Limits on Partisan Gerrymandering Are Like Guardrails for Redistricting
Just as guardrails only keep you from driving off the road but don’t ensure good driving within them, preventing egregious partisan gerrymanders won’t ensure fair congressional representation. My analysis shows that even if the Supreme Court finally imposes limits on partisan gerrymandering, a systemic, artificial Republican tilt, reinforced by the SCOTUS decision, will mean that the only chance that Democrats will have of achieving legislative control of the US House will be through wave elections.
As hard as it is to believe, partisan gerrymandering is de facto legal today, because it is not specifically against the law and hasn’t yet been ruled unconstitutional.[1] So, it is tremendously important that the Supreme Court finally puts some limits on egregious partisan gerrymandering, by upholding the Whitford case out of Wisconsin and the related partisan gerrymandering case before it now from Maryland (and maybe North Carolina and/or Pennsylvania, before all is said and done).[2]
The proposed measure of partisan asymmetry at the center of the Whitford case is called the ‘efficiency gap’ (EG). It is a simple, powerful way to identify too much partisan asymmetry — too much partisan bias, in lay terms — and prevent states from metaphorically “driving off the road” with egregiously gerrymandered congressional districts.
However, upholding an EG threshold in Whitford etc. won’t “transform” or “fundamentally change” congressional elections, as widely reported in the media. While counterintuitive at first, a careful analysis of the likely impact of a great ruling for democracy in Whitford reveals that it would likely reduce net Republican “tilt” from 20–23 seats an election to 12–15 seats. This remaining tilt would become judicially protected though, and it would be three times bigger than the “slack” in the system — the expected margin based on vote shares of the parties in the most recent Presidential elections. So, even with a positive Whitford ruling, there will still be many ways that states can wreck their figurative shiny new car of democracy while still staying on the road. When the dust settles, the party that gets fewer votes will still be able to win more seats and control Congress.
The simple fact is that even if a) there’s no additional voter suppression, b) the 2020 Census is as good as it was in 2010, and c) the problem of big money in politics isn’t any worse than it’s ever been, SCOTUS outlawing egregious partisan gerrymandering won’t solve the bigger problem of misrepresentation in Congress. Because a 12–15 seat Republican tilt will remain — the net accumulation of many small deviations from proportional delegations in the 43 states with two or more congressional districts–Democrats will still need wave elections to ever have a chance of overcoming this deficit and achieving legislative control of the US House.
Degree vs. Impact of Partisan Asymmetry
To begin to gauge the likely impact of Whitford, you need to understand two metrics.
The first is the ‘efficiency gap’, the proposed standard for discriminatory effect at the heart of the Whitford case. The EG measures the degree of partisan asymmetry in a set of electoral districts, using the relative wasted votes for parties. It is described in the introductory article “Here’s How We Can End Gerrymandering Once and for All” [http://bit.ly/2C7Ojim].
As opposed to a 7–8% threshold proposed for state legislative elections, the EG authors recommended using a 2-seat threshold for congressional elections. Their reason for proposing a seat-based standard for congressional elections is sound: What matters for control of Congress is aggregate seats at the national level. Seats are the currency of control, and in our highly polarized politics, the party with 218 seats controls Congress. Congress has winner-take-all dynamics now: Parties win and lose congressional elections.
There are two problems with a seat-based threshold though:
- First, because states have widely different numbers of congressional districts, a seat-based standard would imply wildly different levels of allowable disproportionality. Two seats for a state with 6 districts represents 1/3 of the population, while two seats for California represents less than 8%. To treat states equally with respect to partisan asymmetry, the standard for the degree of bias has to be expressed in percentage terms.
- Second, a seat-based standard conflates the degree of bias measured by the EG in percentage terms (EG_%) and the impact of that bias, and the EG expressed in terms of seats — by multiplying EG_% by the number of representatives for a state — is not an accurate reflection of the impact of the bias.
Hence, to measure the impact of partisan asymmetry, you need a second metric.
Congressional elections have an Electoral College-like indirectness, because districts arbitrarily subdivide states’ non-uniformly distributed populations. As a consequence, vote shares don’t translate directly into seat shares. In other words, wins/losses can be expected or unexpected. Hence, the legislative impact of partisan asymmetry in a state is measured by the number of unexpected seats won/lost (UE_#), which is the difference from a proportional delegation based on the share of votes.
