DRA stuff

[muon2]

Here is my effort a 5-0 Democratic Oregon. Blue voted for obama by about 40k votes in 2012, making it D+2 (estimation).
(Cross-posted from AAD)

A nearly even result in 2012 translates to R+2. Obama got 52% of the two-party vote in 2012. A district that was even in 2012 is two points more Pub than the nation as a whole, hence R+2.

[muon2]

Here is my effort a 5-0 Democratic Oregon. Blue voted for obama by about 40k votes in 2012, making it D+2 (estimation).
(Cross-posted from AAD)

A nearly even result in 2012 translates to R+2. Obama got 52% of the two-party vote in 2012. A district that was even in 2012 is two points more Pub than the nation as a whole, hence R+2.

I got that number by taking Obama's margin in 2012 (216k votes) and dividing by 5. I get 43.2k for a district to match state's 2012 PVI (D+3). I decided to be conservative and say D+2 because the 40k figure was on the higher end of my estimated Obama margin in the district.
I'm not using the two-party vote measure here, hence the confusion.

It isn't hard to use the two party vote share. Just take the Obama vote and divide it by the total that went to Obama+Romney. For example in OR '12, Obama got 54.24% and Romney got 42.15%. .5424/(.5424+.4215) = .5627 or 56.27%. Nationally in '12 Obama's fraction was .5101/(.5101+.4715) = .5197 or 51.97%. Third parties took a larger share of the vote in OR than in the US, so the effect is to raise Obama's numbers in OR compared to the US. The result would be a PVI of D+4 for the state in 2012.

Technically the PVI should be the average of 2008 and 2012. Even if you just use one cycle but use the actual two-party vote it will reduce confusion.

[muon2]

Your imgur images aren't working without copying the link to a separate tab. There's a problem with imgur on this site that was described here.

Also, it would help to not make one of the districts white, since that's the color of the neighboring states.

[muon2]

So I decided to see what a Boise-centered House seat would look like. Republicans back in 2011 shot down the idea.
I was able to split no counties. The Boise-centered House seat would have 782,407 people, while the other seat would have 785,175 people.

The blue seat is 39% Obama, 58.8% McCain; the other seat is 33.4% Obama, 64.1% McCain.

I posted a couple of ID plans a while ago (if you search you can find plans for just about every state). The first is a particular favorite of mine since it is whole county with a population deviation of only 1 person. The second plan keeps the deviation to only 146 while keeping Boise-Nampa whole.

It would be nice to have a clear statement of what the exact role of regions is, vis a vis county/transportation link splits. How does inserting the concept of regions end up with better maps with a system that would apply to all 50 states?  Is part of it an attempt to respect metro regions even when it ends up with more county splits.  The mathematics of this all is quite daunting.

The simple answer is that it reflects the number of county splits. If D is the number of districts and R is the number of regions then the minimum number of county splits S = D - R. The links provide a constraint on region formation so that one doesn't have a split like this for ID where one county (Lemhi) has no connection to any other in CD 1 even though the districts are whole county with a population variance of 1.



The intermediate answer is that regions are easy to form and configure. By using the links to measure erosity it provides a another way to judge a plan in addition to both county splits and population equality. Public mappers can easily follow a plan based on regions. Links tend to be more dense in areas joined by communities of interest so such a region-based plan using erosity will tend towards districts that also follow the big-picture communities of interest. For example, it could lead to this plan that is better connected with a deviation of 146 that keeps the Boise metro largely intact.



The complex answer is that the notion of nodes and links can be extended into counties that need to be split. This creates a hierarchy of measures, one by region and one for the plan itself. The measures can be related at both levels and derive from the same methodology, so it remains reasonably accessible to the public. The principles that would guide well-formed regions then can also guide the formation of districts within those regions. Using ID as an example, this would provide a way of judging the merits of a plan that splits Ada county and Boise city but keeps the north and east separate.

