Fair redistricting: New York

[muon2]

This is my initial submission (copied from the other thread).

I checked the town maps and there is only one county chop, no town chops and does not chop a UCC.

Stats for the districts:
CD 1 (blue): deviation +75; PVI D+6.66
CD 2 (gree): deviation -74; PVI R+0.51

The compact districts have an erosity = 3 by the muon rules using counties as the primary unit and towns as the subunit.

[muon2]

I'll provide some analysis of the submissions to help the panel. Did anyone convert solid4096's submission into DRA?

Analysis terms (low scores are better):

SKEW measures the lean of the delegation compared to the statewide expectation based on PVI. CDs with a PVI of 0 or 1 (tossup) don't count. For ME the expected lean is 0.25 seats for the Dems. A plan with 1 D and 1 tossup (most of the plans) with score a SKEW of 1.

POLARIZATION measure how well the CDs will respond to changes in the electorate. Competitive seats with PVI of 2 to 5 count 1 and uncompetitive seats of 6 or more count 2.

INEQUALITY is a relative measure of the population range.

CHOP sums the number of chops of counties, towns, and urban clusters. Chops that create multiple fragments larger than 5% of a CD add to erosity.

EROSITY measures the irregularity of district borders in terms of the number of broken connections between counties and towns within counties.

Map scores:
TimTurner - S 1; P 2; I 3; C 1; E 7
Singletxguyforfun - S 1; P 2; I 5; C 1; E 14
Torie - S 1; P 2; I 7; C 1; E 7
LimoLiberal - S 1; P 1; I 7; C 3; E 26
Kevinstat 1 - S 1; P 2; I 2; C 4; E 4
Kevivstat 2 - S 1; P 2; I 2; C 1; E 4
jimrtex - S 1; P 2; I 3; C 1; E 7
muon2 - S 1; P 2; I 3; C 1; E 3
starpaul20 - S 1; P 2; I 4; C 1; E 4
solid4096 - S 1; P 2; I 3; C 1; E 7

[muon2]

I finished scoring the plans, and there are some easy reductions to make for the panel. Basically there are these 6 plans.

TimTurner, jimrtex, and Solid4096 have identical plans that chop Kennebec and have scores of 1, 2, 3, 1, 7 on the SPICE metrics.



Singletxguyforfun also only chops Kennebec but leaves off Lincoln into CD 2. This results in a deep macrochop of Kennebec and significantly increases erosity. The SPICE scores are 1, 2, 5, 1, 14



Kevinstat 2 and Starpaul20 both chop Oxford and put Lincoln in CD 2. Kevinstat 2 splits a voting district to move the town of Gilead into CD 1, but is otherwise the same. That shift creates SPICES scores of 1, 2, 2, 1, 4



Torie puts Lincoln in CD 1 and uses less of Oxford to balance population. This stretches out the district and increases the erosity. The SPICE scores are 1, 2, 7, 1, 7.



muon2 puts Lincoln and Sagdahoc in CD 2 and the chop goes in Franklin. Though the boundary is stretched to the north it goes through unpopulated areas that reduces erosity. The SPICE scores are 1, 2, 3, 1, 3.



LimoLiberal chops the Portland UCC and makes a high-erosity border to make CD 1 more competitive which shows up in the reduced polarization. The SPICE scores are 1, 1, 7, 3, 26.

[muon2]

LimoLiberal's map looks ridiculous honestly.   Any Maine map should by default be York+Cumberland+"Other" for ME-1, and then ME-2 just gets whatever is left.

I don't even think there's a main road going through ME-2 from York to Banger...

I checked, and there is a pathway of state and federal highways that can get you from York to Bangor entirely within LimoLiberal's CD 2. That only means they're connected, not that the path is direct.

[muon2]

The panelists have made their choices...! Now's our first round of voting; six maps made it.

