Mid-2014 county population estimates out tomorrow, March 26

[muon2]

As you know and I think concur CoI tends to be the squishiest of metrics for redistricting. However, it's clear to me that the public likes the concept. To the extent that I could quantify other groups that share a CoI, span counties, and are likely to have sufficient population to influence elections, I would do so. I've never thought that urban areas have a unique claim in that regard.

Projecting a BVAP percentage is quite a bit more difficult than projecting the population of a county. The Census doesn't estimate it and I have to use rolling 3-year or 5-year averages from the ACS to make a projection. That's why I only described the BVAP as being in the low 40s, upper 40s or over 50%. A more precise number awaits some more vacation time. Perhaps later this summer.

[muon2]

"That's why I only described the BVAP as being in the low 40s, upper 40s or over 50%."

Where did you do that above? I can't find it in a senior moment. The upper 40's is the most interesting for VRA purposes, where I am not sure it is entirely clear whether such a district must be drawn (assuming it's enough to elect a candidate of the minority's choice).  We know that's a floor where a higher percentage BVAP CD can be drawn, but I am not sure that we know any more than that.

I wasn't able to get one in the upper 40's for 2020. My whole county plan is described in the text as just over 40%, and I described the plan with the Birmingham connection as over 50%.

The MCC minus Barbour on the extreme east has a 2010 population of 473K and a 2010 BVAP of 55.7%. The quota for a CD was 683K so an extra 210K was needed from other counties. For example jimrtex which took the MCC, completed the Montgomery UCC then added Monroe, Pickens, and Pike to the to within 1K of the quota and have a 46.4% BVAP.

My projection for those MCC counties in 2020 is 439K, a loss of 34K. The target quota for a 7 CD plan is 708K which is an increase of 25K. The net is that instead of adding 210K more, in 2020 the MCC has to add 269K more. If whole counties are used that will pull that number down from its 2010 level. Using Tuscaloosa as a whole county and chopping the Montgomery UCC might bring it close to 45%. Even then the 2008 Pres result on DRA was only D+1.

[muon2]

I have taken the 7/1/2014 estimates in a spreadsheet and extrapolated them by county to 4/1/2020 assuming a uniform growth rate since 4/1/2010. I also want to preserve whole counties and keep UCCs covered with the minimum number of CDs. I'm not wild about the Chesapeake crossing, but at least there's a ferry connection, even if it's only in the summer. Without the peninsula split, I would have to introduce some chops in Hampton Roads.

This is what I get for a 12 CD VA:



Except for NoVa all CDs are whole county and within 1% of the quota. Richmond obeys both the cover and pack for its UCC. Norfolk and NoVa have the minimum cover. CDs 3 and 4 have 33% and 41% BVAP respectively.



The only chops beyond the 1% here are in NoVa. There is projected to be about 216K moved from Fairfax to CD 10 and the rest is split. With this split the political breakdown is 3R, 1r, 2e, 2d, 4D so the SKEW is D+2 and the POLARIZATION is 17.

[muon2]

My map tried to max the BVAP's, and they were at 41.5% and 45.5%. It's not legally mandated, but it is probably what will be done politically. I was unable to avoid a chop for VA-09, but maybe my figures are still off some. There are a lot of counties to count down there.

I guess I'd be a little concerned that the chops into both VB and Portsmouth look like race-based swaps at the expense of split counties/ICs.

Perhaps you haven't given as much growth to Roanoke and Blacksburg as I project in CD 9. I have Roanoke county (+Roanoke and Salem) at 224.9K and Montgomery (+Radford) at 120.7K. Together with Pulaski they are over half the CD. In any case I expect that these estimates will shift over the next half decade.

[muon2]

My map is a race based gerrymander. It's legal, because there is no packing.

I'm not sure I buy this statement. All race-based gerrymandering is illegal if it can be shown to be the predominant reason used to draw the district. It is a strict scrutiny test and race can only be used to advance a compelling state interest. Often the VRA provides that compelling interest, but not here. How do you justify it here, and is your plan narrowly tailored to meet your goal?

You must be logged in to read this quote.

I'm sure it would, but I'll be happy to use this as a counter-example to those who claim a plan drawn primarily to keep whole units of government will invariably favor the Pubs.

[muon2]

I don't like the erosity of your grey CD, but then we do have different philosophies about that. I am willing to chop when the erosity gets to be too much, absent a darn good reason to do otherwise. But then you already know all of this all too well about me!

