You might think these neat lines above would produce fair outcomes and protect the process from bias resulting from gerrymandering. However, all it does is needlessly and unproductively split communities, cities, and counties. Just as importantly, it violates the Voting Rights Act and denies black voters the ability to elect the candidate of their choice.
North Carolina’s cultural geographic regions. Click to enlarge
Proposed non-partisan North Carolina congressional districts. Click to enlarge
With the above geographic regions in mind, let's compare the first map to what a realistic, non-partisan alternative might look like. This map creates a 1st district that is plurality black, in compliance with the VRA as it existed during redistricting. Instead of splitting the three cities of the Piedmont Triad between two districts, it places them neatly in the 9th. Charlotte isn't randomly combined with ultra-white, wealthy exurbs to its south, but the 12th is instead entirely urban and majority-minority. All three of these districts would have now been likely to elect a black representative.
Furthermore, the four cities of the Research Triangle are moved from five districts to just the 4th and 13th. Even outside of big cities, rural areas now conform more closely to the regions of cultural geography shown above, while overall the map strives to reduce any unnecessary division of existing political jurisdictions, such as cities, counties, and small towns. This map leads to superior representation, because it aims to give each region representation based upon its own unique characteristics.
Using single-member districts is itself a flawed way to elect a legislature compared to proportional representation. However, since this is the system we have, we should ultimately want to draw districts based upon these five factors, in order of prioritization:
1. Ignoring partisanship.
2. Compliance with the Voting Rights Act's demand for majority-minority districts.
3. Utilizing communities of interest like shared culture, economic class, etc.
4. Minimization of unnecessary county and municipality splits.
5. Geographic compactness. Not geometrically minimizing district boundary length, but drawing districts so that they don't combine disparate parts of intrastate regions.
I will expand further on why these principles are both necessary and superior to utilizing geometric compactness alone. Additionally I will demonstrate the flaws in the political science work that relies on compactness to assert that gerrymandering is relatively unimportant to the maintained Republican control of the US House.
It is a fact of American political geography that Democrats are more geographically clustered than Republicans. Inner cities vote overwhelmingly for Democrats while rural areas and suburbs are not as heavily Republican. This phenomenon and the consequences of it are referred to as geography bias.
The primary political science work cited by proponents of the geography explanation is that of professors Jowei Chen and Jonathan Rodden. Their method involved having a computer draw up thousands of random maps for various states. From there, they determined the statistical likelihood of each outcome using the 2000 presidential result to measure geography bias. If we extended the original North Carolina map to every state, we would see a clear bias in favor of Republicans.
However, this method, is problematic for four reasons.
First, using only the 2000 presidential result to determine the likelihood of a 2012 congressional outcome relies on old data that has changed in 12 years, especially in Florida, which Chen and Rodden focused on. Second, it fails to take into account incumbency, which is still a very important factor in congressional elections. Third, these maps also violate the VRA by ignoring the need for majority-minority districts. Finally, why should we needlessly divide communities and thus weight the absurd the same as what might be logical?
To demonstrate why that last point is a problem, here is a map of two hypothetical cities from a paper by political scientists Michael McDonald, Micah Altman, Brian Amos, and Daniel Smith.
Each block in this map is assumed to be a precinct of equal population, thus there are two cities in the center-left and center-right of the map. Now imagine that we are dividing this map into two districts. The most logical way of drawing it would be to split the map east and west, with each city comprising its own district as pictured in the middle. However, a method of randomly drawing the map could just as easily split it north and south with each city being cut in half as shown on the right. This method makes no sense, but it is just as geometrically compact as the middle iteration. Prioritizing geometric compactness above everything else can lead to illogical outcomes such as needlessly splitting cities.
We can see for ourselves what makes a map better than an alternative, but randomly generating them cannot do that. Additionally, random maps generated by Chen and Rodden cannot account for a fundamentally subjective judgment when it comes to districts and regions protected by the Voting Rights Act. How would a computer determine when and how to draw a district that protects minority voters?
Ultimately, the concept of what is fair is itself subjective and requires human input rather than a computer algorithm. Let’s take a more detailed look at the work from Chen and Rodden. For instance, they note how some randomly generated iterations yielded a second Democratic district in St. Louis, Missouri. But to do that you have to draw fairly crazy lines that make no sense.
Proposed non-partisan Missouri congressional districts, St. Louis closeup
Here are two iterations that show just how absurd the lines have to be to elect a second Democrat. Because the 1st district is protected by the Voting Rights Act, it needs to be heavily black even if African-Americans cannot physically be a majority. This first map shows what an optimal non-partisan rendering might look like. The 2nd district is still solidly Republican, since despite Obama coming close to winning it in 2008, Romney easily carried it by 14 points.
To get a 2nd district that is even competitive for Democrats, you have to split the city of St. Louis. That alone isn't enough though, since you also have to remove the red suburbs in St. Charles County along with part of St. Louis County. In place of those wealthy suburbs, the 2nd takes in more Democratic-friendly exurbs and rural areas to the south in Jefferson County.
Alternate Missouri congressional districts with a competitive 2nd District
This second map still sees Romney win the 2nd by one percent even though Obama won it in 2008. You can only imagine how ugly the lines have to be to actually create a second Democratic seat for then-Rep. Russ Carnahan, who nearly lost a safe district in 2010. This would be pretty blatant gerrymandering and there is utterly no good reason why we should include maps like this in our data set. Geography bias hurts Democrats in Missouri, but we can't conclude from this alone that it's the same in other states.
In creating a set of random maps to measure quantitatively, Chen and Rodden have subjectively selected their own data set by prioritizing compactness and ignoring other factors. In short, this method is an exercise in confirmation bias where looking for the impact of geography bias has simply demonstrated it in their underlying dataset. Furthermore, the two scholars contradicted their own work in a subsequent expert report on Florida's most recent redistricting.
There are effectively infinite ways to draw a map. The choice of building blocks itself is a subjective decision that can impact the ultimate map. For instance, in many states the precincts themselves are gerrymandered, can vary drastically in population, and combine areas of vastly different partisanship. Even non-partisan census blocks put restrictions on the (already massive) number of combinations the computer can generate. You can only imagine the staggering number of combinations given the number of people or acres in each state.
Calculating the likelihood of a district being drawn thus makes little sense, because modern computing power and statistical data make it impossible to produce a sample that could reasonably reflect every single combination. By simply attempting to draw the best map for every state, we can then estimate the most likely outcome resulting from it.
Proposed non-partisan Virginia congressional districts. Click to enlarge
Communities of interest should be the foundation for drawing districts. This term entails the economic, geographic, cultural, and demographic makeup of the district. Combining areas that have commonality is ideal if possible. For instance, note how in Virginia the Shenandoah Valley fits almost perfectly into a single district. Even though this 6th district is quite long, keeping this distinct cultural and geographic region of the state undivided is preferable to neat, square districts that would divide it.
Proposed non-partisan Pennsylvania congressional districts.
Another example of a community of interest can be seen in the Philadelphia suburbs. Bucks County is historically left undivided, but it makes more sense to split it due to differences in socioeconomic class. Rather than have Montgomery and Bucks counties each comprise most of a single district, a north-south division creates one wealthy suburban/exurban 8th district and a 13th district that is much more middle and working class. This iteration is more geographically compact, even if it ignores the actual border length geometry.
Ultimately, creating districts based solely on mathematics serves no rational purpose and just locks in a gerrymander without the name. We should want districts that give voice to the distinct groups of people in our country. Drawing them based on communities of interest, minority rights, municipal integrity, and geographic proximity is how to achieve that end in a non-partisan manner.
Comments are closed on this story.