ASA Connect

Ideal US congressional districts untouched by human hands.  https://github.com/raagnew/IdealCongressionalDistricting

------------------------------
Robert Agnew
------------------------------

What is the utility function that this apportionment maximizes such that you call it "ideal"

------------------------------
Stefan Sharkansky, PhD
Useful Work, Inc.
Bellevue,WA
------------------------------

Original Message:
Sent: 10-03-2021 14:34
From: Robert Agnew
Subject: CONGRESSIONAL DISTRICTING

Ideal US congressional districts untouched by human hands.  https://github.com/raagnew/IdealCongressionalDistricting

------------------------------
Robert Agnew
------------------------------

Mathematically, it minimizes population-weighted total squared distance from census block groups to population-weighted cluster centroids, subject to cluster populations being approximately equal.  This constrained version of the well-known k-means clustering algorithm seems to fit the congressional districting problem very well.  Hence my "ideal" terminology.  I also consider it "ideal" in that the algorithm could potentially remove districting from the hands of politicians.

------------------------------
Robert Agnew
------------------------------

Original Message:
Sent: 10-04-2021 15:35
From: Stefan Sharkansky
Subject: CONGRESSIONAL DISTRICTING

What is the utility function that this apportionment maximizes such that you call it "ideal"

------------------------------
Stefan Sharkansky, PhD
Useful Work, Inc.
Bellevue,WA

Original Message:
Sent: 10-03-2021 14:34
From: Robert Agnew
Subject: CONGRESSIONAL DISTRICTING

Ideal US congressional districts untouched by human hands.  https://github.com/raagnew/IdealCongressionalDistricting

------------------------------
Robert Agnew
------------------------------

Hey Robert!

Thanks for sharing! I've recently become fascinated by this topic. I've actually been playing with a different approach, one that uses hierarchical clustering on principal components of demographic information. 

As an example, here's a start at a district map for NC using my method on county subdivisions.



It starts with computing principal components based on selected demographic data (i.e. percentage of males, percentage within census-defined age brackets, percentage within census-defined race groups, and percentage Hispanic for this example). I made sure to leave one group out with the percentages to avoid too strong of a correlation in the variables. I then selected the principal components accounting for at least 90% of the variation. From that, I created a weighted combination of the distance matrix for the PCs and a distance matrix from the county subdivision centroids (I used equal weighting here). Finally, I used hierarchical clustering with Ward's method to create 14 clusters, corresponding to the number of districts for NC. 

The weighted combination approach was inspired by this R package: https://cran.r-project.org/web/packages/ClustGeo/vignettes/intro_ClustGeo.html. I created this plot using JMP, but I've used R as well. While it's not necessarily as hands-free as your approach since someone would have to decide the number of principal components to use and the weighting scheme, I find it a nice balance in incorporating demographic information into the district creation, giving a nice display of the demographic distribution in the state, operating under the theory that districts should be representative of the state. Of course, it's not perfect, as you can see with the district "islands". Also, it currently does not attempt at keeping populations roughly equal among the clusters, something I'm still working on...

I'm curious as to your thoughts (and the thoughts of anyone else)! I'd especially be interested in any political ramifications of this approach, as I'm certain there are some. 

--------------------------
Caleb King, PhD
JMP
Raleigh, NC
-------------------------

------------------------------
Caleb King
JMP
------------------------------

Original Message:
Sent: 10-03-2021 14:34
From: Robert Agnew
Subject: CONGRESSIONAL DISTRICTING

Ideal US congressional districts untouched by human hands.  https://github.com/raagnew/IdealCongressionalDistricting

------------------------------
Robert Agnew
------------------------------

Thanks for sharing, Caleb.  I'm well aware that other algorithmic approaches to congressional districting have been developed in recent years.  Yours has some interesting additional features.  I like the geo-simplicity of the constrained k-means approach, however.

------------------------------
Robert Agnew
------------------------------

Original Message:
Sent: 10-04-2021 16:59
From: Caleb King
Subject: CONGRESSIONAL DISTRICTING

Hey Robert!

Thanks for sharing! I've recently become fascinated by this topic. I've actually been playing with a different approach, one that uses hierarchical clustering on principal components of demographic information.

As an example, here's a start at a district map for NC using my method on county subdivisions.



