The new Freedom to Vote Act has many interesting pieces, but I want to focus here on the rebuttable presumption of gerrymandering that’s at the core of the Act’s redistricting provisions. In a nutshell, this presumption kicks in if a court finds that an enacted plan (1) exceeds a certain quantitative threshold (2) with respect to certain prior elections (3) according to certain measures of partisan fairness. If the presumption is triggered, a plan can’t be used unless a court ultimately concludes that, actually, the plan was not “drawn with the intent” and does not have “the effect of materially favoring or disfavoring any political party.”
Threshold: I’ll now unpack the different parts of the rebuttable presumption, starting with the quantitative threshold. It’s defined as “partisan advantage or disadvantage in excess of 7 percent or one congressional district, whichever is greater.” A bias score can be converted from a percentage to a number of seats simply by multiplying it by the size of a state’s congressional delegation. For example, Washington has 10 congressional seats. So if a Washington congressional plan has a pro-Democratic bias of 7%, that’s equivalent to a pro-Democratic bias of 0.7 seats.
In practice, the one-seat threshold will be the binding constraint for smaller states (those with 14 or fewer congressional seats). That’s because, in those states, a bias of one seat is always larger than a bias of 7%. On the other hand, the 7% threshold will be the binding constraint for larger states (those with 15 or more congressional seats). That’s because, in those states, a bias of 7% is always larger than a bias of one seat. Note also that these thresholds don’t allow for any rounding. A bias of 1.1 seats (in a smaller state) or 7.5% (in a larger state) exceeds the limit.
Prior elections: Of course, partisan bias doesn’t exist in the abstract. It has to be calculated using particular election data. The Act specifies exactly what data should be employed for this purpose: the two most recent presidential elections and the two most recent Senate elections in a state. An enacted plan’s bias has to be computed with respect to each of those four elections. The plan is presumptively unlawful if it exceeds the applicable threshold (7% or one seat) in “2 or more of the 4 elections assessed.”
Consider Washington again. Its recent presidential and Senate elections have been remarkably consistent. In 2016, Hillary Clinton and Patty Murray each won 59% of the two-party vote. In 2018, Maria Cantwell won 58% of the two-party vote. And in 2020, Joe Biden won 60% of the two-party vote. These four elections provide the data to be used to evaluate any new Washington plan. Any new plan can exceed a bias of one seat in at most one of the four elections. Any new plan that exceeds a bias of one seat in two or more of the four elections is presumptively invalid.
Partisan fairness measures: This leaves the question of how to measure partisan bias. The Act states that the only metrics that can be consulted are “standard quantitative measures of partisan fairness that relate a party’s share of the statewide vote to that party’s share of seats.” The term “standard” does the work of excluding newfangled metrics that aren’t accepted in the academic literature and may even have been devised for litigation purposes. More significantly, the other italicized phrase excludes metrics that don’t specify an optimal seat share for a party’s given statewide vote share. Partisan asymmetry is thus excluded as a metric since it permits any vote share to result in any seat share (as long as, if the parties’ positions were flipped, the same seat share would follow from that vote share). The mean-median difference and the declination are also excluded because they’re not calculated using a party’s seat share.
On the other hand, the efficiency gap is plainly included. In its preferred form, it’s calculated using the formula S – (2 * V), where S is the difference between a party’s seat share and 50% and V is the difference between a party’s statewide vote share and 50%. Also included, at least as long as it’s considered a “standard” measure, is a plan’s deviation from proportional representation. That deviation is computed by simply subtracting a party’s seat share from its statewide vote share.
In situations where the efficiency gap and disproportionality both point in the same direction, consulting two metrics instead of one is unproblematic. What about when they disagree—when a plan’s efficiency gap in a given election is above (below) the threshold but a plan’s disproportionality is below (above) that line? This is where the Act, admirably detailed as it is, finally runs out of steam. In my view, the better approach is to count a strike against a plan only when it exceeds the applicable threshold under both the efficiency gap and disproportionality. If a plan is above the threshold using one measure, but below it using another, I wouldn’t call that a strike. In effect, this approach lets states choose whether they prefer to aim for a low efficiency gap or for low disproportionality. States wouldn’t be compelled to achieve a low efficiency gap and low disproportionality simultaneously—an impossible goal in certain circumstances.
Return to Washington one more time. As noted above, Democratic candidates in the four reference elections received about 60% of the two-party vote. Given this statewide Democratic vote share, minimizing the efficiency gap would entail Democrats winning seven of ten seats, while minimizing disproportionality would entail Democrats winning six of ten seats. Under my preferred reading of the Act, Washington could use either of those benchmarks depending on whether the state wanted a modest winner’s bonus for the majority party (provided by the efficiency gap) or no winner’s bonus at all (per disproportionality). Many outcomes—like nine or more, or four or fewer, Democratic seats—would result in both efficiency gaps and disproportionality values above the applicable threshold. But a considerable range of other outcomes would be allowed because they would produce a sufficiently low score on at least one metric.