Edit: I took a closer look at the ID state highways and found I counted a road across Boise County linking SR 55 and SR 21 as a state highway when it is not. Therefore, by my narrow rule for connections there is no way to divide ID without cutting the Boise metro. It does show the utility of the state creating an altered list of connections. It would permit connections from Gem and Adams to Boise and thus to eastern ID without going through Ada County, or allow the state to forbid that type of district by excluding that alteration. More reason to provide a county split mechanism as well as one for whole county regions.

[muon2]

Any road connection at all between the two ends of my district?

Only a dirt road through the mountains between Valley and Custer counties. Boise county has the first paved road that connects coming south from the Panhandle.

[muon2]

Any road connection at all between the two ends of my district?

Only a dirt road through the mountains between Valley and Custer counties. Boise county has the first paved road that connects coming south from the Panhandle.

Interesting.
Of course the smaller seat is Boise MSA+Twin Falls MSA+Gooding County. That leaves that dirt road as the only road connection. I haven't gave much thought to the question of a dirt road counting as a road connection, which I usually have in my maps.

We have a long history of discussion on what counts as a connection going back to 2012. You can see some of that in my post on ID from that period. If you read the current local vs regional road thread you will see how that has evolved to the present.

[muon2]

My effort at 5 D seats. The blue seat is 46.9 W/46.1 B, 61.3 O/38.1 McC, 57.7 D/42.3 R. The green seat is 62.3 W/31.6 B, 49.9 O/49.2 McC, 51.2 D/48.8 R. The purple seat is 47.2 O/51.5 McC, 53.8 D/46.2 R. The red seat is 47.5 O/51.2 McC, 51.5 D/48.5 R. The yellow seat 46.6 O/51.7 McC, 52.4 D/47.6 R.
The teal seat, in contrast, is 33.3 O/65.5 McC, 38.5 D/61.5 R. The rest of the state, equal to 3 CDs, is 30.1 O/68.4 McC, 37.3 D/62.7 R.

You are relying on state-oriented Dems voting D for Congress, too. That's been getting harder in southern states over the last few cycles. The rough PVI's for your 5 D districts are D+8, R+3, R+6, R+6, R+6. Three of those seats would be considered pretty safe for Pubs barring a wave year for the Dems. The R+3 would be lean to likely R depending on the candidates.

[muon2]

It looks like you split up the black areas of Columbus. You can get 25-29% BVAP in a Franklin CD. That counts as an influence district to them.

[muon2]

It looks like you split up the black areas of Columbus. You can get 25-29% BVAP in a Franklin CD. That counts as an influence district to them.

I didn't consider the possibility. I guess I could have done something similar to RL CD-3.
I drew the lines in Franklin County without respect to ethnicity and with erosity as first.
What would a CD optomized for blacks look like? (without shifting voters in any other counties in any way)

The first OH congressional map in 2011 passed with a simple partisan majority. That made it vulerable to a popular referendum. To avoid that the pubs reached a compromise with the OH Legislative Black Caucus to improve BVAP in Columbus. The result is OH-3. It's very erose, but has a 29% BVAP.



This has the same BVAP, but without the erosity. Some of the erosity here is due to municipal lines.

[muon2]

It looks like you split up the black areas of Columbus. You can get 25-29% BVAP in a Franklin CD. That counts as an influence district to them.

I didn't consider the possibility. I guess I could have done something similar to RL CD-3.
I drew the lines in Franklin County without respect to ethnicity and with erosity as first.
What would a CD optomized for blacks look like? (without shifting voters in any other counties in any way)

The first OH congressional map in 2011 passed with a simple partisan majority. That made it vulerable to a popular referendum. To avoid that the pubs reached a compromise with the OH Legislative Black Caucus to improve BVAP in Columbus. The result is OH-3. It's very erose, but has a 29% BVAP.



This has the same BVAP, but without the erosity. Some of the erosity here is due to municipal lines.

