North Carolina 2020 Redistricting

[Sorenroy]

I've hit a bit of an impasse when trying to draw state house districts. The state uses county groupings to theoretically help prevent gerrymandering and ugly, stringy districts. If a county can fully contain one or more districts within it, it cannot be divided. Counties that are too small or are outside the acceptable bounds for districts (for example, a county with 150,000 people is too big for one district but too small for two) are put into county groupings, trying to maximize the number of two-county pairs, then maximizing three-county pairs, ect. However, I don't know how to actually go about doing this. The blue, non-contiguous counties are counties that fully contain their districts, all the other colors are possible pairs that contain one or more pairings. I've tried going by hand and seeing what possible combinations of the 31 possible pairings work without overlapping or otherwise breaking the system, but this is both time consuming and inefficient. Does anyone know of a way to test what works and what doesn't without going one-by-one through the possible options? I've been enjoying messing around with what I can, but this just becomes tedious.

A "Stephenson" explainer

[Sorenroy]

I wanted to see how badly I could gerrymander the NC State House for the Republicans while abiding by (most of) the redistricting criteria of the legislature.

• Equal Population: The largest district is 4,335 people above the ideal district, or 4.98% and the smallest district is 4,341 people below the ideal district, or 4.99%.
• Contiguity: All districts are fully contiguous by the criteria's definition, i.e. some are only contiguous by thin strips of water
• Counties, Groupings, and Traversals: All districts in this map are drawn with the consideration of the Stephenson ruling. That is, the number of county splits is minimized as much as possible by creating pairs or sets of counties when counties are unable to fully house districts within themselves. I ended up using Duke's set of optimal county clusters after finding that doing so on my own would be unfeasible (without software built for this kind of thing). In instances where multiple county clusterings were optimal, I used the clustering that best fit the other criteria. Additionally, as per the rules, if any county was large enough to fully encapsulate a district by itself, it is done. This did result in several snaking districts used to absorb the extra population of counties only slightly over the population limit in some instances, which would be unavoidable even under an ungerrymandered system. I did not consider double bunking at all, so no accommodations were needed to allow for it here.
• Racial Data: Racial data was not used in the creation of these districts.
• VTDs: The criteria doesn't really lay out what the voting districts, or VTDs, are that should only be split when necessary. I interpreted this as requiring minimal precinct spitting. Across the whole state, only one precinct is split, and this is out of necessity. Including it in either possible district means making one district smaller than the 5% limit imposed by the equal population requirement.
• Compactness: Without the ability to run some type of algorithm the sift through all the possible districts and come up with the most compact option, selecting for compactness was just unfeasible. I would have liked to do more with this, but it's just not something I could have done.
• Municipal Boundaries: In areas where Republicans already beat the gerrymandering criteria and nothing could be done to gerrymander further in their favor, I did my best to maintain municipalities. In the case of municipalities too large to fit into one district or municipalities that interlocked with others to create a cluster of precincts too large to fit into one district, I did my best to minimize (1) the number of municipality splits and (2) minimize the number of districts any municipality was split into. I tried to maintain larger municipalities when possible, but if breaking one larger municipality would maintain two smaller ones, I used the raw number of splits as the deciding factor. If there were multiple ways to divide districts with minimal splits, I then went back to making each district as Republican as possible.
• Election Data: As I fully expect the legislature to do, I ignored this criteria. Given that the point of this exercise was the gerrymander for the Republicans, not looking at the Republican vote share was not doable.
• Member Residence: Something that I already think is a stupid rule that only works to protect incumbents elected in previously gerrymandered districts, I wasn't going to weaken my map just to protect them in this hypothetical. Any districts that do serve to protect incumbents are purely happenstantial.
• Community Consideration: Given that this requirement is already explicitly made secondary to the other goals of redistricting, and given that I have no way of getting community input for a hypothetical map, there really is no community consideration given here.

Since there is no generally accepted level for what is considered safely gerrymandered, I figured that a 60-40 split for the national environment as established by the Popular Vote Index would be a decent place to start. Abiding by Duke's county clusters and not splitting precincts, this is the best I could come up with. I will continue using Abdullah's formatting for things until it is no longer the most elegant way to show this type of information.

