
A Surprising Granularity when Coarsening Continuous Variables with Coarsened Exact Matching
Authors: Aran Canes, Jigar Shah,
Presenting Author: Aran Canes*
Iacus, King and Porro introduced the well-known method of Coarsened Exact Matching (CEM) in 2011. As they state, “The basic idea of CEM is to coarsen each variable by recoding so that substantively indistinguishable values are grouped and assigned the same numerical value.”
We’ve used CEM in retrospective observational studies for many years. What we wanted to know is whether common continuous variables, such as age, could be coarsened into large enough categories to allow for a reasonable number of matches, and thus the evaluation of the ATE, as well as be small enough so that the limits are substantively indistinguishable.
We were able to test these assumptions by looking at the effect of coarsening continuous variables on a prior version of the outcome. By creating increasingly more granular categories until the limits were statistically insignificant predictors of the prior outcome we tested the “Substantively indistinguishable” assumption. The number of categories for certain variables was in excess of one hundred—making it impossible to perform CEM and meet all assumptions.
Given the increasing use of CEM, we believe these results are important as showing that common-sense or intuitive coarsening of continuous variables may be violating the assumptions of CEM and leading to inaccurate measures of the treatment effect.