Ohio Redistricting Competition

[jimrtex]

I also end up with 6 districts at over 20% vote difference (5 R and 1 D), and I still have no districts between 10% and 20%. However, I've now been able to make two more districts under 5% for a total of 8. That cost some compactness, so this past weekend I've been tweaking boundaries to get my compactness back over 0.028.

What constitutes an entry?  Is it the list of files in the tutorial, or the the items in the contest rules?  The items in the contest rules can be generated easily enough from the files in the tutorial.

[muon2]

I also end up with 6 districts at over 20% vote difference (5 R and 1 D), and I still have no districts between 10% and 20%. However, I've now been able to make two more districts under 5% for a total of 8. That cost some compactness, so this past weekend I've been tweaking boundaries to get my compactness back over 0.028.

What constitutes an entry?  Is it the list of files in the tutorial, or the the items in the contest rules?  The items in the contest rules can be generated easily enough from the files in the tutorial.

I've asked that question, and it seems that it is not quite either. The tutorial is closer to correct in that only one map showing the districts is needed, not a separate map for each district as the rules suggest. However, there are files that should be copied that are not listed in the tutorial screen shot. I was told that copying all files that matched the name of the version to be submitted would catch the the files needed for the entry.

[jimrtex]

I also end up with 6 districts at over 20% vote difference (5 R and 1 D), and I still have no districts between 10% and 20%. However, I've now been able to make two more districts under 5% for a total of 8. That cost some compactness, so this past weekend I've been tweaking boundaries to get my compactness back over 0.028.

What constitutes an entry?  Is it the list of files in the tutorial, or the the items in the contest rules?  The items in the contest rules can be generated easily enough from the files in the tutorial.

I've asked that question, and it seems that it is not quite either. The tutorial is closer to correct in that only one map showing the districts is needed, not a separate map for each district as the rules suggest. However, there are files that should be copied that are not listed in the tutorial screen shot. I was told that copying all files that matched the name of the version to be submitted would catch the the files needed for the entry.

Is the style of the map important?  I didn't follow the tutorial exactly.  I found it a lot easier to work with a solid fill for the districts, rather than an outline map.  I turned off cities and towns and only showed county boundaries.

I was able to get an Even district, so that I am now 9R-8D-1E

I tried smoothing my districts in the Cincinnati and Dayton areas and my compactness score is just shy of 0.027.  I could probably go a big higher, but I flipped the partisan index of one district, and got tired of splittling cities.  I'd be really surprised if a plan went over 0.030 unless they split some extra counties.

[jimrtex]

I'm curious to see people's maps. Can you post them?

My original entry (PDF)

My final entry (PDF)

14 county fragments. 

Median compactness: .026903

Competitiveness: 12

< 5% = 6 districts  +2
5-10% = 6 districts +1
10-15% = 0 districts x 0
15%+ = 6 districts x -1

Representational Fairness (Party Quota):

9 R, 8 D, 1 E = 52.9% - 52.1% = 0.08%

Details of Split Counties

Cuyahoga (PDF)
Summit (PDF)
Trumbull - Mahoning (PDF)
Guernsey (PDF)
Franklin (PDF)
Madison (PDF)
Hancock (PDF)
Montgomery (PDF)
Hamilton (PDF)

[muon2]

I'm curious to see people's maps. Can you post them?

My original entry (PDF)

My final entry (PDF)

14 county fragments. 

Median compactness: .026903

Competitiveness: 12

< 5% = 6 districts  +2
5-10% = 6 districts +1
10-15% = 0 districts x 0
15%+ = 6 districts x -1

Representational Fairness (Party Quota):

9 R, 8 D, 1 E = 52.9% - 52.1% = 0.08%

Details of Split Counties

Cuyahoga (PDF)
Summit (PDF)
Trumbull - Mahoning (PDF)
Guernsey (PDF)
Franklin (PDF)
Madison (PDF)
Hancock (PDF)
Montgomery (PDF)
Hamilton (PDF)

What are you using for the county shots? Is it internal to ArcGIS or are you just cropping from the full state PDF?