Anyway, these maps are more about how legislatures interested in good government maps due to political power splits, or courts, will draw the lines, as opposed to following your unique rules (and up to a point mine). (In regard, the Jimtex definition of metro areas is totally unique, and I don't think most folks would view Stafford and Spotsylvania as part of the DC metro area.) Among other things, I suspect courts are more interested in "art" than you are.

Your art has always differed from mine. For example I find that your NW peninsula out of CD 5 and jut and chops around Richmond to add far more erosity than the clean L-shape of my CD 7. I do hope you are not abandoning the idea of measurable metrics to judge maps.

edit: The Census agrees with jimrtex that Fredricksburg and the two adjacent counties are part of the DC metro.

edit2: I think I found the math discrepancy. First the April to July shift is 0.25 of a year. Then the formula to get the annual rate is (estimate/census)^(1/4.25) - 1. The projection is census*(rate + 1)^10. I also had to correct for the fact that Bedford city was separate from the county in 2010, but they are together for the county estimate.

[muon2]

My eye tells me that this new version of 6 wouldn't be much worse than the old one for many conventional versions of compactness, and is probably about the same on the muon erosity scale. Since the new version reduces the erosity of 5 (and improves its compactness) it would be preferred by any system IMO. My guess is that it might get better still by rotating populations in Nelson, Louisa and Nottaway.

[muon2]

Counties in the MSAs are divided by the Census into central and outlying counties. The MSAs had too many rural outlying counties that made it in solely due to relative numbers of commuters in and out, not based on the density of populated areas. Some outlying counties had population centers that were deservedly in the metro, so using central counties only was unsatisfactory. We spent weeks haggling over the UCC definitions to balance those two points and come to a consensus agreement. I'm not going to abandon them to aid one map over another

Torie, what do you have for your projected VA CD now?

[muon2]

Counties in the MSAs are divided by the Census into central and outlying counties. The MSAs had too many rural outlying counties that made it in solely due to relative numbers of commuters in and out, not based on the density of populated areas. Some outlying counties had population centers that were deservedly in the metro, so using central counties only was unsatisfactory. We spent weeks haggling over the UCC definitions to balance those two points and come to a consensus agreement. I'm not going to abandon them to aid one map over another

Torie, what do you have for your projected VA CD now?

732,463.1096 per CD.

I am well aware of Jimtex's rationale. It even makes some sense in theory. But this exercise was not about maxing the Muon2 score, but rather what seemed like the best map that would fly politically, and keep both sides reasonably happy, while still being a good map. (This is more than a theoretical exercise if Gov. McAuliffe gets reelected, as seems more likely than not.) Your map runs the risk of jettisoning two black congressmen, including one existing one, and also violates the bridge or full time ferry rule (seasonal ferries do not count, and I checked that issue myself, before just abandoning the rule (not the law in VA obviously), due to other considerations, just as you did). Anyway, even under your rules, putting aside our mutual cheat on the ferry issue, I think I win on the erosity front, so it is not as if I am knocked out of the box from a pareto optimality standpoint. The bottom line is that your map will never, ever, be enacted into law, as I am sure that you probably agree.  Mine might.

Hmm. We are still off by quite a bit. I get 734,390 for a CD. If we can compare here are three counties in my 2020 projection: Appomattox 15,703; Arlington 255,785; Augusta 74,014.

I appreciate that you are trying to thread a political needle of sorts in VA. It seems to me that neutral criteria either recognize that a state has section 2 obligations or that it doesn't. If it does then the path is clear. But here, I am prone to agree with you that it does not have a section 2 obligation. How then does one construct a neutral map that still provides an opportunity for minorities to elect a candidate of choice? You don't seem to like my idea of MCCs as a CoI to balance UCCs - contiguous sets of counties (or subdivisions for chopped counties) that exceed a certain minority fraction (40% BVAP in this case).

VA at one point discussed going to a commission system. I remain convinced that commissions should have firm rules to guide their mapmaking. It is on a commission basis that I'm looking at the 2020 projections.

[muon2]

All the maps I've posted for 2020 in this thread have assumed work by a neutral commission. That doesn't mean I expect any of the states to adopt said commissions by then. It was a point of interest in some states going into 2010 to see what their states might look like if drawn neutrally. I assume that interest will be there for the next cycle as well.

In the case of VA, Gov McDonnell went so far as to create a Bipartisan Redistricting Commission in 2011, though it had no real authority. That suggests to me that VA would be interested to see how neutral criteria would affect their state.

