Princeton Election Consortium

Innovations in democracy since 2004

Outcome: Biden 306 EV (D+1.2% from toss-up), Senate 50 D (D+1.0%)
Nov 3 polls: Biden 342 EV (D+5.3%), Senate 50-55 D (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

North Carolina’s new House plan still has at least half the partisan skew of the gerrymandered map

September 14th, 2019, 3:31pm by Sam Wang

NC House remedial map as amended 13 Sep 2019(Links: shapefiles, PlanScore, spreadsheet of analytics, and DRA)

A three-judge state court in North Carolina has ordered that both House and Senate legislative maps, which it identified as a partisan gerrymander, must be redrawn by next Wednesday. Last night the House took a step toward meeting that deadline – but the handiwork so far (shown above) still contains enough partisanship to raise an eyebrow.

On Friday night, the state House voted on a remedial legislative map to oversee its own members’ districts. To become law, it also has to pass the Senate. (Under North Carolina law, the governor has no say in the plan, making the state unusually vulnerable to single-party manipulation of the redistricting process.) If passed, the plan will go to a court-appointed special master for evaluation – and possible redrawing, if he finds that the General Assembly’s map is insufficient.

Here at the Princeton Gerrymandering Project, we have noticed algorithmic biases in the process used to generate the remedial map. Using the engine and additional analysis, we furthermore find that the map still contains between one-half and two-thirds of the partisan advantage that was present in the illegal gerrymander.

The first step in drawing the remedial House map (as well as the Senate map, which is not done yet), was to select it on Thursday, bit by bit, from a lottery machine. This act of showmanship caught people’s attention and was criticized. But the showmanship hides a deeper problem. The real issue is whether the balls that were put into the lottery machine were fair in the first place. To make an analogy to card games, a stacked deck can’t produce a fair deal. And I have some questions about the deck.

First, let’s consider the remedial map as initially drawn by lottery, which I will refer to as “House basemap 12 September”:House initial basemap 12 September
(Links: shapefiles, PlanScore, spreadsheet of analytics, and DRA)

This remedial map is a composite of seven maps obtained from Jowei Chen, a political scientist at the University of Michigan who served as an expert witness in the case. The maps were drawn according to an algorithm programmed by him, selected from a group of 1,000 randomly-generated maps on the basis of overall statewide scores of geometry. Why seven? The House committee used one map for each of the seven county clusters that the Court identified as illegally drawn.

Chen’s algorithm chugs along, building districts from chunks of counties and cities. It doesn’t ask whether communities of interest are kept together. It produces a few thousand plans, which pales in comparison with faster and more sophisticated map-drawing approaches that explore millions or billions of possibilities (see Auto-Redistrict, as well as work by Wesley Pegden, Kosuke Imai, Moon Duchin, Jonathan Mattingly, and others). However, Chen’s maps have the advantage that they were used by the Court to create a baseline of expectations for what a blind procedure would do.

But a blind procedure can contain algorithmic bias, a well-known risk in modern computing. This automatic drawing procedure tends to favor Republicans, not only because cities are packed with Democrats, but also because cities vote more lopsidedly for Democrats than rural areas vote for Republicans in rural areas. So a the algorithm tends to pack Democrats.

Humans could easily avoid this pitfall – which one might think the Court wanted the legislature to do. Indeed, the Court instructed the legislature to draw lines from scratch. An algorithm with hidden biases most definitely does not start from scratch.

In a map biased by the algorithm’s rules, a majority vote for Democrats could easily elect a majority of Republicans. That only happens in close elections – but North Carolina is a closely divided state, and closely divided states are where partisans can gain the most by committing a gerrymander.

Here is how the plans perform according to PlanScore’s model based on 2016 Presidential vote and 2018 state House elections. (PlanScore’s model gives a statewide vote of 52% Republican, 48% Democratic; I have corrected the seat count to reflect an equally divided 50%-50% vote).

Broadly speaking, the remedial plans are an improvement over existing maps. But as measured by PlanScore metrics, one-half to two-thirds of the 2018 plan’s partisan asymmetry remains. Or to put it conversely, partisan gerrymandering considerably amplifies the natural effects of geography.

But the average effects of blind geography can also be prevented. Think back to the remedial Congressional map put into place in Pennsylvania in 2018, in which the gerrymandered Congressional delegation was 13 Republicans, 5 Democrats, but the redrawn map elected 9 Republicans and 9 Democrats – and was still compact in appearance and pleasing to the eye.

There is another problem: The map chosen by the General Assembly might not look after minority representation. In North Carolina, blacks are a substantial minority, comprising 22% of the population. In such a significant population, it might be desirable to make sure they have adequate opportunities to elect legislators that reflect their priorities.

As a rough measure of opportunity-to-elect districts, I counted the number of districts with at least 40% black population and at least 55% expected Democratic vote. (A proper analysis requires a more complex analysis called ecological regression.) The 2018 map contained 21 such districts. The remedial map started off with 20 opportunity districts, and after amendment went up to 22 districts. One would have thought that undoing the partisan gerrymander would have led to more opportunity districts for black voters.

Drawing a map that achieves partisan and racial fairness, and is also compact and meets other criteria, is possible. But it requires considerable skill and thought from those who draw the lines. The House and Senate committees didn’t draw the lines themselves, as instructed. And this House plan is still pretty partisan. So I would not be surprised if the Court stepped in to draw its own map.

On the other hand, this map does not deviate from the middle of the pack of Chen’s simulations, which also show a pro-Republican bias. So in a literal sense, this plan would not set off the alarms that were defined in court. So that map could be accepted – a partial victory for Republicans.


One puzzle is why legislators didn’t add even more partisanship back into the remedial map. After all, North Carolina politics is famously partisan and bitter. On Wednesday, Republicans staged a budget vote during ceremonies commemorating the 18th anniversary of the 9/11 attacks, in violation of a promise they made to Democrats. So they do not lack for nerve.

One answer may be that partisan gerrymandering is a team sport, but with the line-drawing process done in the open, legislators couldn’t collude. They couldn’t do as the late redistricting genius Thomas Hofeller did: pack Democrats extra tight in a few districts while spreading Republicans more thinly across a much larger number of districts. By drawing lines in open committee, legislators were reduced to saving their own skins, as evidenced by this video of a legislator attempting to draw himself a safe district:

He was caught – and soon thereafter announced his retirement when his current term ends. Even if he had succeeded, an every-man-for-himself mentality would not increase the number of wins.

Anyway, we will be able to tell what happens by scoring plans as they emerge. That’s because of release of geographic data. Data openness allows citizen groups and us to score plans. That’s why projects like can play a critical role in the redistricting cycle ahead.


P.S. Here is the overturned gerrymander:

Tags: Redistricting

2 Comments so far ↓

Leave a Comment