The Population Deviation is 9.97%, and it reflects the 2020 Census.

54/100 on Dave's Proportionality Index
39/100 on the Compactness Index
100/100 on County Splitting
71/100 on the Minority Representation index
15/100 on Dave's competitiveness index

The map above shows results from the 2020 U.S. Presidential Election.

DRA Link.


Partisan Breakdown by Election

2016 Gubernatorial Election: 74R to 46D

2016 Presidential Election: 75R to 45D

2020 Gubernatorial Election: 69R to 51D

2020 Presidential Election: 73R to 47D

2016/2020 PVI: 75R to 45D


While I didn't go into this attempting any specific seat distribution, it did end up being the case that the 60th seat is R+10 PVI. With it being this close, I decided to make a quick edit to make a Republican majority as safe as possible. The only change that needed to be made was in the Columbus-Robeson county pair. The PVI of both counties combined is R+9.82, but in order to bring one of the two districts within the pair over R+10, the weaker set had to be lowered to R+9.65. The change below (following the same set of constraints and goals listed above) represents two almost identical districts with only one municipal split in Fairmont that only cuts across one unpopulated block. This changes the districts to be R+9.81 (blue) and R+9.84 (green).

DRA Link.

Beyond that, it's worth looking at the one necessary precinct split. The 18,203 person Havelock precinct in Craven county sits between the 78,616 person block of District 3 (in purple) and the 71,587 person block of District 13 (in brown). It is the only precinct between the two sections and thus is forced to be divided so that both districts can hit their minimum allowed population. Unfortunately, 16,621 of those 18,203 people live in the town of Havelock itself, so it isn't even possible to keep the precinct's only municipality whole. I ended up deciding to split the precinct down the Croatan National Forest on the left, and then split the city down Slocum Creek, the river that runs through the center of the town.

However, without detailed data on how the different blocks vote, it is technically possible that the land contained within District 3's boarders houses all the Democrats in the district and none of the Republicans. In this hypothetical, District 3 could drop as low as R+9.02. The workaround is to snake District 3 through the upper part of the precinct and into its neighboring Harlowe, the only precinct in Craven county fully withing District 13. By doing this, you can still split one precinct while guaranteeing both districts remain above R+10.

With both of those changes made, the 61st district lies at R+9.81, creating a safely gerrymandered State House that could easily withstand even a 2008 Obama-level national environment.

[Sorenroy]

Here is a second draft of a Republican gerrymandered NC State House. For this map, I decided to prioritize minimizing city (as defined by Dave's Redistricting App) splits above gerrymandering for districts. As such, while this map uses the same guidelines as my previous post, I am somewhat revising this point:

• Municipal Boundaries: In areas where Republicans already beat the gerrymandering criteria and nothing could be done to gerrymander further in their favor, I did my best to maintain municipalities. In the case of municipalities too large to fit into one district or municipalities that interlocked with others to create a cluster of precincts too large to fit into one district, I did my best to minimize (1) the number of municipality splits and (2) minimize the number of districts any municipality was split into. I tried to maintain larger municipalities when possible, but if breaking one larger municipality would maintain two smaller ones, I used the raw number of splits as the deciding factor. If there were multiple ways to divide districts with minimal splits, I then went back to making each district as Republican as possible.

Now, the hierarchy of goals is as follows:

• Minimize County Splits: Both this and the previous map followed this rule. Both maps abide by Duke's set of optimal county clusters, keep districts fully within counties when population allows, and minimize the points at which counties are split when need be. While it is not explicitly written that a county couldn't have multiple tendrils pull from it if it already needed to be split, I have seen no drafts, NC or otherwise, that allow for this. As such, while I don't necessarily see any value in contiguous precincts within a county if a district is otherwise connected, I also see no reason to violate that "rule."
• Minimize "Voting District" Splits: Again, this is something that I also abided by in the previous map. As before, I avoided splitting precincts to make this map. While I made mention of the single necessary split in between Carteret and Craven in the previous post, it is actually implemented here.
• Minimize Municipality Splits: The third criteria used, and the goal of this map, was trying to minimize city splits. Without breaking up precincts, this can often become difficult as even smaller "cities" can reach across several highly populated precincts or can share precincts with other cities causing chains of cities to run together. As such, unlike splitting counties or precincts, splitting cities was unavoidable. While I did not attempt to split cities at a minimum number of points, as with county or precinct splits, I did try to minimize the number of districts each city was split into. I split as few cities as possible and split cities as little as possible if required.
• R+10 Gerrymandering: Again, there is no definitive definition of what makes a safe district. For the purposes here, after minimizing county, precinct, and city splits, I tried my best to maximize the number of safe Republican districts on the assumption that a Popular Vote Index score of R+10 would probably be enough to keep almost any district safe.
  