It starts with computing principal components based on selected demographic data (i.e. percentage of males, percentage within census-defined age brackets, percentage within census-defined race groups, and percentage Hispanic for this example). I made sure to leave one group out with the percentages to avoid too strong of a correlation in the variables. I then selected the principal components accounting for at least 90% of the variation. From that, I created a weighted combination of the distance matrix for the PCs and a distance matrix from the county subdivision centroids (I used equal weighting here). Finally, I used hierarchical clustering with Ward's method to create 14 clusters, corresponding to the number of districts for NC.

The weighted combination approach was inspired by this R package: https://cran.r-project.org/web/packages/ClustGeo/vignettes/intro_ClustGeo.html. I created this plot using JMP, but I've used R as well. While it's not necessarily as hands-free as your approach since someone would have to decide the number of principal components to use and the weighting scheme, I find it a nice balance in incorporating demographic information into the district creation, giving a nice display of the demographic distribution in the state, operating under the theory that districts should be representative of the state. Of course, it's not perfect, as you can see with the district "islands". Also, it currently does not attempt at keeping populations roughly equal among the clusters, something I'm still working on...

I'm curious as to your thoughts (and the thoughts of anyone else)! I'd especially be interested in any political ramifications of this approach, as I'm certain there are some.

--------------------------
Caleb King, PhD
JMP
Raleigh, NC
-------------------------

------------------------------
Caleb King
JMP

Original Message:
Sent: 10-03-2021 14:34
From: Robert Agnew
Subject: CONGRESSIONAL DISTRICTING

Ideal US congressional districts untouched by human hands.  https://github.com/raagnew/IdealCongressionalDistricting

------------------------------
Robert Agnew
------------------------------

Interesting ideas with goals consistent with my musing (wishful thinking?) concerning how to adjust the representation of those elected, right up there with term limits, longer terms, and accompanying salary increases in order to reduce the incidence of building political machines.

Some years ago a US representative I knew posed a challenge to me (knowing that was a statistician and therefore assuming I knew all things mathematical): Can you come up with a algorithm that can be objectively applied that minimizes boundary irregularities of congressional districts? My naïve thought of a potential solution was a score reflecting unacceptable boundary irregularity.

I lacked the time, and most importantly the mathematical skills, to undertake such. Perhaps someone else has already done this or feels challenged by the goals of the challenge.

------------------------------
Brent Blumenstein
------------------------------

Original Message:
Sent: 10-04-2021 20:58
From: Robert Agnew
Subject: CONGRESSIONAL DISTRICTING

Thanks for sharing, Caleb.  I'm well aware that other algorithmic approaches to congressional districting have been developed in recent years.  Yours has some interesting additional features.  I like the geo-simplicity of the constrained k-means approach, however.

------------------------------
Robert Agnew

Original Message:
Sent: 10-04-2021 16:59
From: Caleb King
Subject: CONGRESSIONAL DISTRICTING

Hey Robert!

Thanks for sharing! I've recently become fascinated by this topic. I've actually been playing with a different approach, one that uses hierarchical clustering on principal components of demographic information.

As an example, here's a start at a district map for NC using my method on county subdivisions.



It starts with computing principal components based on selected demographic data (i.e. percentage of males, percentage within census-defined age brackets, percentage within census-defined race groups, and percentage Hispanic for this example). I made sure to leave one group out with the percentages to avoid too strong of a correlation in the variables. I then selected the principal components accounting for at least 90% of the variation. From that, I created a weighted combination of the distance matrix for the PCs and a distance matrix from the county subdivision centroids (I used equal weighting here). Finally, I used hierarchical clustering with Ward's method to create 14 clusters, corresponding to the number of districts for NC.

The weighted combination approach was inspired by this R package: https://cran.r-project.org/web/packages/ClustGeo/vignettes/intro_ClustGeo.html. I created this plot using JMP, but I've used R as well. While it's not necessarily as hands-free as your approach since someone would have to decide the number of principal components to use and the weighting scheme, I find it a nice balance in incorporating demographic information into the district creation, giving a nice display of the demographic distribution in the state, operating under the theory that districts should be representative of the state. Of course, it's not perfect, as you can see with the district "islands". Also, it currently does not attempt at keeping populations roughly equal among the clusters, something I'm still working on...

I'm curious as to your thoughts (and the thoughts of anyone else)! I'd especially be interested in any political ramifications of this approach, as I'm certain there are some.