Panelists, copy the list as follows and vote YES or NO to moving the maps to the final round. Do add comments for explanation if possible. Consider compactness, representativeness, constituent similarity, and to a lesser extent, splits (really just towns for this state; please do not be fanatical about counties). A "no" vote is not necessarily saying the map in question is bad, just that in context of the goal and the other maps you wouldn't recommend it for the final round. Hyperlinks are provided for reference. I vote:

Starpaul20 - Y
Torie - Y
Singletxguyforfun - Y
TimTurner et al. - Y
muon - N
LimoLiberal: N

LimoLiberal's is just kinda funky in its shape. muon's is fine, but imo Sagadahoc is better suited for ME-01 than Franklin and Oxford (particularly their northern, sparsely populated areas)

\

As someone who spent some time there (including getting an engagement stone), the bulk of the population of Oxford county is much more connected to SW ME (and NH) than it is to Bangor and the lumber country of northern ME. OTOH, Bath comes across more like the coastal communities to its east than it does the coastal towns SW of Portland.

[muon2]

All the yes ones all look more or less the same and Uon is right about southern oxford and southern franklin

I know I was really talking about the northern parts. I especially like Starpaul20’s way of executing the split. muon’s map isn’t bad but I want to narrow down the maps for the final round

But virtually nobody lives in the northern parts of Oxford and Franklin, though they are certainly larger by area. Are you suggesting that to curry your vote on future states we should favor the interests of the larger proportion of geographical area rather than the larger proportion of population in a county?

[muon2]

If a plan is considering UCCs the only one with multiple counties in NH is the Portsmouth/Boston UCC consisting of Rockingham and Strafford. Hillsborough is separate.

In NH counties exist, but are less important than in ME. The Census Bureau has a measure that is like MSAs built on counties, but is built around towns instead. They are called NECTAs and here is a map showing them in NH. The red, orange and yellow areas are subdivisions of the Boston NECTA.

[muon2]

One of the criteria is that towns be not merely contiguous, but also connected. Here is a map showing which towns are connected by road (excluding the northern end in Coos).

[muon2]

One of the criteria is that towns be not merely contiguous, but also connected.

It is?

When you suggested that contiguity by water required a bridge, I suggested that contiguity by land should have a road, too, and no one objected.

[muon2]

My first submitted plan [muon2-A] aims to minimize county and UCC chops, create highly competitive districts, while minimizing erosity and deviation. The deviation is 29, PVI R+0.93, D+0.96, erosity 28 (town-based). There are 5 NECTA chops.




My second submission [muon2-B] recognizes that the NECTAs are a better measure of community of interest than counties, so it keeps them together with no NECTA chops. That results in 3 county chops plus the UCC chop. The districts are slightly less competitive: PVI D+1.56, R+1.73. It does allow for lower erosity: 24 (town-based). The deviation is only 1 person.

[muon2]

This is basically a compact urban-rural scheme, with the densely populated southeast in the second district. I made this map very carefully from looking election maps, trend maps, and population distribution maps
PVI is D+0.8 and R+0.86, so v competitive for both
also, anyone who separates Dover and Portsmouth is a MONSTER

If I might suggest, you should swap the town of Epping (Rockingham) with the town of Lee (Strafford). It doesn't change the shape of your districts in a meaningful way and it reduces the deviation from 1966 to 115. It's the sort of detail that matters in court challenges, too.

I agree about Portsmouth and Dover. There's a reason that Rockingham and Strafford are in the same UCC. Though I would add Rochester to that pair, since it's more attached to Dover economically than Dover is to Portsmouth.

[muon2]

Here's a table for the submitted NH plans as I fill each in. The erosity is based on the town connection map. The NECTA chop is based on the NECTA map reflecting Census groupings of towns. The key is S:Skew, P:Polarization, I:Inequality, CC:Chop (Counties/UCC), CN:Chop (NECTA), E:Erosity. Low scores are better.

Plan	-S-	-P-	-I-	-CC	-CN	-E-
Solid4096	0	0	5	1	4	29
Singletxguyforfun	0	2	4	3	2	21
TimTurner	0	0	2	3	4	30
cvparty	0	0	4	4	3	23
Sol	0	0	3	5	3	20
HCP	0	2	2	2	2	25
LimoLiberal	0	0	3	2	4	30
Starpaul20	0	2	3	2	2	27
Gallatine	0	2	7	4	2	19
muon2-A	0	0	2	1	4	28
muon2-B	0	2	1	4	0	24
jimrtex	0	2	5	4	3	24

[muon2]

I believe this is the final table for NH. The SPICE scores allow simple comparison between plans. They aren't designed to pick a winner, but they are designed to weed out weak plans.

