Measuring Gerrymandering’s Effect on City Representation

City populations are predominately liberal, which means Democrats sort themselves into dense urban areas, and packing and cracking techniques exploit this fact to gerrymander.

It’s difficult, however, to quantify how cities suffer under a gerrymandered map: district lines don’t obey city borders, city populations could be combined with other populations for legitimate reasons, such as promoting competitive elections, and Chen and Rodden’s unintentional gerrymander makes a strong case that even good faith mapping efforts could still disadvantage city constituents.

Despite these difficulties, cities demand close scrutiny when assessing gerrymandered maps.

Cities — defined here as Metropolitan Statistical Areas — present clear communities of interest for the purposes of redistricting, being defined by the Census Bureau as urban areas plus surrounding counties with “a high degree of social and economic integration” with the urban core.

There’s legislative appetite for preserving communities of interest too, with 24 state legislatures mentioning this criterion in their laws or constitutions.

And, most obviously, we should be concerned about the Congressional representation of such socio-economically cohesive portions of the electorate that comprise 85.

3% of the US population.

¹Popular measures for identifying gerrymanders in the current literature — such as asymmetry of seats-votes functions and sampling / distributional approaches — are laudable approaches for identifying maliciously-drawn maps, but they’re weaker at isolating harm done to one geographic subset of the state population.

And while some traditional redistricting goals, like compactness and preserving county borders, correlate with the notion of community of interests, the benefits of these approaches are usually evaluated in terms of preserving state-wide symmetry or proportionality (where the state-level vote for each party translates to a proportional amount of seats in the legislature, i.


if 60% of a state’s residents vote for Democrats, 60% of the legislature — or thereabouts — are Democrats).

What’s lacking from these is a measure that drills down into the effect of gerrymandering on geographically concentrated populations within states.

Representational Harm of CitiesThis post offers up one such measure, namely a quantification of how gerrymandered maps inflict what I’m dubbing “representational harm” on cities.

Here’s the main idea: let’s consider representational harm for a city to occur when its constituents are of one ideological persuasion but its representatives are solidly of the opposite ideology, and then let’s compare some coarse measures of political beliefs of city constituents with those of their representatives, looking at how this varies between gerrymandered and non-gerrymandered states.

Doing so yields the comparison below:Here, negative / positive scores indicate liberal / conservative, with the vertical orange line highlighting a score of “0” — the ideological center.

The measures stem from a 2014 Pew Survey for city constituents (many thanks to my friend David McClendon for pulling the Pew data together) and DW-Nominate scores of the 113th Congress for representatives.

Each data source was standardized to approximate an apples-to-apples comparison.

² Red lines indicate that the city is liberal but the representatives are, on average, conservative, and blue lines indicate the reverse.

The more glaring cases of this reversal constitute representational harm done to city constituents, under any common theory of representation.

³ Austin is solidly liberal, for example, but it is represented by very conservative voices in Congress.

The division of states into “gerrymandered” and “non-gerrymandered” divides those that exceed a 7% efficiency gap threshold from those that do not, with some minor qualifications.

⁴ Admittedly, an efficiency gap greater than 7% does not guarantee a map is gerrymandered —it improperly flags proportional maps as gerrymanders, for example— but it’s a useful proxy here.

There’s strong circumstantial evidence bolstering this categorization too, since every state categorized as “gerrymanders” has faced serious legal challenges.

Two of them —North Carolina and Virginia — were deemed racial gerrymanders by the Supreme Court and two others — and Florida and Pennsylvania — were overturned by their respective state supreme courts on the grounds that they violated their state constitutions.

Texas, meanwhile, came very close to a Supreme Court loss at the congressional level, and four others — Ohio, Maryland, Michigan, and Wisconsin — have been challenged on partisan gerrymandering grounds.

⁵So, accepting I’ve identified them correctly, these graphs suggest that gerrymandered states are more likely to have differing polarity between city constituents and their representatives.