--------------------------
Caleb King, PhD
JMP
Raleigh, NC
-------------------------

------------------------------
Caleb King
JMP

Original Message:
Sent: 10-03-2021 14:34
From: Robert Agnew
Subject: CONGRESSIONAL DISTRICTING

Ideal US congressional districts untouched by human hands.  https://github.com/raagnew/IdealCongressionalDistricting

------------------------------
Robert Agnew
------------------------------

Thanks Robert! The constrained k-means is certainly straightforward, which I like. I think there are still some areas where human intervention could affect the outcome, though perhaps not as easily as the current system. The one area that stands out is the use of census blocks; these can easily change from census to census (and have done so in the past). This could create artificial changes in the districts not necessarily related to population. Also, how might the districts change if blocks were used instead? Or census tracts? The appropriate level of granularity would need to be agreed upon if there were significant changes to the districts, which would require human intervention to ultimately decide. 

On a related note, after seeing your district maps, it reminded me of similar results I had seen online. I did some searching and re-located this work done by fivethirtyeight on different approaches to creating district maps and their political ramifications. One of them is very similar to the constrained k-means results you show. They also reference another website (see here) which explores several different approaches as well along with much more detailed analyses.

------------------------------
Caleb King
JMP
------------------------------

Original Message:
Sent: 10-04-2021 20:58
From: Robert Agnew
Subject: CONGRESSIONAL DISTRICTING

Thanks for sharing, Caleb.  I'm well aware that other algorithmic approaches to congressional districting have been developed in recent years.  Yours has some interesting additional features.  I like the geo-simplicity of the constrained k-means approach, however.

------------------------------
Robert Agnew

Original Message:
Sent: 10-04-2021 16:59
From: Caleb King
Subject: CONGRESSIONAL DISTRICTING

Hey Robert!

Thanks for sharing! I've recently become fascinated by this topic. I've actually been playing with a different approach, one that uses hierarchical clustering on principal components of demographic information.

As an example, here's a start at a district map for NC using my method on county subdivisions.



It starts with computing principal components based on selected demographic data (i.e. percentage of males, percentage within census-defined age brackets, percentage within census-defined race groups, and percentage Hispanic for this example). I made sure to leave one group out with the percentages to avoid too strong of a correlation in the variables. I then selected the principal components accounting for at least 90% of the variation. From that, I created a weighted combination of the distance matrix for the PCs and a distance matrix from the county subdivision centroids (I used equal weighting here). Finally, I used hierarchical clustering with Ward's method to create 14 clusters, corresponding to the number of districts for NC.

The weighted combination approach was inspired by this R package: https://cran.r-project.org/web/packages/ClustGeo/vignettes/intro_ClustGeo.html. I created this plot using JMP, but I've used R as well. While it's not necessarily as hands-free as your approach since someone would have to decide the number of principal components to use and the weighting scheme, I find it a nice balance in incorporating demographic information into the district creation, giving a nice display of the demographic distribution in the state, operating under the theory that districts should be representative of the state. Of course, it's not perfect, as you can see with the district "islands". Also, it currently does not attempt at keeping populations roughly equal among the clusters, something I'm still working on...

I'm curious as to your thoughts (and the thoughts of anyone else)! I'd especially be interested in any political ramifications of this approach, as I'm certain there are some.

--------------------------
Caleb King, PhD
JMP
Raleigh, NC
-------------------------

------------------------------
Caleb King
JMP

Original Message:
Sent: 10-03-2021 14:34
From: Robert Agnew
Subject: CONGRESSIONAL DISTRICTING

Ideal US congressional districts untouched by human hands.  https://github.com/raagnew/IdealCongressionalDistricting

------------------------------
Robert Agnew
------------------------------

Thanks again, Caleb.  I had seen Dave's work in the past and his algorithm may very well be similar to mine.  The reason I focused on census block groups, between census tracts and minute census blocks, is that they offered granularity, but not too much, and Census offered both population and longitude-latitude geo-coordinates for them.  I understand that things can change from census to census, but you have to have a stake in the ground somewhere.  In many of the states, census block groups average about 1,400 in population.  It would be easy to adapt my algorithm to census tracts.  Going down to census blocks is another matter.  My linear programming (transportation version) sub-algorithm would probably be flummoxed.  There is another way to approach it, but I didn't pursue it since this level of granularity didn't seem warranted.  Even with census block groups, my algorithm ran for 13 hours on California, probably because of memory issues on my PC.