  
  
  The Population Deviation is 9.98%, and it reflects the 2020 Census.
  
  65/100 on Dave's Proportionality Index
  42/100 on the Compactness Index
  99/100 on County Splitting
  74/100 on the Minority Representation index
  22/100 on Dave's competitiveness index
  
  The map above shows results from the 2020 U.S. Presidential Election.
  
  DRA Link.
  
  
  Partisan Breakdown by Election
  
  2016 Gubernatorial Election: 73R to 47D
  
  2016 Presidential Election: 76R to 44D
  
  2020 Gubernatorial Election: 62R to 58D
  
  2020 Presidential Election: 70R to 50D
  
  2016/2020 PVI: 77R to 43D
  
  
  Given the restriction that the each district minimize splits of all kinds, there is much less flexibility in how much different areas can be gerrymandered and much less flexibility in shoring up the 61st seat. As such, there is no improvements that can be made to make the 61st seat more favorable to the Republicans, at least under a Popular Vote Index criteria. As such, while the map in the previous post was able to create a R+9.81 balancing seat, this map can only achieve a R+7.12. Additionally, while the previous map had 60 R+>10 PVI seats, this one only has 55.
  
  That does not mean that this was an unsuccessful gerrymander, just that it's a weaker one. Cooper, even with a 4.5% win in 2020, still lost the 60th seat by a margin of 3.65% and the 61st seat (important because the Republicans won the Lieutenant Governor's seat) by 3.99%.
  
  
  There are three things I want to highlight in this map. The fist, very briefly, is a set of unnecessary splits. Both the Anson-Union and Harnett-Johnston county pairs had municipalities split where unnecessary given the secondary criteria of the first map and the primary criteria of this one. As such, while the lines of those county pairs changed in this map, they are better than the ones given in the first map and would apply to that map as well.
  
  Second, another mistake made in the first map was in Cumberland county. As a bit of a peak into my personal process, the best way I know how to make these maps is to find all the precincts in a given range of PVI and then stitch them together to make a full safe district. Sometimes I go back in and try to find deviations in turnout to get population balances right, but most of the time that just happens on the margins. In this specific instance though, it was something I did not catch until coming back and making this second map. There is a precinct in Cumberland county called G11B that has 23,496 people, 18,738 that are voting age, but only had 1,846 people vote for president in 2020. My best guess as of why is that the precinct houses the Pope Army Airfield, but I haven't noticed anywhere else where a military base has its population catalogued as living there but where the forces don't vote in the state. Additionally, I don't think it's a mistake. The State Board of Elections shows the same precinct boundaries and the same number of votes for each candidate in the precinct. And, while there is no real primary source for population by voting precinct, the districts don't make sense population-wise unless G11B is counted the way DRA shows.
  
  
  
  DRA Link.
  
  Finally, now that the legislature has started to draft some of its own maps, we have a point of comparison and analysis. For now, lets consider Cleveland and Moore counties. They have populations of 99,519 and 99,727 respectively, and both share the characteristic of being roughly in the center of a multi-county cluster (Cleveland in a nine county, seven district cluster and Moore in a five county five district cluster). In the proposed criteria on redistricting from earlier this year is the following line:
  
  

  If a county is of sufficient population size to contain an entire congressional district within the county’s boundaries, the Committees shall construct a district entirely within that county.

  
  To me, this would seem to indicate that both counties should encapsulate one full district with the rest going to one or more of its neighbors in its grouping. As such, the districts in my map include snaking districts that look like gerrymanders given their odd shape, but are actually necessary to properly balance population. Perhaps the ugliest of these districts, 110, snakes through four different counties to collect excess populations. On the positive end, this means that communities within their own counties remain intact and represented by someone who owes none of their votes to other areas. On the other, that leaves mangled districts like the one below that are represented by someone that does not truly have roots in any major part of the community.
  