[muon2]

I also end up with 6 districts at over 20% vote difference (5 R and 1 D), and I still have no districts between 10% and 20%. However, I've now been able to make two more districts under 5% for a total of 8. That cost some compactness, so this past weekend I've been tweaking boundaries to get my compactness back over 0.028.

What constitutes an entry?  Is it the list of files in the tutorial, or the the items in the contest rules?  The items in the contest rules can be generated easily enough from the files in the tutorial.

I've asked that question, and it seems that it is not quite either. The tutorial is closer to correct in that only one map showing the districts is needed, not a separate map for each district as the rules suggest. However, there are files that should be copied that are not listed in the tutorial screen shot. I was told that copying all files that matched the name of the version to be submitted would catch the the files needed for the entry.

Is the style of the map important?  I didn't follow the tutorial exactly.  I found it a lot easier to work with a solid fill for the districts, rather than an outline map.  I turned off cities and towns and only showed county boundaries.

I was able to get an Even district, so that I am now 9R-8D-1E

I tried smoothing my districts in the Cincinnati and Dayton areas and my compactness score is just shy of 0.027.  I could probably go a big higher, but I flipped the partisan index of one district, and got tired of splittling cities.  I'd be really surprised if a plan went over 0.030 unless they split some extra counties.

I agree with you about solid fill. I use a 50% transparency and it's easier to read than thick border lines. I submitted a version with the full precinct and county lines, but I also made a county-only version.

When I got home yesterday, I decided to deconstruct my map and attack it from a different angle. My initial map started by minimizing county fragments, and I was able to get a 8-fragment map that I reported on previously. That map had a compactness of 0.028 and competitiveness of 14. When the new rules came out, I noted that ties would be handled in bunches, but ties were unlikely to happen in compactness, so that was the new starting point for a map that took about 24 hours to produce.

I ignored all community boundaries except for those that cross county lines, and I made substantial use of the rule that allowed those cross-county cities to avoid creating county fragments, which eventually got me back to 9 fragments. I also allowed a number of districts to get thin and contorted so that I could push up the compactness for the other districts. This gets me to a compactness score of almost 0.033. I took 6 districts as uncompetitive with -1 scores but swapped pieces to get all others within the magic 5% for a competitiveness score of 18. I still have a 9-8-1 distribution and by luck the even district is exactly even.

To be honest, my map from last week looks better to me, but the new map uses the rules more effectively. I'll posted them both along with the aforementioned county detail so you all can take a look.

[jimrtex]

What are you using for the county shots? Is it internal to ArcGIS or are you just cropping from the full state PDF?

I zoomed in to the county and then printed from ArcGIS (when you print, you need to set CutePDF as the printer.  When you print it will ask you the destination of the file.  I send them to Final_Plan and then copied them to my system.

I'm not exactly sure how to control the image size.  CutePDF is apparently set up to generate 8-1/2 x 11 inch images, and my statewide maps came out different sizes.  When you print, it will print areas outside the screen image, particularly vertically.

[jimrtex]

When I got home yesterday, I decided to deconstruct my map and attack it from a different angle. My initial map started by minimizing county fragments, and I was able to get a 8-fragment map that I reported on previously. That map had a compactness of 0.028 and competitiveness of 14. When the new rules came out, I noted that ties would be handled in bunches, but ties were unlikely to happen in compactness, so that was the new starting point for a map that took about 24 hours to produce.

I ignored all community boundaries except for those that cross county lines, and I made substantial use of the rule that allowed those cross-county cities to avoid creating county fragments, which eventually got me back to 9 fragments. I also allowed a number of districts to get thin and contorted so that I could push up the compactness for the other districts. This gets me to a compactness score of almost 0.033. I took 6 districts as uncompetitive with -1 scores but swapped pieces to get all others within the magic 5% for a competitiveness score of 18. I still have a 9-8-1 distribution and by luck the even district is exactly even.

I think 0.033 is exceptionally good.  I only have 3 districts that are that high, and only 4 over 0.030.  I suspect that the mean would be better than the median for scoring, since the maximum (for a circle) is only 0.079, and an approximation of a circle using a grid of square townships is 0.049, and the minimum is 0.000 there isn't too much problem with extreme values, but there would be some reward for avoiding them.

