Eric McGhee and I just posted a new article, The Measure of a Metric, that discusses the measurement of partisan gerrymandering and that’s forthcoming in the Stanford Law Review. In this post and a couple more, I thought I’d highlight the paper’s main contributions.
One of them is to identify a series of criteria that can be used to evaluate gerrymandering metrics. For decades, there was no need for such formal assessment, because social scientists mostly used a single measure (partisan bias). Over the last few years, though, gerrymandering metrics have proliferated, and now include the efficiency gap, the mean-median difference, the difference between the parties’ average margins of victory, the declination, and others. So it’s important today, in a way it wasn’t previously, to think rigorously about what we want and don’t want from a measure.
One thing we want, in our view, is consistency with the following principle: If a party wins more seats while receiving the same votes, a metric should indicate a larger advantage for this party. This principle stems from what we see as the defining characteristic of partisan gerrymandering: manipulating the relationship between votes and seats so that a party is able to translate its popular support into legislative representation more effectively. Whatever else a measure does, it should capture this conceptual core.
Our second criterion is distinctness from other electoral values. Redistricting implicates not just partisan fairness but also electoral competitiveness, minority representation, population equality, and several other concerns. But it’s only partisan fairness that’s at the heart of partisan (as opposed to bipartisan or racial) gerrymandering. So it’s only partisan fairness that should be revealed by a metric.
Third, we think a measure should be broad in its scope. In other words, it should be usable whether a state is red, blue, or purple; whether turnout is roughly equal or varies sharply from district to district; and whether two or more than two parties are competing for office. Gerrymandering is possible in all of these electoral environments, so a metric should not be foiled by any of them.
Our fourth and final criterion is consistency with American electoral history. Every measure implies a certain ideal: a perfect score (usually zero) indicating that neither party benefits from, or is disadvantaged by, a district map. This ideal, in our opinion, should be one that plans have actually achieved with some regularity in previous elections. Otherwise a metric would suggest that most prior maps were gerrymanders—and its adoption would be so disruptive as to be infeasible.
That’s it for today; in the next couple days I’ll address how various measures perform under these criteria.