Introducing the Reverse Gerrymandering Index: US House Elections 1872-2022

• In Event: Electoral and Voter Behavior

Abstract

In this paper, I present a new measure of gerrymandering called the Reverse Gerrymandering Index (RGIx) and then use it to trace gerrymandering in all US House of Representative elections from 1872 to 2022.

A critical problem with current measures is that they are built on weak theories of gerrymandering, which leads to significant bias. For example, the popular efficiency gap measure (Stephanopoulos and McGhee 2015) treats any inefficiency across an entire state as evidence of gerrymandering, leading to the problematic conclusion that vote differences in opposite sides of states (e.g., districts in downtown Manhattan and upstate NY) are caused by gerrymandering. Similarly, geographic measures like Polsby-Popper (Polsby and Popper 1991) assume that oddly shaped districts are gerrymandered, even though there can be many reasons for odd district shapes. Besides, legislatures have gotten better at drawing highly gerrymandered compact districts. More recent measures based on Monte Carlo simulations (Altman and McDonald 2011; Chen and Rodden 2015; Fifield et al. 2020) perform better than older approaches, but they require so much subdistrict-level data that it can rarely be applied beyond recent US elections.

I argue that the key characteristic of gerrymandering is that it intentionally changes the percent votes across districts that are situated next to each other. For example, packing, the most common form of gerrymandering, is achieved by drawing districts so that one party has a supermajority of support in one district while another party has a slim majority of support in all the neighboring districts. For example, North Carolina’s 12th congressional district in 2012 had a Democratic vote of 80% while its five neighbors had a Democratic vote of between 39% and 47%, leading the Republicans to win a far higher percent of seats than votes.

RGIx works by reversing this gerrymandering of votes. Votes for each party in each district are broken into small groups, and then these groups are randomly moved to either their original district or to one of the adjacent districts. This smoothens the vote across neighboring districts, simulating what the vote might have been without gerrymandering. These smoothened votes act as an expected value for each district, and the RGIx score for each district is then calculated as the actual vote minus its expected value.

These RGIx scores can be applied to legislative election research in various ways, including by using them as a variable in statistical analyses. RGIx has a range of other advantages: for example, it gives single-district states a score of zero instead of treating them as missing values, it can handle multimember and at-large districts common before the 1960s, and it can be applied to electoral systems with more than two parties. Because the only information it requires is the percent vote per party per district and what district are adjacent to each other, RGIx is also easy to apply to historical, comparative, and local elections.

In this paper, I will show how RGIx provides much more detailed results on US House elections over the past 150 years. For example, while some argue that there has been a pro-Republican shift in gerrymandering since the Gingrich era, RGIx shows that this shift actually began a decade earlier, after the 1980 census. RGIx also shows that gerrymandering barely existed for the first half of the twentieth century and then started to rise in the 1950s, a decade before Reynolds v Sims (1964). The percent of packed districts continued to rise for both parties through the 1960s and 1970s, and then after the 1980 apportionment the percent of Democratic packed districts (which favor the Republican Party) continued to rise while the percent of Republican packed districts (which favor the Democratic Party) declined rapidly. Following the 2000 reapportionment, the percent of Democratic packed districts also began declining, but since both Democratic and Republican gerrymanders continued to decline at the same rate, the Republican advantage remained steady into the 2022 election. In other words, the current Republican gerrymandering bias was not simply caused by the Democratic Party’s inaction, but instead by a range of other factors. One of those factors, as RGIx shows, was the South switching over to the GOP and then becoming the most gerrymandered region within the US.

In summary, the goal of the RGIx project is to produce a family of measures that researchers can use to analyze gerrymandering in various ways. I also plan to build publicly available software in Python and R that calculates RGIx statistics for its users.

Briefly about myself: I’m an associate professor at Valdosta State University. I was a postdoctoral fellow at Harvard’s Center for Quantitative Research, a visiting research scholar at Columbia University, and a software developer at Princeton University. My most recent book is The Demise and Rebirth of American Third Parties (2018).

Author

• Bernard Tamas, Valdosta State University