I'll be traveling and offline for the next few days.

------------------------------
Robert Agnew
------------------------------

Original Message:
Sent: 10-05-2021 11:50
From: Caleb King
Subject: CONGRESSIONAL DISTRICTING

Thanks Robert! The constrained k-means is certainly straightforward, which I like. I think there are still some areas where human intervention could affect the outcome, though perhaps not as easily as the current system. The one area that stands out is the use of census blocks; these can easily change from census to census (and have done so in the past). This could create artificial changes in the districts not necessarily related to population. Also, how might the districts change if blocks were used instead? Or census tracts? The appropriate level of granularity would need to be agreed upon if there were significant changes to the districts, which would require human intervention to ultimately decide.

On a related note, after seeing your district maps, it reminded me of similar results I had seen online. I did some searching and re-located this work done by fivethirtyeight on different approaches to creating district maps and their political ramifications. One of them is very similar to the constrained k-means results you show. They also reference another website (see here) which explores several different approaches as well along with much more detailed analyses.

------------------------------
Caleb King
JMP

Original Message:
Sent: 10-04-2021 20:58
From: Robert Agnew
Subject: CONGRESSIONAL DISTRICTING

Thanks for sharing, Caleb.  I'm well aware that other algorithmic approaches to congressional districting have been developed in recent years.  Yours has some interesting additional features.  I like the geo-simplicity of the constrained k-means approach, however.

------------------------------
Robert Agnew

Original Message:
Sent: 10-04-2021 16:59
From: Caleb King
Subject: CONGRESSIONAL DISTRICTING

Hey Robert!

Thanks for sharing! I've recently become fascinated by this topic. I've actually been playing with a different approach, one that uses hierarchical clustering on principal components of demographic information.

As an example, here's a start at a district map for NC using my method on county subdivisions.



It starts with computing principal components based on selected demographic data (i.e. percentage of males, percentage within census-defined age brackets, percentage within census-defined race groups, and percentage Hispanic for this example). I made sure to leave one group out with the percentages to avoid too strong of a correlation in the variables. I then selected the principal components accounting for at least 90% of the variation. From that, I created a weighted combination of the distance matrix for the PCs and a distance matrix from the county subdivision centroids (I used equal weighting here). Finally, I used hierarchical clustering with Ward's method to create 14 clusters, corresponding to the number of districts for NC.

The weighted combination approach was inspired by this R package: https://cran.r-project.org/web/packages/ClustGeo/vignettes/intro_ClustGeo.html. I created this plot using JMP, but I've used R as well. While it's not necessarily as hands-free as your approach since someone would have to decide the number of principal components to use and the weighting scheme, I find it a nice balance in incorporating demographic information into the district creation, giving a nice display of the demographic distribution in the state, operating under the theory that districts should be representative of the state. Of course, it's not perfect, as you can see with the district "islands". Also, it currently does not attempt at keeping populations roughly equal among the clusters, something I'm still working on...

I'm curious as to your thoughts (and the thoughts of anyone else)! I'd especially be interested in any political ramifications of this approach, as I'm certain there are some.

--------------------------
Caleb King, PhD
JMP
Raleigh, NC
-------------------------

------------------------------
Caleb King
JMP

Original Message:
Sent: 10-03-2021 14:34
From: Robert Agnew
Subject: CONGRESSIONAL DISTRICTING

Ideal US congressional districts untouched by human hands.  https://github.com/raagnew/IdealCongressionalDistricting

------------------------------
Robert Agnew
------------------------------

I did some work about 10 years designing geographic clusters for sampling in the National Children's Study that has obvious applications to building Congressional Districts (should not be behind a paywall):

https://www.sciencedirect.com/science/article/pii/S1047279710004576?casa_token=ooiLPyTMFYUAAAAA:H-KzVFYdT6d7e2m0ljalkA_4Qcb1hyzddyJNE8hSJJ7G8za60sfK2eoOW751CEzJx8som4giiw

Your more recent work may be more sophisticated.  As a side note, Michigan has a citizen commission to draw non-partisan maps.  I tried to interest them in my work, but I think "untouched by human hands" was considered a bad thing.