The usual way to interpret the scores is that if any plan has values that are in at least one case higher and in no case lower than another plan then it would be eliminated from consideration. I've highlighted those plans in the table. For example Starpaul20 has higher inequality and erosity than HCP while all other values are equal (red). Starpaul20 would be eliminated based on its SPICE scores. Similarly TimTurner and LimoLiberal are eliminated by muon2-A on chops and erosity (green). jimrtex would be eliminated by cvparty on polarization, inequality and erosity (blue). By using a smaller set of scores, such as just one of the chop scores or ignoring the polarization additional eliminations can be made to get a final competition set for voting.

Here's a table for the submitted NH plans as I fill each in. The erosity is based on the town connection map. The NECTA chop is based on the NECTA map reflecting Census groupings of towns. The key is S:Skew, P:Polarization, I:Inequality, CC:Chop (Counties/UCC), CN:Chop (NECTA), E:Erosity. Low scores are better.

Plan	-S-	-P-	-I-	-CC	-CN	-E-
Solid4096	0	0	5	1	4	29
Singletxguyforfun	0	2	4	3	2	21
TimTurner	0	0	2	3	4	30
cvparty	0	0	4	4	3	23
Sol	0	0	3	5	3	20
HCP	0	2	2	2	2	25
LimoLiberal	0	0	3	2	4	30
Starpaul20	0	2	3	2	2	27
Gallatine	0	2	7	4	2	19
muon2-A	0	0	2	1	4	28
muon2-B	0	2	1	4	0	24
jimrtex	0	2	5	4	3	24

Edit: Solid4096 was overlooked and is now added. It would also be eliminated based on SPICE scores from muon2-A.

[muon2]

Was my plan graded?

no, muon probably missed you cuz you just put a link

I did miss the link since it was in before the other maps. It is now scored with the other plans.

Based on the full scoring alone, there would be 7 plans that would go forward to the vote. If at some point you are concerned about the time for responses for both nominations and voting, this is a method that could replace the nomination step. It also would alleviate concern that the panel might unduly protect its own maps by nomination.

[muon2]

Based on the full scoring alone, there would be 7 plans that would go forward to the vote. If at some point you are concerned about the time for responses for both nominations and voting, this is a method that could replace the nomination step. It also would alleviate concern that the panel might unduly protect its own maps by nomination.

Possibly, although there are nuances in the maps that aren't fully captured in your SPICE method/scores. Also, we aren't too concerned about how much deviation there is; the .5% requirement is just to ensure that the districts are about the same, nothing more. I don't think we're going past that and looking at +100 districts versus +500 districts. Making that distinction would tend to interfere with more important goals like communities of interest

That was my point about using a reduced set rather than the full number of values. For instance one can say that inequality cannot be used alone to eliminate a plan; it can be left as a tie breaker. That way some panelists may use it in deciding their vote, while others may not. Historically when we've done this type of exercise there are roughly equal numbers who feel or don't feel that differences under 0.5% deviation matter.

Usually we've focused on chops and erosity as the prime measures with skew considered to avoid an unduly partisan plan. In working on some states we found that adding the chop score to the inequality score gave a better result than chops alone.

The problem we found with communities of interest is that panelists rarely agreed on subjective measures. It's hard on real commissions and especially hard here on Atlas where many don't have any firsthand experience with the states they're judging. The work on UCCs was to create an objective way to determine what parts of an urban area belonged together as a CoI and how to balance that against maintaining whole counties in rural areas.

[muon2]

I mean, surely there has to be a reason New Hampshire's CDs have had the same basic design since before the 20th century, no?

In many states there's a lot of history behind certain districts. If there's no gain from extensive gerrymandering, that history is going to influence a legislature drawing districts. Minimal change is a recognized neutral redistricting principle. WV successfully used that principle to uphold their current map. Then there are states like IA that intentionally start from a blank map every 10 years - the NH districts would probably change a lot each decade if they used the IA system, and that is essentially what we are doing in this thread.

