More than thirty years ago, Sandy Levinson wrote an article called “Gerrymandering and the Brooding Omnipresence of Proportional Representation.” The article is old but its title was as apt as ever in today’s oral arguments in Rucho and Lamone, in which the Justices asked question after question about proportionality. Justice Alito wondered if there was any way a legislature could pick among computer-generated maps except to “choose one that honors proportional representation.” Justice Kavanaugh inquired whether “proportional representation [is] a judicially manageable standard” and whether the Equal Protection Clause might “require something resembling proportional representation.” Justice Gorsuch asserted that any effect test would ultimately hinge on “how much deviation from proportional representation is enough.” In sum, the Justices and the advocates referred to proportionality almost seventy times.
In response to all this discussion, it’s important to be clear about two things. First, proportional representation is an entirely inappropriate benchmark for an electoral system like ours. And second, none of the Rucho or Lamone plaintiffs’ proposals would actually require proportionality. The reason proportional representation is an improper baseline for single-member districts is that—even if designed without any partisan intent—they rarely produce it. Rather, single-member districts (even if neutrally drawn) generally yield superproportional representation: a seat share for the majority party that is substantially larger than its vote share. Specifically, over the last half-century, American congressional and state legislative maps have averaged a seat-vote responsiveness of almost exactly two. That is, a one percent increase in a party’s vote share has typically led to a two percent rise in the party’s seat share (not the one percent boost required for proportionality).
What accounts for this pattern of superproportional representation? In brief, that most maps contain substantial numbers of competitive districts: indeed, enough such districts to make parties’ seat shares change at roughly double the rate of their vote shares. Suppose that a plan has ten districts, for example, of which Party A won five in the last election with 50% of the statewide vote. If Party A earns 60% of the vote in the next election, it should usually expect to claim 70% of the seats (not the 60% necessary for proportionality). This is because, on average, two of the seats previously won by the opposing party (not one) would flip to Party A as its vote share increases by 10%. Both of these seats would be competitive enough to change hands given an electoral shift of this size.
Proportionality is thus an unsuitable benchmark because it’s an unrealistic expectation for single-member districts. Fortunately, none of the Rucho or Lamone plaintiffs’ methods would insist on it. Start with the random generation of district maps by a computer algorithm that ignores partisanship but satisfies all nonpartisan criteria. If an enacted plan is compared to an array of simulated maps, that is plainly different from measuring the plan’s “deviation from proportional representation” (to use Justice Gorsuch’s phrase). The median simulated map might be one that achieves proportionality. Or it might not be. It all depends on the state’s political geography and districting goals, and how they affect the maps the algorithm produces.
In North Carolina, for instance, the median simulated map is roughly proportional. It includes seven Democratic districts out of thirteen, for a Democratic vote share slightly above fifty percent. In Maryland, on the other hand, the median simulated map (like most plans historically) exhibits hyperproportionality. It contains six Democratic districts out of eight (or seventy-five percent), for a Democratic vote share close to sixty percent. Accordingly, the comparison of enacted plans with randomly generated maps in no way amounts to the imposition of proportional representation. What the method does push states toward is the typical outcome of a redistricting process that takes into account their political geographies and legitimate line-drawing objectives.
Next, consider the measures of partisan asymmetry on which some of the Rucho plaintiffs rely. None of them are equivalent to disproportionality either. The efficiency gap tallies the parties’ respective wasted votes, district by district. It doesn’t even examine the parties’ statewide seat and vote shares—let alone subtract one from the other (as disproportionality does). Likewise, partisan bias indicates how different the parties’ seat shares would be if they each earned the same share of the statewide vote. Any degree of disproportionality is acceptable, as long as each party receives the same “winner’s bonus” if it’s in the majority. And the mean-median difference simply compares a party’s mean vote share, across all of a map’s districts, to its median vote share. The party’s seat share plays no role in this calculation.
There’s a good reason why all these metrics are distinct from disproportionality. It’s that they’re rooted in the quite different concept of partisan symmetry. Symmetry, unlike proportionality, doesn’t require the parties’ seat and vote shares to be equal. Rather, it more flexibly asks that the parties be able to translate their popular support into legislative representation with approximately equal ease. Also unlike proportionality, symmetry is an appropriate aim for single-member districts. In fact, over the last fifty years, American congressional and state legislative maps have had average efficiency gaps, partisan biases, and mean-median differences of almost exactly zero. This family of measures, then, is no surreptitious attempt to import foreign notions of proportional representation. It’s an effort, instead, to nudge American maps in the direction of their own historical norm—which has always been symmetry but not proportionality.