I still think this is an important question for neutral mapping efforts using defined criteria:

It seems to me that neutral criteria either recognize that a state has section 2 obligations or that it doesn't. If it does then the path is clear. But here, I am prone to agree with you that it does not have a section 2 obligation. How then does one construct a neutral map that still provides an opportunity for minorities to elect a candidate of choice?

[muon2]

I don't like the erosity of your grey CD, but then we do have different philosophies about that. I am willing to chop when the erosity gets to be too much, absent a darn good reason to do otherwise. But then you already know all of this all too well about me!

Anyway, these maps are more about how legislatures interested in good government maps due to political power splits, or courts, will draw the lines, as opposed to following your unique rules (and up to a point mine). (In regard, the Jimtex definition of metro areas is totally unique, and I don't think most folks would view Stafford and Spotsylvania as part of the DC metro area.) Among other things, I suspect courts are more interested in "art" than you are.

Your art has always differed from mine. For example I find that your NW peninsula out of CD 5 and jut and chops around Richmond to add far more erosity than the clean L-shape of my CD 7. I do hope you are not abandoning the idea of measurable metrics to judge maps.

edit: The Census agrees with jimrtex that Fredricksburg and the two adjacent counties are part of the DC metro.

edit2: I think I found the math discrepancy. First the April to July shift is 0.25 of a year. Then the formula to get the annual rate is (estimate/census)^(1/4.25) - 1. The projection is census*(rate + 1)^10. I also had to correct for the fact that Bedford city was separate from the county in 2010, but they are together for the county estimate.

Right you are. The census excel list does not have Bedford City listed. So with that correction, my numbers for the VA total and per CD are 8,806,268.366 and 733,855.6972, respectively. So we are still off from each other.

If you have the numbers that go into the three counties I listed, I can perhaps reverse engineer our discrepancy.

[muon2]

It looks like you are not using the 7/1/14 Census estimates for VA. Are you using ACS numbers instead?

You must be logged in to read this quote.

You may have detected me cringing when I wrote about the use of the ferry in my VA map. I only relented in an apples to apples comparison with your work and jimrtex's comments. I'll be happy to return to the stricter rules for connectivity.

I appreciate your suggestion about two maps emphasizing different features. As I noted at the beginning of this series, the best is probably to reduce chops and erosity by allowing inequality to rise to to a few percent. There's a cumulative error of a about a half percent per year due to changing patterns of migration, births and deaths. So 2014 data used for 2020 CDs should probably allow a 3% variance to the quota. I've run studies to show that a map of IA drawn with minimal deviation and one drawn with 0.5% maximum deviation will have statistically similar deviations after 2 1/2 years, equal to the time between the census and the first general election.

[muon2]

The link to the correct numbers was in my earlier post.

My idea since the OH competition has been similar to what you have. I like the bargaining idea which I hadn't considered. I have to think about the Survivor aspect of the proposal, which is certainly novel. However, I don't think a computer as map maker works.

Redistricting is an example of a NP-complete problem known to computer science. It basically says you can't guarantee that any algorithm will converge to an optimal solution or set of optimal solutions in polynomial time - that means the time grows exponentially as the state becomes more complex and counties require chops. It also means you can't readily defend in court that the computer will beat a neutral mapper.

I favor making the computer the judge of what is EPO, a task well suited to NP-complete problems. Anyone, including someone using a computer algorithm, can submit a plan to be judged. A human commission would support the judging algorithm to guard against hacks and to make sure all legal niceties are followed including the VRA.

[muon2]

I'm still concerned as to how a computer draws a VRA-compliant map for IL, NY, or CA. In IL at the legislative level it would be extremely difficult to code that. Notwithstanding any of that, why not open up the process to the public in parallel with any computer algorithm that runs. It seems to me that would work in favor of the negotiation aspect in your proposal.

[muon2]

They do look quite similar. I estimated CVAP numbers from the 2013 3-year ACS data for my plan last April. Based on my estimates your 46.4% BVAP should be 50.1% BCVAP.

CD 05: BVAP 46.3%, BCVAP 50.0%
CD 08: BVAP 52.9%, BCVAP 52.4%
CD 09: BVAP 52.8%, BCVAP 50.9%

CD 07: HVAP 54.9%, HCVAP 52.3%
CD 14: HVAP 58.1%, HCVAP 59.3%
CD 15: HVAP 56.5%, HCVAP 57.8%

It is interesting to see the HCVAPs bigger than the HVAPs. The immigrant population in the Bronx is more from non-Hispanic countries.