[muon2]

For those who want to get a start on RI, I would note the following issues. RI has a PVI of D+9.6 and 2 CDs. To get a 0 Skew one district has to be a toss up (PVI less than 1.50). That's hard to do without making a mess of the map, so most plans will have a skew of 1 for the Dems. Polarization can be reduced by keeping one CD under PVI D+5.5 which is a competitive district that leans D.

RI has no county government, and they exist for purely statistical purposes. All local functions are handled by the municipalities. Because of that there is no chop score given for chopping counties, only municipalities. Also all of RI except for the municipalities of Hopkinton and Westerly are in the Providence-Warwick NECTA so every plan is going chop that once, and there's no value in that score.

Municipalities are the primary community of interest in RI, so keeping them whole is important. However, DRA only has 2010 block groups and the municipality name doesn't show up. The City/Town lines feature only works for a few municipalities since in some cases they show a Census place, not a municipality. Here's a map of the municipalities and the connection lines (including bridges and the Block Island ferry). Erosity is measured by the number of connecting lines cut by the border between CDs and cuts through municipalities.

[muon2]

I remember spending a lot of time trying to get a map with equal population (within the 1000 people range) without splitting any municipalities. I think it was impossible.

It's not impossible, and there are good low-erosity plans with under 1000 deviation and no muni chops. I've found at least 2. If you don't care about erosity you can get a plan with a deviation under 10 without a muni chop!

[muon2]

I remember spending a lot of time trying to get a map with equal population (within the 1000 people range) without splitting any municipalities. I think it was impossible.

It's not impossible, and there are good low-erosity plans with under 1000 deviation and no muni chops. I've found at least 2. If you don't care about erosity you can get a plan with a deviation under 10 without a muni chop!

I guess I am just going to sit out with RI.

The way to find them is not with DRA. Get the muni populations from Wikipedia (2010 Census). Copy them into a spreadsheet. Then take the map I posted and make connected groups of munis. Use the spreadsheet to keep track of the population of the group compared to the quota.

[muon2]

CD1 (Rochester, Dover, Concord) - D+2.6
CD2 (Portsmouth, Manchester, Nashua) - R+2.7

Deviation is ±1257.

Basically a choppier version of HCP's map - three county chops (Rockingham, Merrimack, Hillsborough), but all towns are intact and IMO looks better with chops than following Rockingham and Merrimack Lines.

For those of the judges who are uncomfortable with the deviations of this plan, you can perhaps pretend you're voting for this



or this



I'm sorry, I couldn't resist posting these maps, which I made midday yesterday but hadn't gotten around to saving in the Atlas Gallery and then posting here.  You probably can't pretend you're voting on something other than what you're actually voting on.

In fairness to Gallatine, to reduce the inequality as you suggest would increase the erosity by 1 and 3 points respectively.

[muon2]

NECTA's are arbitrary because they largely ignore commuting patterns.

Conjoined Urbanized Areas were separated based on pre-2000 MSA's. Remember that before 2000, MSA's in New England were town-based. Beginning with the 2000 Census, urban areas were delineated based on continuous semi-dense settlement. This in effect would produce a Bosnywash urbanized area stretching from Portland to Richmond. To avoid this, the Census Bureau decided to separate urbanized areas based on the pre-2000 MSA's. NECTA's are based on the urban areas at their core. Because of the scale of towns and also patterns of development and settlement fewer non-densely settled areas are captured based solely on commuting.

This is not analogous to UCC's since even when there is a non-separated urbanized area grandfathered (see Livingston, MI) the proto-MSA can be captured based on commuting. But that is not possible with NECTA's, since Nashua is too big to be captured by Manchester and vice versa.

The NECTA boundary between Springfield and Hartford is clearly misplaced based on commuting patterns (there is only one town in Connecticut where people mostly drive north in the morning).

Since one of the prime responsibilities of RPC's is transportation planning, they inherently provide better definitions of communities of interest.