The net accumulation of UE_# across states enables a party that loses the state-level popular vote to still control Congress, just like a candidate can lose the national popular vote but still win the Presidency. Hence, congressional elections across states have become interdependent, and misrepresentation in Congress is a national, systemic problem, as opposed to just a state, local problem.
The simple Excel formula for calculating the expected number of seats for a party in an election is ROUND((#_districts * vote_share) — ε), 0), where epsilon (ε) is a very small number that allows you to simplify the formula. UE_# is the difference between that number and the actual seats won.
Bias & Impact by State: 2012–2016
With those two metrics in hand, you can explore how the states compare.
Exhibit 1 plots the 43 states with two or more congressional districts, using the average EG _% and the average UE_# for the 2012–2016 congressional elections, on the x-axis and y-axis, respectively. (See “Appendix: Data Methodology.”) States plotted in red had an EG_% favoring Republicans, while states with an EG_% favoring Democrats are shown in blue. States with a dark circular border had an average EG in terms of seats (EG_#) of two or more. States with fewer than 6 districts are plotted with square borders.
Some examples illustrate the possibilities:
- South Carolina (SC, lower right) had a very high degree of bias (EG_%), but it was under the two-seat threshold (EG_#) because of its small number of representatives (7).
- California (CA, upper left) has a low degree of bias (EG %) but contributes a large number of unexpected seats (UE_#) for Democrats, because it has so many districts (53).
- North Carolina and Pennsylvania (NC and PA, middle right) have very high degrees of bias but contribute smaller numbers of unexpected seats for Republicans, because they have much fewer numbers of districts (13 and 18, respectively).
- States with small numbers of districts (lower left) don’t have efficiency gaps defined for them because they have fewer than 6 districts, but they still contribute unexpected wins/losses!
This last point is essential to understanding the broad problem of misrepresentation in Congress and why even a positive ruling in Whitford won’t change things much: all 43 states that have two or more congressional districts can have non-proportional delegations, and one unexpected win/loss in a small, not-gerrymandered state counts just as much for party control of Congress as one seat more from an egregiously gerrymandered state.
Congressional Election Results 2000–2016
Popping up from a state-by-state view, Exhibit 2 shows an aggregate roll up of the congressional elections for 2000–2016. It turns out that two key metrics define essential congressional election dynamics: “slack” and “tilt.”
The first two rows show the number of seats Republicans and Democrats expected to win, based on state-level vote shares (not the aggregate national vote).
The third row shows the expected margin or “slack,” the number of seats the party that should have controlled Congress could have lost and still controlled the House. Slack is a measure of how many net unexpected wins/losses the system can absorb and still yield the expected result. The data show that slack in the previous 2010 Census-redistricting cycle averaged ~11 seats per election, but in the wake of the 2012 redistricting, average slack narrowed to just ~4 seats.
The fourth row shows the net unexpected wins/losses in an election which measures how many net seats a party won that they shouldn’t have, again based on state-level vote shares. Net unexpected wins/losses is the sum of UE_# across states. Hence, it is a measure of the “tilt” of the congressional election playing field. Again, the data show that in the prior Census-redistricting cycle, tilt averaged just ~2 seats and swung back and forth between Republicans and Democrats, but after the 2012 redistricting, tilt averaged over 20 seats always in favor of Republicans.
Setting aside major swings in vote shares — wave elections — what matters for control of the House is the relationship between tilt and slack. Because tilt was greater than slack in 2000 and 2012, Republicans controlled the House when they shouldn’t have. In other words, in the nine elections from 2000–2016, Democrats should have controlled the House four times (2000, 2006, 2008, 2012), but due to tilt being greater than slack only did twice (2006 and 2008). Moreover, tilt padded Republicans’ margins in 2014 and 2016, making them much bigger and more defensible than they should have been, undermining the possibility for bipartisan action.
Likely US House Tilt After Whitford
With that background, you can gauge the likely impact of a positive ruling in Whitford.
To do so, I made two assumptions:
- The Court would establish an EG threshold of 10% — I assumed that SCOTUS would sort through the seat-based threshold confusion and adopt a percent-based standard and would adopt a relatively low threshold because efficiency gaps are functionally equivalent to malapportioned districts.[3]
- States would react to a clear gerrymandering threshold — Specifically, gerrymandered states would tweak their districts to reduce EG_% but only to 90% of the threshold, thereby maintaining significant partisan bias while avoiding judicial scrutiny; and states not gerrymandered according to this threshold would also tweak their districts to possibly increase EG_% to that now safe degree of partisan asymmetry and gain as much partisan advantage as legally possible, again without being exposing themselves to court challenges.