Interesting. What would the PVIs be for these and the other two Columbus seats, if these shifts within Franklin County took place?

The purple CD above was 65.3-33.5 for 2008 Pres. That's a D+12 (65.3%/(.653+.335) - 53.7%). The 53.7% is the Obama 2-party vote in 2008, and it's the baseline for a PVI using only data from that election.

When you say the other two seats, why assume two. A CD with the remainder of Franklin plus Delaware, Union, and Madison is only about 10 K short of a CD and very compact. By my rules, if a county must be chopped the smaller piece(s) should add up to less than 5%. That's 36 K or less for OH CDs. Franklin of course will be larger, but there shouldn't be many large chops (we call them macrochops).

[muon2]

Are the CO PVI's now using the formula I gave you?

[muon2]

Are the CO PVI's now using the formula I gave you?

No. Sorry.
Of course, I do have the underlying files all saved for your interest, and I will try to use that formula when I can get myself to going forward.

If you can't use the formula, don't write down D+/R+ since the reader assumes it means the PVI. Write the popular vote percentages instead. That makes it clear to everyone and they can calculate the PVI for themselves from the percentages.

[muon2]

Are the CO PVI's now using the formula I gave you?

No. Sorry.
Of course, I do have the underlying files all saved for your interest, and I will try to use that formula when I can get myself to going forward.

If you can't use the formula, don't write down D+/R+ since the reader assumes it means the PVI. Write the popular vote percentages instead. That makes it clear to everyone and they can calculate the PVI for themselves from the percentages.

Okay, I will do that instead.
For the record, in that D gerrymander of CO, the teal seat is 50.5 Obama, and the white one is 50.4 Obama.

Thanks, and include both percentages as well. There's a big difference in a district that was 50.5-49.5 and one that was 50.5-47.5 (2% other). That's why the formula adjusts for the two party vote first, then compares to the national 2-party percentage (53.7% in 2008).

[muon2]

In IL over half the population is in three counties (Cook, DuPage, and Lake). How many states don't have half the population in 10 counties or fewer?

[muon2]

Not just the VRA, but you split reservations in AZ. That's generally frowned upon in neutral maps. It can be more important to keep reservations together than counties.

[muon2]

Not just the VRA, but you split reservations in AZ. That's generally frowned upon in neutral maps. It can be more important to keep reservations together than counties.

Which ones did I split?

Gila River IR extends from Pinal (pop 8774)  into Maricopa (pop 2994).
Tohono O'Odham IR is primarily in Pima (pop 7174) with a bit in Maricopa (pop 151) and the part you chopped in Pinal (pop 545).
Fort Apache IR is primarily in Navajo (pop 11,199 in the main VTDs) with parts in Apache (pop 555) and the chopped part in Gila (pop 1676 in the 2 main VTDs).

The boundaries of current AZ-1 and AZ-4 keep all of Fort Apache IR in AZ-1. The boundary between AZ-1 and AZ-3 keep all of Gila River IR in AZ-1 and Tohono O'Odham IR in AZ-3.

[muon2]

Not just the VRA, but you split reservations in AZ. That's generally frowned upon in neutral maps. It can be more important to keep reservations together than counties.

Which ones did I split?

Don't feel bad. I did the same thing when I did my first AZ maps. Lewis Trondheim took the paddle to me, and I still feel the sting from that.

I am most focused on cutting county splits generally. Just look at my maps!

Reducing county splits is good, and if that's the goal the number of splits can be reduced even further. For example, this is a map of OK I posted in 2012 with no county splits and it has 51D-49R for the partisan avg in CD-2. All the CDs are within 0.5% of the quota.



It turns out that many states can be drawn with districts within 0.5% with no county splits except for counties which are larger than a CD. There's even a mathematical relation I derived from posted maps that relates the number of counties in the state, the number of districts, and the typical population deviation.

[muon2]

Yep. It it's easier as the total # of counties/CD ratio increases.

