[EL] “All the Pretty Little Districts: Why You Need to Stop Whining About Gerrymandering”

[Michael P McDonald]

Micah Altman and I have now worked through three states – Florida, Ohio, and Virginia – comparing plans drawn by the public with those drawn by redistricting authorities, and it is clear that it is possible to chew gum and walk at the same time. That is, that it is possible to draw congressional districts that are more compact, respect more political boundaries, are more politically fair, and respect the Voting Rights Act than the plans that were adopted. Why? Well, congressional districts are very large, with over 700K people in the typical state. It is not difficult at all to unpack Democrats from urban areas with such large districts. What about New York City or San Francisco? Yes, it is true that it is difficult to unpack Democrats from these large urban centers, but this is not where the Republicans gained advantage in the last round of redistricting. There is also the well-known phenomenon of a seat bonus accruing to the majority party in a jurisdiction using single-member districts, thus counteracting the inefficient distribution of partisans in heavily Democratic California and New York.

So, what is wrong with the analysis by Hopkins, Hayes, and Sides cited by Jonathan Bernstein? It asks the question what would have happened if the 2008 elections were run in the 2012 districts. As Hirsch published in the Election Law Journal (the unofficial journal for the list-serve), and have others such as Jacobson and myself separately in other peer-reviewed journals, the 2000’s redistricting was heavily pro-Republican, as much as the post-2010 redistricting it turns out. It took an extraordinary pro-Democratic election in 2006 to overcome the Republican structural gerrymandering advantage. Others argue the same is true for the 1990s, especially with regards to minority seat maximization in Southern states that robbed Democrats of seats and led to the Shaw line of litigation; indeed, in 1996 House Democratic candidates also won more votes than Republicans, but failed to win a majority. The Hopkins et al analysis thus poses the wrong counterfactual question: what would have happened if the 2012 elections had been held in similarly pro-Republican districts drawn in the previous decade. Not surprisingly, they find little evidence that Republicans used the recent redistricting to increase their majority (they find a 7 seat gain); indeed, an alternative conclusion is that the Hopkins et al analysis supports Hirsch, Jacobson, and myself who argue the 2000's redistricting was heavily pro-Republican. The correct question to ask is it possible to draw fair districts given the 2010 census data. This is why analyzing alternative legal plans, such as those drawn by the public who tend to approach redistricting in a much different manner than redistricting authorities, is a better way to assess counterfactuals, like it is possible to draw fair and compact legal plans (plans exist that show it demonstrably is). Nowhere in their analysis do Hopkins et al do a geographic analysis, they just assume their conclusion that it is impossible to draw fair and compact districts is true because it is a theory that conforms with their analysis, even though there are alternative theories that also conform with their analysis that they cannot rule out and do not consider.

Some might argue (as has been done at the Monkey Cage blog and elsewhere) that a simulation analysis by Wei and Rodden published in QJPS corroborate the Hopkins et al argument; however, new analysis presented by these same authors in an expert report in support of plaintiffs in the current Florida redistricting litigation now finds it is possible to draw a fair plan in Florida; i.e., their simulations now cover a fair division of the state, whereas their QJPS article analysis did not. Peer-reviewed work by Micah and I show that redistricting is an NP-hard partitioning problem: automated algorithms designed to generate simulations are thus likely biased in unknown ways and cannot be trusted, except where these simulations can show the existence of a counterfactual scenario, not the absence of one, such as Wei and Rodden argue in QJPS. The presence of compact and fair legal plans drawn by humans similarly provide the counterfactual evidence that Wei and Rodden's simulation approach is indeed biased. Furthermore, Mexico’s experience with automated redistricting algorithms demonstrates that humans do better than the computer when optimizing multiple criteria; redistricting is such a complex optimization problem that computer algorithms tend to get stuck in local optima that humans can imagine their way out of. Mexico’s redistricting is also much less data-intensive than the U.S. and thus more amenable to automated solutions, so it is noteworthy that humans can beat the computer there.

As a further aside, prior to conducting our Public Mapping Project where we empowered the public to draw plans through our web-based DistrictBuilder software, I heard voting rights advocates say that the lay persons could not draw districts with a sensitivity to voting rights. Students proved them wrong by drawing districts that demonstrated how to expand minority representation, say in Virginia, where University of Virginia undergraduates first showed how to create an additional minority influence district in southeast Virginia that later became the flashpoint in negotiations between the Democratic-controlled Virginia Senate and the Republican-controlled House. Citizens also proved them wrong in Minneapolis, where community groups used our software to draw city council districts to empower Hispanic and Somali communities.

The subtext is that it is possible for a redistricting authority to state their criteria and let the public, parties, computers - whatever - have at it and draw districts that optimize on the criteria. This is essentially the approach advocated by Ohio reformers, and to a lesser degree (minus the public participation) the approach employed by New Jersey's congressional and state legislative commissions. Given the right mix of criteria, it is possible to produce districts that are the same or better on virtually everything that we might value in redistricting, except if you are the party controlling redistricting.

If you've read this far and want to know more, our Virginia analysis is published here:

Micah Altman and Michael P. McDonald. 2013. "A Half-Century of Virginia Redistricting Battles: Shifting from Rural Malapportionment to Voting Rights and Participation." University of Richmond Law Review 47: 771-831.

An overview of the use of computers in redistricting and the limitations of automated algorithms is here:

Micah Altman and Michael P. McDonald. 2010. "The Promise and Perils of Computers in Redistricting." Duke J. Constitutional Law and Public Policy 5: 69-112.

The Florida analysis is a forthcoming edited volume book chapter and can be shared upon request. The Ohio analysis is nearing completion as a journal manuscript submission. Eventually, we plan to wrap all of these analyses and more into a book. 

============
Dr. Michael P. McDonald
Associate Professor
George Mason University
4400 University Drive - 3F4
Fairfax, VA 22030-4444

phone:   703-993-4191 (office)
e-mail:  mmcdon at gmu.edu               
web:     http://elections.gmu.edu
twitter: @ElectProject