Under this scenario, Exhibit 3 shows that Whitford would likely eliminate ~8 net seats of Republican tilt but still leave ~12 net seats.[4]
Critically, that remaining Republican tilt of 5.5% (=12/218) would then be judicially protected and would still be three times as big as the average slack in the system. The bottom line is that it will still take wave elections for Democrats to have chances to achieve legislative control of the US House.
If that’s not all bad enough, if you add in the net bias of the seven single-district states — which are, in effect, are perfectly gerrymandered! — they add yet another net 3 seats of Republican tilt. Including them, the post-Whitford Republican tilt would be ~15 seats. In other words, Democrats would have to overcome a ~15-seat structural disadvantage to control Congress.
Takeaways
Make no mistake: I believe it is critically important that the Supreme Court finally places limits on partisan gerrymandering, and the efficiency gap is a great tool for that.
But the promise of the cluster of partisan gerrymandering cases before the Court is much more limited than the media has portrayed. Even with a good-for-democracy ruling the fundamental dynamics of congressional elections — tilt vs. slack — won’t change appreciably. Outlawing partisan gerrymandering is necessary but not nearly sufficient, for the House to represent the people directly. If you are in a position of power or influence, that is the much more tempered message to internalize and share.
Similarly, as important as the case is, you shouldn’t put all of your eggs in the Whitford basket. Within the redistricting/fair representation/vote dilution advocacy space, there are several other complementary efforts to continue pushing on, including developing a standard for discriminatory intent,[5] establishing independent commissions to draw district maps, amending state constitutions to ban partisan gerrymandering, and introducing proportional representation voting & electoral systems.[6] And, more broadly, the franchise is under increased threat, by continuing voter suppression efforts, and the upcoming reapportionment is being undermined, by underfunding the 2020 Census.
Just as road safety accrues from a constellation of practices — guard rails, shoulders, lines, banking on curves, speed limits, stop lights, driver training and testing, etc. — many efforts in combination will be required to achieve to fair representation in Congress.
This is the ninth in a series of articles on redistricting and partisan gerrymandering. A big “Thank you” to Mike Mathieu and the oddfellows review group at Front Seat whose feedback on an early version of this work was invaluable.
Appendix: Data Methodology
While the Brennan Center’s analysis in Extreme Maps covers over 85% of congressional districts (=377/435) for the 2012–2016 elections, it only covers states with 6 or more congressional districts. Hence, it excludes over half the states that can have non-proportional delegations (=22/43).
So, for my analysis, I first captured the public congressional election data for all states for the 2000–2016 elections. I imputed results for uncontested races, using a simple (and in the ballpark) 70/30 split heuristic. Then I overlaid the Brennan data for the states with 6 or more districts for the 2012–2016 elections on top of that.
Brennan’s EGs were presented in terms of seats, so I converted those into percentage terms (EG_%) and used those to calculate implied vote shares, given the known seat shares. I then used vote shares for all 43 states with two or more districts to calculate UE_#. Finally, I computed average UE_# by state for 2012–2016 elections. This is the measure plotted on the y-axis of Exhibit 1 and referenced throughout the analysis.
[1] Racial gerrymandering is illegal, so sometimes states argue in court challenges to their skewed districts that they are “only” partisan gerrymanders!
[2] Gill v. Whitford [http://bit.ly/2BbQtfN].
[3] See the first two sections of “Will the Supreme Court Only Close the Door on Partisan Gerrymandering Halfway?” [http://bit.ly/2xKuHOQ] and “Efficiency Gaps as Variable-sized Districts” [http://bit.ly/2i4Mpq5] for further details.
[4] For what it’s worth, I also looked at an EG threshold of two seats that the EG authors recommended for congressional elections: The results are worse than for the 10% scenario.
[5] See “A Unified Standard for Discriminatory Intent and Effect in Partisan Gerrymandering” [http://bit.ly/2AsgOq6].
[6] For example, see FairVote [http://bit.ly/2rjyQv3] and Sightline [http://bit.ly/2DIJXAh].