But I don't think the mathematical relationship was obvious. In any case, for your neutral offerings, why not concentrate on posting maps that really reduce the number of county splits to the absolute minimum?

[muon2]

Yep. It it's easier as the total # of counties/CD ratio increases.

But I don't think the mathematical relationship was obvious.

I was saying that as a general rule, the more counties you have, the less likely you are to have to split one for population equality. It's not really a formula per se.

I'm saying that it is more than a general rule and it does fit to a formula.

A region is a group of whole counties with a whole number districts within the required variance form the quota. A district of whole counties is a region with 1 district. Two districts that are whole counties except that they share a chopped county are effectively a region with 2 districts. Extending that to a whole state, each chop reduces the number of regions in a plan by one.

It turns out that there is a strong correlation between the average number of counties per region in a plan and the inequality. A few years ago we drew up inequality-minimizing plans with no chops and without regard to erosity. I summarized those results for the Forum Redistricting Commission thread in this post.

Let's add to the discussion the I in SPICE: Inequality.

Definition: Quota. The quota is the total population of a state divided by the number of districts rounded to the nearest whole number.
Definition: Deviation. The deviation is the difference between the population of a district and the quota. Negative numbers indicate a district that has a population that is smaller than the quota.
Definition: Range. The range is the difference in population between the largest and smallest district in a plan.
Definition: Average Deviation. The average deviation is the average of the absolute values of the deviations for all districts in a plan.

Background: SCOTUS has set two different standards for districts. Legislative and local districts must be substantially equal and that has been interpreted to be a range not exceeding 10% of the quota. Congressional districts must be as equal as practicable, and for some time that was assumed to mean that only exact equality would do. However, the recent WV case makes it clear that a range of up to 1% of the quota is acceptable when driven by other neutral redistricting factors. Greater than 1% might also be acceptable, but 10% would presumably not be because that is set by a different standard. It's an evolving area in the law.

Item 6. All plans for congressional districts shall have a range not exceeding 1% of the quota. All other plans shall have a range not exceeding 10% of the quota except when otherwise limited by state law.

Background: Some time ago there were some threads that tried to optimize the population equality of districts with no county splits. The result of that exercise was the following graph.



Each square represents a state. New England states used towns instead of counties, and states with counties too large for a district assumed that a whole number of counties would nest inside the large county. The more counties available per district, the closer to equality one could achieve, and the relation is logarithmic in population. The green line represents the best fit to the data. Data for average deviation can be fit as well, but the result is not substantially different other than the scale factor that has the average deviation equal to about 1/4 the range.

The average state has about 72 counties and if one divides that number into 2, 3, 4, etc. districts then one can use the fit from the data in the graph to predict a likely range. That in turn can be built into a table.

Item 7. The INEQUALITY score for a plan is found by taking the range for a plan and comparing it to the table below.

Range	Inequality
0-1	0
2-10	1
11-100	2
101-400	3
401-900	4
901-1600	5
1601-2400	6
2401-3200	7
3201-4000	8
4001-4800	9
4801-5600	10
5601-6300	11
6301-7000	12
7001-7700	13

This score reflects the expected improvement one should get by adding chops to a plan. So adding it to the chop score creates an automatic balance between chops and inequality. It also provides for a Pareto choice vs erosity where plans with no chops are likely (eg IA).

[muon2]

I like your neutral offerings better. I'm sure there will be plenty of more relevant gerrymanders when estimates close to 2020 start to come out.

[muon2]

Michigan fair map

1 (blue): R+1
2 (green): EVEN
3 (purple): R+4
4 (red): D+3
5 (gray) D+9
6 (gold): R+8
7 (teal): D+1
8 (slate blue): D+5
9 (pink): D+2
10 (chartreuse): D+9
11 (aquamarine): D+13
12 (cyan): R+4
13 (dark salmon): 46% White, 48.8% Black; D+23
14 (olive): 43.2% White, 45.4% Black; D+24

That's more like it. Would you like it scored with the other MI maps we all posted a few years ago?