The UCCs are a refinement of the Census MSAs that we originally looked at for determining clusters of counties that represented a community of interest. Nominally the NECTAs are based on the same analysis by the Census as MSAs, at least as I read their defintions:

You must be logged in to read this quote.

Just as there are MSAs where we might spot individual counties that might be better placed elsewhere, I have no doubt that we might find towns that look like they should go elsewhere based on state planning documents. We didn't do that with the MSAs as they became UCCs because that would remove the objective standard developed by the Census. However, if we wanted to take on the exercise of refining the NECTAs using the MSA to UCC metrics that would be a logical step.

[muon2]

I believe this is the final table for NH. The SPICE scores allow simple comparison between plans. They aren't designed to pick a winner, but they are designed to weed out weak plans.

The usual way to interpret the scores is that if any plan has values that are in at least one case higher and in no case lower than another plan then it would be eliminated from consideration. I've highlighted those plans in the table. For example Starpaul20 has higher inequality and erosity than HCP while all other values are equal (red). Starpaul20 would be eliminated based on its SPICE scores. Similarly TimTurner and LimoLiberal are eliminated by muon2-A on chops and erosity (green). jimrtex would be eliminated by cvparty on polarization, inequality and erosity (blue). By using a smaller set of scores, such as just one of the chop scores or ignoring the polarization additional eliminations can be made to get a final competition set for voting.

Here's a table for the submitted NH plans as I fill each in. The erosity is based on the town connection map. The NECTA chop is based on the NECTA map reflecting Census groupings of towns. The key is S:Skew, P:Polarization, I:Inequality, CC:Chop (Counties/UCC), CN:Chop (NECTA), E:Erosity. Low scores are better.

Plan	-S-	-P-	-I-	-CC	-CN	-E-
Solid4096	0	0	5	1	4	29
Singletxguyforfun	0	2	4	3	2	21
TimTurner	0	0	2	3	4	30
cvparty	0	0	4	4	3	23
Sol	0	0	3	5	3	20
HCP	0	2	2	2	2	25
LimoLiberal	0	0	3	2	4	30
Starpaul20	0	2	3	2	2	27
Gallatine	0	2	7	4	2	19
muon2-A	0	0	2	1	4	28
muon2-B	0	2	1	4	0	24
jimrtex	0	2	5	4	3	24

Edit: Solid4096 was overlooked and is now added. It would also be eliminated based on SPICE scores from muon2-A.

Does this say that Gallatine could not be beaten so long as no one beat his erosity score?

The elimination is based on Pareto equivalency. Any plan that is best in a particular aspect is guaranteed to survive. Another plan might also survive if surpasses in one aspect but not in others, that is if the plan is on the Pareto frontier.

[muon2]

Okay, I'm probably going to start a buncha regional threads to speed up this process, although we need to gather a functional, active panel first. How does this look?

You might want to distinguish the panels on some common elements. 20 states have Census-defined county subdivisions recognized by their states, and the rest don't. With a few switches you can group those together. I've kept IA in the Midwest since they only recently dropped their minor civil divisions with the Census. I split the large southern region into two so that TX and FL are in different threads.

[muon2]

I was actually considering using those exact divisions but decided just to go with four. I understand the importance of the groupings in most situations, but these sections are more just about having approximately equal numbers so I didn't think culture and history were really relevant in this context. Five might be a lot to do simultaneously. but if people prefer that, that's fine

It was not about culture and history. It was about the type of geography the states use internally, geography that the Census Bureau recognizes is different among the states. If you want to do more than judge by eyeball, that geography matters. If the panel finds that some rules help clarify that geography it makes sense to confine them to a given thread.

[muon2]

It was not about culture and history. It was about the type of geography the states use internally, geography that the Census Bureau recognizes is different among the states. If you want to do more than judge by eyeball, that geography matters. If the panel finds that some rules help clarify that geography it makes sense to confine them to a given thread.

what I mean is that the four groups were made just to quicken the process, it doesn't actually matter how they're grouped

That's because you assume all states are the same as far as how the rules will govern them. I claim that there are subtle differences and states can be grouped based on those differences. How one can best chop a county to recognize the communities of interest inside is one of the most important differences.