Garrett Dash Nelson is the incoming curator of maps and director of geographic scholarship at the Norman B. Leventhal Map & Education Center at the Boston Public Library.
Using a score he named after the inventor of gerrymandering, one geographer discovers a huge variation in how well congressional districts match up with commuter regions consisting of interconnected urban, suburban, and rural areas.
Everybody knows that gerrymandering is bad because it unfairly stacks the political decks. In addition to the lopsided electoral outcomes, a gerrymandered map is also objectionable because crazy, mangled voting districts in the shapes of corkscrews or tweezers don’t correspond at all to the relevant geographic units in which we actually live. People have emotional and political attachments to all sorts of geographic entities: jurisdictions like states and cities as well as culture regions like the Bay Area or Appalachia. But who ever introduced themselves a proud resident of NH-02, or got a tattoo with the outline of TN-03?
To quantify the political unfairness of gerrymandering, we can count up partisan demographics and look at election results. However, it’s much harder to test how well or how poorly a map matches with the patterns of human geography. Although many states require the districting process to respect so-called “communities of interest,” how to define and demarcate these communities remains an open question, and so these requirements have rarely exerted much legal force.
One way of tracing the geographic structure of communities is to look at the networks that stitch places together. And one type of network that is both particularly important as well as relatively easy to measure and map is commuter patterns. In the study of networks, a “community” is defined as a group of objects that have a relatively dense set of connections among each other, and a relatively weak set of connections beyond. Although that definition is much thinner than our social and political concepts of “community,” it does have the advantage of being measurable.
Measuring the community structure of a network versus the assigned communities of a voting districting can therefore offer us one possible method of checking whether or not an electoral map preserves “communities of interest.” A “good” map, seen from this perspective, would keep areas that are connected by commuter webs together in the same district as much as possible. A “bad” map, by contrast, would slice up commuter regions, fracturing cities and confecting electoral districts out of places that have relatively less connection with one another.
Network scientists have developed a statistic called “modularity” which measures how well a proposed partitioning matches with the structure of connections within a network. In this case, you can think of it as a ratio between the number of commutes which stay inside of the borders of a single district versus the number of commutes which cut across district lines. In honor of the namesake of gerrymandering, former Massachusetts governor Elbridge Gerry, I’m calling it the “Elbridge score” (Electoral Boundary Resemblance to Identifiable Geography).
I used a data set of millions of commutes—the same one that my colleague Alasdair Rae and I used to redraw the map of the U.S. commuter “megaregions”—to compare congressional maps. What I found was a huge variation in terms of how well congressional districts match up with commuter regions consisting of interconnected urban, suburban, and rural areas. The five best states—Kansas, Indiana, Iowa, Kentucky, and Tennessee—all have a fairly even spread of population and medium-sized urban centers that are distributed at a distance from one another. This makes some intuitive sense. Because the population of a U.S. Congressional district is standardized at around 700,000 people, the easiest states to divide into fair districts will be the ones which are already roughly split up into commuter zones of roughly 700,000 each.
As an example, the map below show’s Indiana’s commuter flows (only the ones that both begin and end within the state) as well as the state’s congressional districts. You can see that each district encloses a fairly self contained set of commutes, usually centered around one of the state’s major cities. Commutes are colored purple if they both begin and end in the same district, but orange if they cross a district—here again, you see a map that does a pretty good job at keeping commutes within district boundaries.
In contrast to these fairly evenly-divided, mostly Midwestern states, the five worst states—Rhode Island, Hawaii, New Hampshire, Idaho, and Nevada—are all characterized by heterogeneous geography and difficult-to-contain urban clusters. For example, the border between Rhode Island’s two congressional districts cuts right through the middle of the city of Providence. In Idaho, downtown Boise lies in the second district, but most of the city’s populous western suburbs lie in the first district. In these cases, as the map of Rhode Island shows, a huge fraction of commuters live and work across district lines.
So why aren’t the states that are the most notoriously gerrymandered, like Pennsylvania and North Carolina, at the bottom of the list of Elbridge scores? Because states in the U.S. are so different from one another—some are big, rectilinear, and evenly spread out, while others have off-kilter concentrations of population and urban development—it’s difficult to compare this score across different states. It’s more useful, then, to test multiple proposed maps in one state against each other. By keeping the underlying geography of a state constant, we can see how well—or how poorly—a proposed map fits a state’s “communities of interest” defined by commuter connections.
For example, the difference in Elbridge scores between Pennsylvania’s current gerrymandered map and the new map which the state Supreme Court handed down in February suggests pretty convincingly that the current map cuts the state up into areas that don’t match very well with commuter-based communities. The gerrymandered map scores just 0.58; by contrast, the state Supreme Court’s map scores 0.65, and another proposed fair map scores 0.66.
Another way to show this same measurement on a diagram rather than map is by making a matrix of a state’s electoral districts, and counting the number of commutes between each district at each of the cells. In a perfect (and impossible) districting scheme, all of the commutes would lie along the diagonal of the matrix, since they would begin and end in the same district. The two figures below show how this looks in two states that each have four congressional districts: Nevada, one of the worst scorers, and Kansas, one of the best. The number at each cell represents the percentage of the state’s total number of commutes. In Kansas, nearly everybody lives and works in the same congressional district. In Nevada, by contrast, a lot of people live in one district and work in another—a consequence of the fact that job-rich Las Vegas lies in its own NV-01 district, while the city’s suburbs lie in NV-03 and NV-04.
It’s important to point out that there’s a big conceptual leap between commuter connections and fully-constituted “communities of interest.” After all, commuter networks don’t count people who are out of the workforce. And they completely miss many of the most significant connections that tie places together, like friendships, family networks, cultural ties, or ecological flows—not to mention the many more subjective, unmeasurable forces that make places and political communities feel like they cohere together.
Trying to match electoral districts to the lived geography of “communities of interest” runs into some complicated choices about what we value. If you’re interested in reforming housing markets or redrawing the jurisdiction of a transit authority, following the geography of commuter areas makes a lot of sense. If, instead, your main goal is to balance out partisanship at a federal level and make elections more competitive, then it doesn’t necessarily follow that districts should follow functional geography. In fact, the opposite might well be true, since Americans increasingly cluster together in politically-likeminded geographies.
For instance, which scenario would be preferable? To vote in a geographically-coherent district together with lots of people with common interests, or in a district that doesn’t make any geographic sense but has an even mix of Democrats and Republicans? To answer that question, you have to consider a lot of other values, like: Are our representatives supposed to represent place-specific communities (in which case matching district boundaries to human geography is crucial) or do they just represent the interests of a national political party (in which case preserving local communities might be much less important)?
That’s an ideological and philosophical problem which most of the debate on gerrymandering has skipped over. We may know what a bad electoral map is like, but a good one is much harder to define.