------------------------------
Michael Elliott
University of Michigan
------------------------------

Original Message:
Sent: 10-04-2021 16:59
From: Caleb King
Subject: CONGRESSIONAL DISTRICTING

Hey Robert!

Thanks for sharing! I've recently become fascinated by this topic. I've actually been playing with a different approach, one that uses hierarchical clustering on principal components of demographic information.

As an example, here's a start at a district map for NC using my method on county subdivisions.



It starts with computing principal components based on selected demographic data (i.e. percentage of males, percentage within census-defined age brackets, percentage within census-defined race groups, and percentage Hispanic for this example). I made sure to leave one group out with the percentages to avoid too strong of a correlation in the variables. I then selected the principal components accounting for at least 90% of the variation. From that, I created a weighted combination of the distance matrix for the PCs and a distance matrix from the county subdivision centroids (I used equal weighting here). Finally, I used hierarchical clustering with Ward's method to create 14 clusters, corresponding to the number of districts for NC.

The weighted combination approach was inspired by this R package: https://cran.r-project.org/web/packages/ClustGeo/vignettes/intro_ClustGeo.html. I created this plot using JMP, but I've used R as well. While it's not necessarily as hands-free as your approach since someone would have to decide the number of principal components to use and the weighting scheme, I find it a nice balance in incorporating demographic information into the district creation, giving a nice display of the demographic distribution in the state, operating under the theory that districts should be representative of the state. Of course, it's not perfect, as you can see with the district "islands". Also, it currently does not attempt at keeping populations roughly equal among the clusters, something I'm still working on...

I'm curious as to your thoughts (and the thoughts of anyone else)! I'd especially be interested in any political ramifications of this approach, as I'm certain there are some.

--------------------------
Caleb King, PhD
JMP
Raleigh, NC
-------------------------

------------------------------
Caleb King
JMP

Original Message:
Sent: 10-03-2021 14:34
From: Robert Agnew
Subject: CONGRESSIONAL DISTRICTING

Ideal US congressional districts untouched by human hands.  https://github.com/raagnew/IdealCongressionalDistricting

------------------------------
Robert Agnew
------------------------------

In my view, "untouched by human hands" is the only way to get a fair, unbiased result.  We need a detached, sensible algorithm, whether it's mine or someone else's.

------------------------------
Robert Agnew
------------------------------

Original Message:
Sent: 10-05-2021 17:31
From: Michael Elliott
Subject: CONGRESSIONAL DISTRICTING

I did some work about 10 years designing geographic clusters for sampling in the National Children's Study that has obvious applications to building Congressional Districts (should not be behind a paywall):

https://www.sciencedirect.com/science/article/pii/S1047279710004576?casa_token=ooiLPyTMFYUAAAAA:H-KzVFYdT6d7e2m0ljalkA_4Qcb1hyzddyJNE8hSJJ7G8za60sfK2eoOW751CEzJx8som4giiw

Your more recent work may be more sophisticated.  As a side note, Michigan has a citizen commission to draw non-partisan maps.  I tried to interest them in my work, but I think "untouched by human hands" was considered a bad thing.

------------------------------
Michael Elliott
University of Michigan

Original Message:
Sent: 10-04-2021 16:59
From: Caleb King
Subject: CONGRESSIONAL DISTRICTING

Hey Robert!

Thanks for sharing! I've recently become fascinated by this topic. I've actually been playing with a different approach, one that uses hierarchical clustering on principal components of demographic information.

As an example, here's a start at a district map for NC using my method on county subdivisions.



It starts with computing principal components based on selected demographic data (i.e. percentage of males, percentage within census-defined age brackets, percentage within census-defined race groups, and percentage Hispanic for this example). I made sure to leave one group out with the percentages to avoid too strong of a correlation in the variables. I then selected the principal components accounting for at least 90% of the variation. From that, I created a weighted combination of the distance matrix for the PCs and a distance matrix from the county subdivision centroids (I used equal weighting here). Finally, I used hierarchical clustering with Ward's method to create 14 clusters, corresponding to the number of districts for NC.

The weighted combination approach was inspired by this R package: https://cran.r-project.org/web/packages/ClustGeo/vignettes/intro_ClustGeo.html. I created this plot using JMP, but I've used R as well. While it's not necessarily as hands-free as your approach since someone would have to decide the number of principal components to use and the weighting scheme, I find it a nice balance in incorporating demographic information into the district creation, giving a nice display of the demographic distribution in the state, operating under the theory that districts should be representative of the state. Of course, it's not perfect, as you can see with the district "islands". Also, it currently does not attempt at keeping populations roughly equal among the clusters, something I'm still working on...