Cleveland, Austin, and the DC suburbs of Virginia stand out as liberal cities whose representatives nonetheless are conservative.

Baltimore, while just barely conservative, has extremely liberal representatives.

Most cities have a substantial ideological gulf between their representatives and constituents, but the gerrymandered states seem particularly bad with the wholesale ideological reversal between the two.

This, combined with the magnitude of these reversals, is exactly what “representational harm” looks like.

Limitations and DifficultiesThis measure can by no means replace existing analytical tools used to detect gerrymandering in the first place.

Representational harm could arise under many circumstances: restrictive election laws, low voter turnout, malicious redistricting plans, among other causes.

The existence of representational harm doesn’t prove gerrymandering occurred.

Instead, the most we can assert here is: there’s a correlation between states broadly construed as gerrymanders and representational harm of their cities, not just the state as a whole.

The scope of this measure is simply to quantify the effects of gerrymandering more concretely for a specific subset of state populations.

There are technical problems to consider as well.

We lack DW-Nominate standard errors, so statistical testing is beyond the scope of this post for now.

I’d like to test that the difference between representative and constituent ideology is statistically significant, but we could also use standard errors to test other potentially significant differences.

For example, the average absolute difference between representatives and constituents in gerrymandered states is 0.

497, and it’s 0.

365 in non-gerrymandered states, suggesting that gerrymandering causes greater ideological discrepancies, but we can’t have much confidence in this without standard errors.

There’s also room for debate as to how well each of these data sources tracks the same underlying ideological trait.

Can roll-call votes of representatives really be compared to survey responses of constituents?.This requires assuming that each of these measures indirectly quantify the same ideological beliefs — certainly something up for debate, and I’m more than happy to use alternate proxies for ideological belief in the future.

The other, more theoretical problem facing this measure stems from the concerns listed initially in this post.

Congressional districts frequently span strange segments of the state population, where groups of common geographic areas — like counties, precincts, cities, etc.

 — are not necessarily coextensive with congressional district borders.

Congressional representatives, especially within gerrymandered districts, “represent” no clearly cohesive subpopulation within a state.

So, focusing on cities is inherently unfair to the population of constituents outside city borders, though a similar error would occur if you chose to focus on any common geographically-based demographic within a state.

I’ve chosen to focus on cities, and the claim stands that these gerrymandered maps have detrimental effects on city representation.

Still, someone could try to compare non-city constituent belief with their representatives, which might reveal a different instance of representational harm.

Finally, all of this should be taken into consideration among other traditional redistricting criteria, such as compactness, majority-minority districts, and competitiveness.

Creating majority-minority districts may require some cities to suffer representational harm, but that also bolsters the representation of marginalized state subpopulations.

Maps emphasizing compactness (or preserving communities of interest) could ensure adequate representation of city constituents, though they could also waste Democratic votes in inadvertently packed districts, or produce a relatively non-responsive Congress (meaning significant changes in voting behavior do not translate to significant changes in representation).

And competitiveness might require contorted districts that extend far beyond city borders, but could accomplish other desirable democratic ends at the state level, such as a responsive Congress, more accurate state-wide representation, and fewer wasted votes across both parties.

We have some concrete examples of these conflicts too.

San Antonio stands out in the graphs above for representational harm done to its relatively conservative constituency, and yet, according to 538’s Atlas of Redistricting project, five of the six districts covering San Antonio are majority-minority districts — the highest such percentage of any city considered in this post — and one is competitive.

This compares starkly with Austin, with only two majority-minority districts out of seven total, and no competitive districts among them.

Arizona significantly “cracks” Phoenix too, but half of its congressional districts serve greater ends, with two majority-minority districts and two competitive ones.

We can see this phenomena in a map pretty clearly:⁵What’s most ironic about the Arizona case is that it’s actually got one of the closest correspondences between constituent and representative ideologies, despite having a map where the combined geographic coverage of districts cutting into Phoenix’s borders nearly spans the whole state.