[muon2]

MN was a state on the bubble to lose a seat in 2010, but they dodged a bullet and kept 8. The forecast for 2020 is that MN will lose a seat, though they are still on the bubble. I'll assume they drop to 7 as I use the 2014 data to project county totals in 2020.

The Twin Cities UCC will have about 4 1/4 seats in a seven seat map. This plan preserves the cover count of that UCC and keeps deviations within 5%. The Minneapolis area (orange) has 2 districts.

How close to the cusp vis a vis the population projections is the map below in play as between the blue and teal CD's, which has less erosity I wonder. In both maps, MN-01 moves discernibly more Pub, but it moves more in that direction in the map below.

Your line is definitely better on the erosity measure by 2 (6 vs 8 on the CD1-CD6 line). I drew it that way to better spread the inequality. My projections for CD1+CD6 are that they are about 20K short of the quota. My split left CD1 under by 7K and CD6 under by 13K. Your split puts CD1 over by 9K and CD6 under by 30K. However CD3+5 is over by 33K so you could just chop into Anoka or Hennepin to reduce inequality. Similarly CD4 is 21K to high and CD2 is 24K too low, so a chop into Washington would be required.

[muon2]

When I did my MN map I recognized how far out the estimates were from 2020, so I made two approximations. One is that I allowed the inequality to rise above 0.5%, up to as much as 5%. Second was that I focused on cover but not pack rules. Metro areas and urban densities will likely be revised by the next Census, so the definition of the UCC will possibly change. Erosity tradeoffs for pack and inequality are dependent on population estimates within 0.5% of the actual value, so I punted on pack and strict inequality here. I also didn't have muni projections when I made the map. I can look at it again as if my projections were within the 0.5% tolerance and UCCs are unchanged over this decade.

[muon2]

Here's my take on the 2020 projection assuming the 2014 estimates give the exact answer. That means that all CDs now must be within 0.5% of the quota. I apply the same chop rules as at the end of the MI exercise. Each county chop counts 1, each muni chop in a county counts 1, each CD above the minimium for UCC pack or chop counts 1, each chopped county that is over 5% of a CD is a macrochop and requires erosity to consider the local connections of munis within the county.

I started with my raw map, and worked on the Twin Cities area first. I built a chop into Washington that kept all munis whole. It's not a macrochop so the only extra erosity is due to there being two fragments and additionally one fragment is regionally connected to Ramsey. That's a net increase of 2 erosity over the pure county line, and 1 more erosity than Torie's version which appears to trade a chop of Cottage Grove for the reduced erosity. CD 4 ends up -1412 (-0.17%) under quota.

The next task was constructing CD5 with Minneapolis and south Hennepin. Staying within quota and avoiding chops led me to the more erose arrangement using Brooklyn Center instead of Golden Valley plus Robbinsdale which add about 4200 in population. As drawn CD 5 ends up +2734 (+0.34%) over quota.

Removing CD 5 from Hennepin plus Anoka left a population that was just over 30K above the quota for a CD. Hennepin is macrochopped perforce, so an additional chop into it will use munis for erosity. The erosity of the Hennepin-Wright border due to the Hennepin macrochop is 3 with connecting highways at Independence, Rockford, and Rogers (nb in 2012 Rogers annexed all of Hassan twp). The chop in Tories plan has an erosity of 8 between the munis in western Hennepin. A chop into Anoka at Ramsey is not a macrochop and also follows the only highway link between Sherburne and Anoka, so there is no erosity hit, just the chop. CD 3 ends up +1797 (0.22%) over quota.



Working outward there is the issue of the pack count that I violate by pushing CD 2 into Rice and LeSueur. That choice is a trade of a pack point for an extra chop such as Wright in Torie's version. I gain some erosity by adding those two counties, but not as much as I would with the tri-chop of Hennepin. My CD2 is -1282 (-0.16) under quota.

CD 1 is whole county and to bring it within quota I took Torie's original line and shifted Yellow Medicine. That puts it at +15 (+0.00%) over quota.

CD 7 is also whole county and I had to swap Mille Lacs and Becker to get it within quota. It ends up -1151 (-0.14%) under quota. The remainder is CD 6 and it's at -702 (-0.09%) under quota.

