Symposium: The efficiency gap is a measure, not a test

Eric McGhee is a research fellow at the Public Policy Institute of California. PPIC has not reviewed or endorsed this blog post, and takes no position on its use for any purpose. The views expressed are McGhee’s personal views, and not those of PPIC.

One of the more unusual aspects of Gill v. Whitford is the role played by a new measure of partisan advantage called the efficiency gap. In district-based elections, some votes are “wasted” in the sense that they do not contribute to a victory. These include any votes cast for a losing candidate, as well as those cast for a winner in excess of the number required to win. The party with more wasted votes is at a disadvantage because its support does not translate as efficiently into representation. Parties that gerrymander try to saddle the opposing party with more wasted votes by “packing” that party’s supporters into a small number of districts it will win handily and “cracking” the rest into districts that the party narrowly loses. The EG is just a summation of all the packing and cracking in a plan. It takes the difference between the total wasted votes for each party and divides it by the total number of votes cast.

The plaintiffs in this case relied on the EG (among other measures) in the lower court, and the lower court itself referenced the measure in its decision to strike down the Wisconsin map. Since then, the EG has been attacked from a number of different angles. I have no idea whether the Supreme Court will approve of the EG or even rely on it at all. But I can say that virtually every criticism of the EG fundamentally misunderstands what it does and was intended to do.

I invented the EG. I am not a legal scholar but a political scientist, so I developed it to be a measure: a means of systematically capturing a particular idea. I later joined forces with Nicholas Stephanopoulos, who is a legal scholar (and a very good one). (Stephanopoulos is also counsel to the challengers, but I am not working with him on this case.) We folded the EG into a possible legal test: a set of rules that clearly articulates how to weigh competing interests and principles in a wide range of situations so justice is as swift and certain as possible.

Measures and tests have different goals. A measure is effective if it reflects the essence of the idea it intends to capture and does so consistently across a wide range of circumstances. For that reason, measures benefit from precision. The more a measure is tailored to one specific concept, the more likely it is to capture that concept every time it is applied.

Legal tests are effective if they incorporate all the values and principles people care about, are clear about which should be preferred, and offer thresholds for action. For that reason, tests benefit from breadth. A test that fails to factor in all the relevant considerations will likely run into problems.

Because tests go broad and measures go narrow, a good legal test will probably end up measuring more than one thing. That does not make any of the individual measures it uses “wrong” any more than a thermometer is wrong because it tells you how hot the air is and not how hot you actually feel. (For the latter you likely need the relative humidity as well.) It simply means the test requires more than one measure to describe the complex constellation of things it cares about.

As a measure, the EG is very good at its job. It summarizes relative efficiency, one of the key mechanisms behind partisan advantage in a single-member-district system. In so doing, it satisfies what I call the “efficiency principle”: It confirms that a party that wins more seats without winning more votes is better off as a result. This principle may sound obvious, but many existing measures of partisan advantage fail to satisfy it consistently. The EG not only satisfies the efficiency principle, it does so in a wide range of contexts. It works for both competitive and uncompetitive states, with or without variation in district size or turnout, and ignoring or incorporating minor parties.

In short, the EG is excellent as a measure of efficiency. No argument I have seen disagrees with this basic point. Rather, the criticisms take one of three tacks: They question the EG’s implications, they assert it should measure more than one thing, or they implicitly or explicitly take issue with one legal test or another. The first two misunderstand what good measurement is supposed to do, while the last mistakenly assumes that disagreements with a test cast doubt on the measures it uses.

The first line of attack criticizes aspects of the EG that are not design features so much as implications of the concept it measures. For instance, many have said the EG is too volatile. It is true that there is rarely a single EG value for the life of a redistricting plan. But if one agrees that packing and cracking are important features of a gerrymander and that the EG measures them, then it is the advantage from packing and cracking that is volatile and the EG just reflects that. The EG’s volatility is a feature, not a bug: The courts should want any and all uncertainty front and center before they decide to intervene.

Likewise, many have taken issue with what the EG does not measure. It does not directly measure (among other things) majority party entrenchment, competitiveness, racial or ethnic minority representation, district shapes, the durability of any partisan advantage, or whether the redistricting authority intended to benefit either party when it drew the maps. But a good measure does not try to bite off more than it can chew. The more one freights a measure with multiple concepts, the less it measures any one of them effectively.

That is not to say the EG is useless for measuring some of these ideas. Durability can be estimated quite well by adjusting the election outcomes in plausible ways to see if a particular EG advantage survives. Any defensible assumptions about voter loyalty to parties can be built into this exercise. However, although such “sensitivity testing” uses the EG, it really amounts to a different measure – one that quantifies the potential volatility of a plan’s partisan advantage. Stephanopoulos and I have strongly advocated such testing to ensure that a particular EG value is not ephemeral.

Sensitivity testing can also explore majority entrenchment or competitiveness. But there are other measures that capture those ideas more directly, and a strong case can be made that those measures should be used for those concepts instead. Any remaining features of a redistricting plan should be identified, if at all, with other measures. Multiple measures are best for multiple concepts.

The final line of attack takes issue with any one of the specific legal tests that use the EG and pins any identified problems on the EG itself. An example is the test Stephanopoulos and I proposed, which says that if a plan’s EG is greater than eight percent (or two seats for congressional plans) and the sensitivity testing reveals a persistent advantage for one party, the burden shifts to the state to defend the plan as necessary for some other reason. If it cannot, the plan is struck down.

I like this test, but neither it nor any other is handed down on stone tablets. Reasonable people can disagree about specifics. In fact, Stephanopoulos and I explicitly left the details of ours open to future revision. Most importantly, decisions about legal tests have no bearing whatsoever on the quality of the measures employed. To dismiss the EG based on disagreements with the test sends baby and bathwater out the window together.

Once one accepts this distinction, many criticisms of the EG are revealed as complaints about tests in disguise. Take the claim that the EG advocates “super-proportionality” because it implies a 2-1 seats-to-votes ratio (i.e., a party should get an extra two percent of the seats for every one percent increase in vote share). This ratio itself is a mathematical implication of the measure and not a design feature, but regardless, it only constrains to the extent that it is strictly applied. If a party can instead enjoy an EG of eight percent in its favor (as Stephanopoulos and I suggested), a wide range of ratios becomes permissible. Even eight percent is just a suggestion: The Supreme Court can set the threshold as loosely as it wants, widening the potential range even more.

The same is true for claims that the EG will upend American elections. The disruptive potential is entirely a function of the test that is applied. A test that prioritized partisan fairness and offered little wiggle room around the ideal 2-1 ratio would have large effects; a test that prioritized other interests in redistricting and legitimized large EG values would have small ones.

Even the concerns about what the EG does not measure are really arguments about what a potential legal test should include. The Supreme Court might use the EG as one criterion for defining an unconstitutional gerrymander and then include others such as district shape, competitiveness, minority representation and the like – each with its own legitimate measure. In fact, the EG test Stephanopoulos and I proposed allowed for such a mixture of ideas (albeit after the EG was employed to identify a problematic case). The Supreme Court has a lot of flexibility.

At the end of the day, measures are tools. If well designed, they provide a consistent, systematic means of knowing how much one has of some specific thing. Whether that information is useful for governing is a matter of values and principles that a measure cannot resolve. The justices must decide for themselves what makes the most sense for the country. Whatever they decide, the EG can be a flexible and powerful tool for guiding their decision.