This goes to show that it may be possible to balance several redistricting priorities at the same time, and really drives home the Supreme Court’s ruling in Shaw v.

Reno that oddly shaped districts serve only as an indication that gerrymandering may have occurred.

Strange shapes — even those dissecting cities — aren’t proof of malice.

⁶ConclusionContrasting the ideological leanings of city constituents with their congressional representatives gives us a rough measure of the representational harm done by gerrymandering to cities, but it’s currently limited by (1) technical data difficulties, like a lack of standard errors and (2) theoretical difficulties surrounding any analysis of geographic subpopulations inherent in the redistricting process itself.

City representation may also need to be sacrificed for other legitimate redistricting goals and compact districts tightly surrounding cities aren’t the only way to get good city representation.

With these points in mind though, this hopefully serves as an initially promising quantification of the representational harm done to cities via gerrymandering.

I used the estimated 2012 Metropolitan Statistical Area populations and divided the sum of this by 313,914,040, the 2012 estimated total population.

Some technical accompaniment for these graphs and data sources: Sampled city constituents in the Pew study responded to four questions tracking political ideology, which were averaged for each respondent, standardized based on the full sample mean and standard deviation, and then averaged for each city.

The first and second dimensions of DW-Nominate scores were combined for all representatives, again standardized at the National level, and then averaged across any representative whose district intersected at least in part with city borders.

This standardization makes the two measures roughly comparable; think of them as kinds of z-scores that are not necessarily normally distributed.

Pew’s survey design sampled from Metropolitan Statistical Areas — our “cities” here.

These MSAs can span multiple states, and have been divided into their state-level parts where appropriate.

The DC MSA, for example, includes parts of Maryland and Virginia, so two distinct rows in these graphs represent these different portions of DC.

I don’t want to get too sidetracked here, but for those interested, theories of representation ask the more philosophical question of how (or how should) representatives stand-in for their constituents.

The two dominate theories claim representatives should be either a trustee or a delegate.

The trustee theory of representation says constituents hire representatives to make decisions in their stead, even if some of the “trustee’s” decisions are contrary to their beliefs.

The delegate theory says representatives should mirror the beliefs of their constituents and act in a strictly faithful manner in line with the majority of their base.

My point here is representational harm, as I’m defining it, occurs under either of these theories, though the trustee theory is perhaps less clear.

Still, I’d say this kind of harm is closer, metaphorically, to a lawyer who concedes their client’s guilt to a jury against the client’s express wishes (which, by the way, they can’t legally do, thanks to a recent Supreme Court case).

I’m using the New York Times analysis of the states with an efficiency gap of more than 7%, but this is from October 2017, and I’m using 2014 data.

Georgia, Virginia and North Carolina have drawn new maps since 2014 , so I calculated the efficiency gap for the 2014 election of those states, ignoring third party candidates and uncontested seats.

Georgia is 11.

4, Virginia is 8.

5% and North Carolina is, uh, 20.

8%, all pro-Republican.

I’ve also included Florida, since their 2014 map was explicitly overturned by their state Supreme Court as a state-wide partisan gerrymander, violating the Florida Constitution, even though its efficiency gap score is under 7%.

For those interested, the general map is here.

The highlighted districts here are those with at least some portion within one of the cities considered in this post.

The colors indicate if the representative is liberal or conservative, and the lighter / darker shading indicates if they are above or below the median within their respective liberal / conservative grouping nation-wide.

So, a darker blue means that representative is in the more-extreme-half of all liberal representatives, lighter red means that representative is in the less-extreme-half of all conservative representatives, and so on.

The dashed lines delimit the boundaries of all districts, and the green lines show the borders of the city cut by state.

Note too that the converse is not true: the existence of a gerrymander does not imply that there must be oddly shaped districts.


. More details

Leave a Reply