After I did my second plan, I had a gap of about 0.003 between my 9th and 10th best districts (ie the median didn't really characterize my districts well).  Simply by adjusting a single district, I was able to get a substantial improvement in the median.

Have you tried the report mechanism in ArcGIS?  Most of the time it looks like it is putting out an empty report.  I did get it to work with my original map, but it didn't output in a neat format, and it took an awful lot of massaging to get it into a text file and then a spreadsheet.  I'd be peachy keen if there is a way to simply generate a .csv file from the attribute tables.

There is currently a proposed constitutional amendment regarding reapportionment before the Ohio legislature, that has a mild competitiveness characteristic.  At present, the legislative districts are drawn by a board comprised of 4 members appointed by legislative leaders of the 2 parties, and 3 members appointed by 3 state executive officers (IIRC governor, SoS, and one other), which tends to give a partisan bias. 

Ohio Constitution (see Article11)

The proposed amendment would change the composition of the board, and also bring congressional redistricting under its responsibilities, and apply some of the same standards that are currently used for legislating redistricting.  The current legislating standards are intended to minimize splitting of counties, and also other subdivisions (townships, cities, etc.).

Anyhow the current map is kind of interesting:

Current Ohio Districts

If you search for district numbers on the PDF of the house map, and then refer to the constitution it will make sense (districts consisting of a single county have the lowest numbers, then districts in multi-district counties are numbered in order of county population, and then the highest numbered districts are the rest of the state).

[muon2]

When I got home yesterday, I decided to deconstruct my map and attack it from a different angle. My initial map started by minimizing county fragments, and I was able to get a 8-fragment map that I reported on previously. That map had a compactness of 0.028 and competitiveness of 14. When the new rules came out, I noted that ties would be handled in bunches, but ties were unlikely to happen in compactness, so that was the new starting point for a map that took about 24 hours to produce.

I ignored all community boundaries except for those that cross county lines, and I made substantial use of the rule that allowed those cross-county cities to avoid creating county fragments, which eventually got me back to 9 fragments. I also allowed a number of districts to get thin and contorted so that I could push up the compactness for the other districts. This gets me to a compactness score of almost 0.033. I took 6 districts as uncompetitive with -1 scores but swapped pieces to get all others within the magic 5% for a competitiveness score of 18. I still have a 9-8-1 distribution and by luck the even district is exactly even.

I think 0.033 is exceptionally good.  I only have 3 districts that are that high, and only 4 over 0.030.  I suspect that the mean would be better than the median for scoring, since the maximum (for a circle) is only 0.079, and an approximation of a circle using a grid of square townships is 0.049, and the minimum is 0.000 there isn't too much problem with extreme values, but there would be some reward for avoiding them.

After I did my second plan, I had a gap of about 0.003 between my 9th and 10th best districts (ie the median didn't really characterize my districts well).  Simply by adjusting a single district, I was able to get a substantial improvement in the median.

The median can be quite different than the mean. My mean is about 10% lower than my median. I let five districts drop to a range from 0.0112 to 0.0176, but I used their contorted shapes to make four of those into +2 competitive districts, including my tied district. The fifth was a -1 district but surrounds the two compact districts involving Franklin. The remaining 11 districts were all above 0.028. In the end I could just concentrate on the shape of those districts involved with the median.

I don't like the shape of some of my districts, and my earlier map is better in that regard. The best formula might involve both the mean and standard deviation of the compactness.

[muon2]

Here is my map submitted to the competition and a link to the pdf with precinct lines.



As I noted earlier I used five municipalities that cross county lines to balance population and eliminate county fragments: Hunting Valley - D14 Cuyahoga; Minerva - D16 Carroll; Plain City - D7 Union; Columbus - D7 Fairfield; Milford - D2 Hamilton. This also eliminated the apparent single fragments left in those counties larger than a district. That left nine fragments - four in Summit, three in Montgomery, and 2 in Lucas.

