A new paper by Jeffrey Buzas and Gregory Warrington. Here’s the abstract:
We introduce simulated packing and cracking as a technique for evaluating partisan-gerrymandering measures. We apply it to historical congressional and legislative elections to evaluate four measures: partisan bias, declination, efficiency gap, and mean-median difference. While the efficiency gap recognizes simulated packing and cracking in a completely predictable manner (a fact that follows immediately from the efficiency gap’s definition) and the declination does a very good job of recording simulated packing and cracking, we conclude that both of the other two measures record it poorly. This deficiency is especially notable given the frequent use of such measures in outlier analyses.