  
  
  The state legislature has opted to take their districts in another way. Instead of abiding by the rule to keep districts within single counties when possible, they have opted to minimize these stretched out districts. In both cases, rather than having a district collecting stray population on the edges of the map, they have opted to split the counties to be represented by different state legislators. In the case of Moore, this means taking a county that could have had its own district and splitting it into three separate districts making up about half (or less) of each of them. On the positive end, this means that there are no snakey, ugly districts connecting unrelated areas. On the other, that leaves counties that could have had their own representative split into multiple groups.
  
  https://twitter.com/lekhashupeck/status/1449100029076066304
  
  To me, the state legislature's idea is probably the better option of the two. Even split three ways, it might be the case that Moore is better represented, as it now has three different representatives rather than just the one and the partial one of the snake. For the people who would have been in the snaking district, it also means that their rep is likely more keyed into their concerns rather than dashing from place to place along the road. This is just to explain why these areas have such an odd difference between what my map shows and what the Representatives are coming up with.

[Sorenroy]

Statehouse minority leader is being moved from a Biden +22 to Biden +900 votes district.

https://twitter.com/ForwardCarolina/status/1450891182280691718

North Carolina R's literally pull a Mattis!

I think this tweet does not understand the redistricting process at play here in North Carolina. Given the requirements made around clustering counties, it is inevitable that Reives would no longer sit in a massively Biden-favoring district. Last cycle, Chatham county, a county Biden won by 11.5%, was paired with Durham county, home of the city of Durham. Durham gave Biden 80% of the vote to Trump's 18% and the three precincts taken from the county to bring Reives's district up to population all roughly split that way too. Now, after the 2020 Census, Durham was optimally matched with Person county to its north. Chatham is now in a five county cluster of which it is by far the most Democratic of those counties. Even if you were to try and create the most Biden district possible in the cluster while abiding by the assembly's own rules, Trump would still win it 49.6% to 49.1%, and that district wouldn't even include the Reives residence. In fact, the only reason that the current draft map gives Reives a Biden district at all is because the NC GOP, in violation of its own redistricting criteria, is splitting Moore county, something that shouldn't be done given Moore's population.

The NC GOP moderately used deviation to add an extra precinct to shift it from Biden +6 to Biden +1.  The rest of the districts in the cluster are relatively underpopulated while this one is nearly 5%(max legal) overpopulated. Really not the biggest gerrymander in the history of the world but it is funny to note.

Yeah. Rereading the thread on Twitter, I think my criticism is actually more on what you added in your post. The district no longer being Biden+22 or anywhere near that was inevitable. The extra work done by the NC GOP to press the population up to its limit as the other districts remain far smaller than needed is fairly blatant.

[Sorenroy]

Is there a program that can create every possible combination of whole precinct districts under a given population criteria?

I've been working on a full NC-Senate map ahead of the redraws that does not split any precincts (or minimizes splits if they are required) and stumbled on this: The Granville + Wake County Cluster is painfully small for its allotted six Senators. It has a total population of 1,190,402 and each district must have at least 198,349 people. Once each district has that number, there are only 308 people that can be shifted around to make the maps work. If I can't find a way to automatically sort through the junk districts that don't fit the population requirements, I'm looking at dozens of hours combing through precinct-by-precinct manually to see what works. Does anyone have any ideas here?

[Sorenroy]

Btw, didn't NC Democrats screw themselves sometime in the 90s or 2000s by taking away the governor's power to veto maps?

Not only did they do that, it was arranged by Roy Cooper.

Woof. These maps are disgusting and I don't support gerrymandering regardless of which party does it, but it's very hard for me to feel bad here.

The work that Cooper did in 1997 was a compromise with Republicans giving the governor veto power. Vetos on maps weren't taken away, they were never granted to the governor to begin with.

Either way, to say you don't feel bad because an action was taken 26 years ago is silly. There are people who were born and have lived there whole lives in North Carolina since that bill was passed. They had no say in Cooper's actions at the time, even indirectly through voting.