The three counties split were critical to reach a competitiveness rating of 18, partisan balance of 9R-8D-1E within 0.8% of the ideal and a median compactness of 0.03295. The competitiveness required 6 districts at the worst ranking of -1 each, and the other 12 at the best possible score +2 each. Splitting Toledo in Lucas was needed to make both D5 and D9 Dem within 5%. The split between D10 and 11 in Cuyahoga was designed to keep both districts highly compact but make D10 just within the 5% margin for the Dems. Similarly the split in Franklin and Hamilton  kept those four districts very compact and made 2 R and 1 D very competitive districts. The Summit split was designed to make D14 and D16 just barely within 5% for the Dems while keeping them compact and it created an exact tie between the parties for D13.

Here is the complete table of parameters for the 18 districts:

District	Compactness	Competitiveness
D1	0.03478	0.62% R +2
D2	0.03245	31.76% R -1
D3	0.03442	1.22% R +2
D4	0.04674	32.30% R -1
D5	0.02811	-2.28% D +2
D6	0.01204	3.78% R +2
D7	0.03075	-3.08% D +2
D8	0.03485	32.54% R -1
D9	0.01121	-3.46% D +2
D10	0.03712	-4.96% D +2
D11	0.03344	-65.08% D -1
D12	0.01669	27.38% R -1
D13	0.01143	0.00% E +2
D14	0.03872	-5.00% D +2
D15	0.03912	1.78% R +2
D16	0.03763	-4.96% D +2
D17	0.01763	-3.50% D +2
D18	0.02950	15.56% R -1

[jimrtex]

Here is my map submitted to the competition and a link to the pdf with precinct lines.

It is interesting that D7 and D16 have such good compactness scores.  D7 is about what you might get with 3 ordinary counties stuck together.

D6 and D18 illustrate problems with the compactness measurement.  D6 eats up all the Ohio River in a low scoring district, while D18 builds up a lot of area in the whole counties that can balance the perimeter in Summit County.

Is it feasible to drop the Democratic percentage of D14 and D18, or were you just barely able to get under 5%?  What is the population split of Summit County between the districts?

[muon2]

It is interesting that D7 and D16 have such good compactness scores.  D7 is about what you might get with 3 ordinary counties stuck together.

D6 and D18 illustrate problems with the compactness measurement.  D6 eats up all the Ohio River in a low scoring district, while D18 builds up a lot of area in the whole counties that can balance the perimeter in Summit County.

Is it feasible to drop the Democratic percentage of D14 and D18, or were you just barely able to get under 5%?  What is the population split of Summit County between the districts?

District	Compactness	Competitiveness	%pop Summit	% Dem Summit
D13	0.01143	0.00% E +2	22%	54%
D14	0.03872	-5.00% D +2	11%	41%
D16	0.03763	-4.96% D +2	46%	58%
D18	0.02950	15.56% R -1	21%	61%

I added the Summit info to my table for the 4 districts it supports. D18 is a in some sense a throw away district, since it figured neither into competitiveness nor compactness. I used it to absorb Dem votes in Akron from the other three districts.

I probably could have swapped Dem votes from D16 to D18 by stretching D18 east through more of Akron and putting areas from Barberton into D16. D13 had little wiggle room in its precincts since it was my even district for voting, though I could alter the shape as much as I liked.

D14 was my toughest problem. D14 was too Dem without Summit so I had to add a majority of Rep votes there, but I couldn't stretch across the to western Summit without hurting compactness. The addition of Hunting Valley in Cuyahoga was critical here. It was a net 122 Rep votes to D14, it added 590 people to D14 that ended up at -0.43% population, and it triggered the county crossing rule so that the part of D13 in Cuyahoga no longer counted as a fragment.

Your observations on compactness are quite correct. The Ohio river bends would hurt any district's compactness, so it was best to lump them all into one district and aim to make it competitive instead. I spent a lot of time on Franklin to find the ideal split since either D7 or D15 could be majority Dem. Because D15 was smaller it was more sensitive to wiggles in it's boundary, so that if I added the Fairfield part of Columbus to D15 it would drop substantially in compactness, where the effect of D7 was much less.

[muon2]

As I look at Summit more, I wonder if I couldn't improve my score. If I pull D18 across central Akron and shift D16 to sit more in Barberton and SW Summit I might put some of the D's from D16 into D18. What I wonder is whether I could keep D16 above 0.034 compactness while shifting D18 to under 15% R. That would increase competitiveness to 19 while holding the compactness at the same value.