[muon2]

Michigan fair map

1 (blue): R+1
2 (green): EVEN
3 (purple): R+4
4 (red): D+3
5 (gray) D+9
6 (gold): R+8
7 (teal): D+1
8 (slate blue): D+5
9 (pink): D+2
10 (chartreuse): D+9
11 (aquamarine): D+13
12 (cyan): R+4
13 (dark salmon): 46% White, 48.8% Black; D+23
14 (olive): 43.2% White, 45.4% Black; D+24

That's more like it. Would you like it scored with the other MI maps we all posted a few years ago?

Sure.

In order to score your map, I need to recreate it to measure the chops. The scoring rules treat chops more harshly if they exceed 5% of the population of a CD. Some of your county chops look like they slice through townships and cities (which are equivalent to townships in MI). We also treated maps more harshly if they split multiple townships and cities in a single county.

Do you have the population deviations for each CD? That would tell me if I've recreated it correctly.

[muon2]

Michigan fair map

1 (blue): R+1
2 (green): EVEN
3 (purple): R+4
4 (red): D+3
5 (gray) D+9
6 (gold): R+8
7 (teal): D+1
8 (slate blue): D+5
9 (pink): D+2
10 (chartreuse): D+9
11 (aquamarine): D+13
12 (cyan): R+4
13 (dark salmon): 46% White, 48.8% Black; D+23
14 (olive): 43.2% White, 45.4% Black; D+24

That's more like it. Would you like it scored with the other MI maps we all posted a few years ago?

Sure.

In order to score your map, I need to recreate it to measure the chops. The scoring rules treat chops more harshly if they exceed 5% of the population of a CD. Some of your county chops look like they slice through townships and cities (which are equivalent to townships in MI). We also treated maps more harshly if they split multiple townships and cities in a single county.

Do you have the population deviations for each CD? That would tell me if I've recreated it correctly.

I don't, but I could (a) email you the drf files or (b) load them later and tell you the deviations.
Which do you prefer?

You can load them later. However, I can tell your score will suffer from the large number of township/city chops - particularly in the large counties. Townships and cities are subunits, and once more than 5% of a county is chopped into other CDs then each subunit chop counts against the score, just like a county chop. You may want to make some edits first to fix some of the more obvious subunit chops.

Here's the thread with the other scored MI maps.

[muon2]

Well the reason you might see some townships chopped is because I was looking at DRA's municipal lines tool very closely to avoid municipal splits.
Does one take precedence over the other?

DRA shows lines for Census places. Those include cities which are equivalent to townships, and villages which are subsidiary to townships. There is a nice map on the thread I linked that show the county subunits for Wayne, Oakland and Macomb counties.

On DRA you can tell what are subunits by the voting district (precinct) number.  In MI the first three digits are the county number. The next four digits are the county subunit, either township or city. The remaining digits identify the voting district in that subunit.

[muon2]

Well the reason you might see some townships chopped is because I was looking at DRA's municipal lines tool very closely to avoid municipal splits.
Does one take precedence over the other?

DRA shows lines for Census places. Those include cities which are equivalent to townships, and villages which are subsidiary to townships. There is a nice map on the thread I linked that show the county subunits for Wayne, Oakland and Macomb counties.

On DRA you can tell what are subunits by the voting district (precinct) number.  In MI the first three digits are the county number. The next four digits are the county subunit, either township or city. The remaining digits identify the voting district in that subunit.

Wait so that is what they truly mean by 'city and town lines'?

Yes. In MI some are county subdivisions and others are just Census places that don't count as separate county subdivisions. Each state is different based on their own laws. DRA just loaded the place shapes for all states and didn't distinguish between types of municipalities. Wikipedia is usually accurate about which munis fall into which category. If you use Wikipedia as a reference for MI, it's townships and cities that matter for the chop count.