[muon2]

The fundamental problem for any erosity or compactness measure is to have it make sense low density and high density areas. These are some principles I think any system should accommodate:

1. Compact shapes should be generally preferred over non-compact shapes.
2. Borders that respect natural barriers to travel should be preferred.
3. Borders that respect accepted political boundaries should be preferred and county borders should have priority over whole cities.
4. An otherwise compact shape should not be penalized for an erose boundary that arises solely from natural or accepted political boundaries.
5. Similarly-shaped districts should have comparable scores whether they cover large rural areas or small urban areas.

This last part is hard to get right. Consider the MN districts we drew. My MN-5 is certainly more erose than yours. To measure that one needs to consider the impact of the shapes of the individual munis. However, a simple chop to balance populations between two rural districts doesn't need that scrutiny.

So when does a chop rise to the level requiring a county to be viewed by its divisions? During the MI exercise we came up with 5% as a reasonable threshold.

Like all thresholds it gas consequences. Its weakness is when a rural district goes into an already chopped urban county such as your MN CD-2. It probably creates a general bias from constructing such districts, but many of the posts in other threads have noted that its not bad to disfavor such rural-urban districts. This philosophy manifests itself more clearly in the cover-pack rules, but it shows up here as well.

[muon2]

"Its weakness is when a rural district goes into an already chopped urban county such as your MN CD-2."

My MN-02 was not a rural CD. I take your point about the hostility of rural CD's intruding into urban areas, that does not obtain with respect to intra metro area CD's. Whether there is any merit to drawing a distinction is yet another matter. Indeed, I macro chopped precisely because I wanted more pure urban focused CD's.

I understand your point, I blanked on the CD-6 into Wright as opposed to the CD-2 into Hennepin. I know the area well and except for the exurb Rogers, most of your CD-2 in Hennepin is rural, so that helped me lose sight of your objective.

Erosity and compactness measures generally don't know the population make up of the geographic units in the district. The exceptions are compactness measures that use the distances to the center of population, but they aren't particularly good at measuring erosity. We spent a lot of time with Grand Rapids and Detroit area districts to find a scoring balance in macrochops that reasonably approximated what the eye suggested. I'm not sure it makes sense to go back to the drawing board.

Did you look at swinging all those chops east so that CD-2 macrochopped Washington and then CD-4 went into Anoka?

[muon2]

Is it possible to get all of the Red River into a single district?

The current 5:3 split requires inclusion of St. Cloud into the metro area. In a 4:3 split, St. Cloud goes back to the outstate districts, and logically goes to the western district, which will have to make up population to the southern district.

Here's a version that keeps the Red River together by stretching the St Cloud CD east to WI.

[muon2]

I'm not sure how I score erosity for MN-05 with your proposal. Yours is better shaped than mine and should get credit for that.

I used an early system for our AL discussion that had no use of subdivisions or their shapes, just the main county chop. I found that system resulted in no penalty for a badly shaped chop in a large population county, and it could be easily gerrymandered. Invoking subdivisions for erosity in large counties was my solution.

[muon2]

One does not credit for keeping the Red River basin together, but it does appear by doing so, erosity goes down in the map above.

One thing I like about the scoring system is it allows a mapper to try to achieve certain geographic goals, and if the result stays on the Pareto frontier it can go into the mix.

You must be logged in to read this quote.

When we were looking at MI for the purposes of macrochops one situation arose where a county had two ordinary chops that would have been a macrochop had they been added together. This was felt to be an undesirable gaming of the system, so I adjusted my rule to look at the sum of all chop fragments less the most populous one to determine if a macrochop existed. The result is that once a county is macrochopped it is treated as a collection of subdivisions for all districts in that county when calculating erosity.

I looked at your Wright chop in more detail and it is perilously close to a macrochop in 2020 (39K). It is fine under this exercise, but might not be by 2020.

[muon2]

I like your idea. I've adjusted it to equalize populations using my projections to 2020.




In a macrochopped county all the subdivision units are treated as linked to each other if they are locally connected (ie any paved public roads) between their town halls. Units in a macrochopped county are linked to an adjacent county if there is a numbered highway that crosses the border between them.

[muon2]

I like the first AZ plan.

The reservations are indeed a problem area. They are like munis that cross county lines, but there's more desire to keep them together than munis at county lines. At a minimum I would agree that there should be no chop penalty for an area that is exactly coincident with the reservation land in a county. At one point jimrtex and I suggested treating reservations as separate counties since that would then create a chop penalty for reservation splits, but not for chopping into the counties to keep them whole. It all depends on how strong one wants to make the incentive to keep reservations whole.