Unfortunately the account is locked out now that the competition is closed, so I can't test my hypothesis.

[jimrtex]

As I look at Summit more, I wonder if I couldn't improve my score. If I pull D18 across central Akron and shift D16 to sit more in Barberton and SW Summit I might put some of the D's from D16 into D18. What I wonder is whether I could keep D16 above 0.034 compactness while shifting D18 to under 15% R. That would increase competitiveness to 19 while holding the compactness at the same value.

Unfortunately the account is locked out now that the competition is closed, so I can't test my hypothesis.

There is an a really ugly precinct just below the SW corner of Barberton.  Going west around it, you alnost pinch off D18, and then swinging back across central Akron you may push D16 compactness way down.

Akron appears to have a lot of block jitter in its precinct boundaries, probably caused by Ohio's tight limits on precinct sizes.  1400 is the maximum that a local election authority can use (vs 5000 in Texas).  So when you get close to a limit, you either have to shift the boundaries, or split a precinct in two precincts that are closer to around 1/2 the limit.

In 2008, the average registration per precinct in Akron was 887 per precinct.

ps Summit County Board of Elections has a pretty neat set of nested election maps for county, cities/towns, wards, and precincts with streets.

You might be able to do better by swapping areas between D13 and D18, depending on how close you are to the 0.05% population thresholds.  There are a lot of places that you can look for Democrats, including Cuyahoga County to bring D13 back into a tie (especially if you are willing to accept less than a perfect tie, my "E" district has a 7-vote plurality).

[muon2]

As I look at Summit more, I wonder if I couldn't improve my score. If I pull D18 across central Akron and shift D16 to sit more in Barberton and SW Summit I might put some of the D's from D16 into D18. What I wonder is whether I could keep D16 above 0.034 compactness while shifting D18 to under 15% R. That would increase competitiveness to 19 while holding the compactness at the same value.

Unfortunately the account is locked out now that the competition is closed, so I can't test my hypothesis.

There is an a really ugly precinct just below the SW corner of Barberton.  Going west around it, you alnost pinch off D18, and then swinging back across central Akron you may push D16 compactness way down.

Akron appears to have a lot of block jitter in its precinct boundaries, probably caused by Ohio's tight limits on precinct sizes.  1400 is the maximum that a local election authority can use (vs 5000 in Texas).  So when you get close to a limit, you either have to shift the boundaries, or split a precinct in two precincts that are closer to around 1/2 the limit.

In 2008, the average registration per precinct in Akron was 887 per precinct.

ps Summit County Board of Elections has a pretty neat set of nested election maps for county, cities/towns, wards, and precincts with streets.

You might be able to do better by swapping areas between D13 and D18, depending on how close you are to the 0.05% population thresholds.  There are a lot of places that you can look for Democrats, including Cuyahoga County to bring D13 back into a tie (especially if you are willing to accept less than a perfect tie, my "E" district has a 7-vote plurality).

I don't think I can get any help in Cuyahoga. I had very little flexibility in D10 and D11 to make both above 0.034 compactness and get D10 to be competitiveness +2. A swap of only two precincts between D11 and D13 is enough to bring down D11's compactness. That's the problem with this definition of compactness on small area districts.

[minionofmidas]

I understand how it came about, but I shudder to look at your 13th district.

[muon2]

I understand how it came about, but I shudder to look at your 13th district.

It's certainly not as bad as IL 04 or IL 17.

[minionofmidas]

I understand how it came about, but I shudder to look at your 13th district.

It's certainly not as bad as IL 04 or IL 17.

Damning yourself with faint praise?

[muon2]

Just in ...

You must be logged in to read this quote.

[doktorb]

As a Brit not accustomed to the way redistricting is carried out in the US, this thread and others like it is a confusing and bewildering education!

[Brittain33]

I just got the announcement with the three winning plans. Congratulations, muon2!

[muon2]

I just got the announcement with the three winning plans. Congratulations, muon2!

Thanks, I have to note that they somehow picked up my next to last map submission which doesn't match my signature. I've asked them to correct that or tell me why they picked what they did.