I'm curious as to your thoughts (and the thoughts of anyone else)! I'd especially be interested in any political ramifications of this approach, as I'm certain there are some.

--------------------------
Caleb King, PhD
JMP
Raleigh, NC
-------------------------

------------------------------
Caleb King
JMP

Original Message:
Sent: 10-03-2021 14:34
From: Robert Agnew
Subject: CONGRESSIONAL DISTRICTING

Ideal US congressional districts untouched by human hands.  https://github.com/raagnew/IdealCongressionalDistricting

------------------------------
Robert Agnew
------------------------------

I agree 100%.  But census counts and district boundaries are at the core of political survival for all legislative politicians (excepting the US Senate) regardless of party or ideology, so it is almost impossible to keep their "hands off".  Moving the task to so-called non-partisan commissions doesn't fully ameliorate these tendencies.

------------------------------
Michael Elliott
University of Michigan
------------------------------

Original Message:
Sent: 10-06-2021 07:31
From: Robert Agnew
Subject: CONGRESSIONAL DISTRICTING

In my view, "untouched by human hands" is the only way to get a fair, unbiased result.  We need a detached, sensible algorithm, whether it's mine or someone else's.

------------------------------
Robert Agnew

Original Message:
Sent: 10-05-2021 17:31
From: Michael Elliott
Subject: CONGRESSIONAL DISTRICTING

I did some work about 10 years designing geographic clusters for sampling in the National Children's Study that has obvious applications to building Congressional Districts (should not be behind a paywall):

https://www.sciencedirect.com/science/article/pii/S1047279710004576?casa_token=ooiLPyTMFYUAAAAA:H-KzVFYdT6d7e2m0ljalkA_4Qcb1hyzddyJNE8hSJJ7G8za60sfK2eoOW751CEzJx8som4giiw

Your more recent work may be more sophisticated.  As a side note, Michigan has a citizen commission to draw non-partisan maps.  I tried to interest them in my work, but I think "untouched by human hands" was considered a bad thing.

------------------------------
Michael Elliott
University of Michigan

Original Message:
Sent: 10-04-2021 16:59
From: Caleb King
Subject: CONGRESSIONAL DISTRICTING

Hey Robert!

Thanks for sharing! I've recently become fascinated by this topic. I've actually been playing with a different approach, one that uses hierarchical clustering on principal components of demographic information.

As an example, here's a start at a district map for NC using my method on county subdivisions.



It starts with computing principal components based on selected demographic data (i.e. percentage of males, percentage within census-defined age brackets, percentage within census-defined race groups, and percentage Hispanic for this example). I made sure to leave one group out with the percentages to avoid too strong of a correlation in the variables. I then selected the principal components accounting for at least 90% of the variation. From that, I created a weighted combination of the distance matrix for the PCs and a distance matrix from the county subdivision centroids (I used equal weighting here). Finally, I used hierarchical clustering with Ward's method to create 14 clusters, corresponding to the number of districts for NC.

The weighted combination approach was inspired by this R package: https://cran.r-project.org/web/packages/ClustGeo/vignettes/intro_ClustGeo.html. I created this plot using JMP, but I've used R as well. While it's not necessarily as hands-free as your approach since someone would have to decide the number of principal components to use and the weighting scheme, I find it a nice balance in incorporating demographic information into the district creation, giving a nice display of the demographic distribution in the state, operating under the theory that districts should be representative of the state. Of course, it's not perfect, as you can see with the district "islands". Also, it currently does not attempt at keeping populations roughly equal among the clusters, something I'm still working on...

I'm curious as to your thoughts (and the thoughts of anyone else)! I'd especially be interested in any political ramifications of this approach, as I'm certain there are some.

--------------------------
Caleb King, PhD
JMP
Raleigh, NC
-------------------------

------------------------------
Caleb King
JMP

Original Message:
Sent: 10-03-2021 14:34
From: Robert Agnew
Subject: CONGRESSIONAL DISTRICTING

Ideal US congressional districts untouched by human hands.  https://github.com/raagnew/IdealCongressionalDistricting

------------------------------
Robert Agnew
------------------------------