Redistricter Discussion and Map Thread

[kwabbit]

If you guys weren't aware, @Redistricter on twitter has been developing a new redistricting site, an analog to DRA. It's in a closed beta stage right now, so you have to request access, but it looks to be much more powerful than DRA already. It will include far more data, such as elections before 2016 and demographic data on education, income, sex, industry, etc. It does cost $5 per month, but that's well worth it to people who use DRA a lot.

I was hoping this thread could be used to show off some of the unique features of the app since I'll be playing with a lot.

Below is a map of Colorado with equal numbers of bachelor degree holders, 214k. The most populous is 1.1 million, the least is 530k. I chose Colorado to do this with first since it's easier to manage than Texas. Right now the app has CO, TX, and DE, but TX is the state with the most data.

[kwabbit]

Here is Georgia with an equal number Bach degree holders per district- about 170k each. North Fulton and Mid Fulton + North DeKalb are 67 and 69% college educated respectively, only needing 360k and 334k to hit the quota. South Central GA needs 1359k to hit the quota, at 18% college educated. Since all of the Atlanta districts had to contract, this goes 8-6 Dem in 2022.

[kwabbit]

Yeah some of this redistricr stuff seems a bit sus; are these hard stats or just estimates?

Having election data going back to 2000 is pretty cool though.

I'd assume it's ACS. The five year estimates are pretty accurate typically for areas of about 5k people. I don't know how the precinct data is generated, but it shouldn't be much off. Not more so than a few year old census. The census isn't even a headcount either. A lot of it uses regression techniques that are more accurate than trying to get a headcount.

[kwabbit]

I don't remember asking people about their highest level of college completed when I worked for the Census Bureau. Where does this precise data come from?

I do appreciate including 2012 numbers, I think DRA used to have that but threw it away for no good reason (for every state but Kansas for some reason). What other elections are included before 2016?

Most of the alternate data probably comes from the ACS. It's the Census Bureau's primary product at this point.

[kwabbit]

Here is Georgia with an equal number Bach degree holders per district- about 170k each. North Fulton and Mid Fulton + North DeKalb are 67 and 69% college educated respectively, only needing 360k and 334k to hit the quota. South Central GA needs 1359k to hit the quota, at 18% college educated. Since all of the Atlanta districts had to contract, this goes 8-6 Dem in 2022.

Is it even possible to draw a R gerrymander yielding a majority for Republicans under this sort of paradigm?

An 8-6 is pretty simple by mixing around districts 12 and 6 in North Atlanta and 2, 8, and 1 in South GA. Probably wouldn't hold the decade though.

[kwabbit]

I’d be interested to see the maximum woman district and maximum man district.

Right now there’s only 4 states. In TX you can make a 57% male district by collecting all of the ranching and oil rigs. Maybe in CA which isn’t added yet you could make a strongly Male one from farm workers in the CV.

[kwabbit]

I'm doing TX with equal numbers of bachelor's degree holders and holy hell does Houston have extreme socioeconomic segregation. The Houston arrow is about 70% college educated and the rest of the city is about 15%.

[kwabbit]

Here's my attempt at Texas with equal numbers of Bach holders, about 158k per. This was much more challenging, both because 38 districts is difficult and the fact that the app got more and more unstable as I progressed. There is no autosave at the moment, so I had to manually save quite a bit which is time consuming since you have wait. Otherwise at the end I could go only go a few minutes before it crashed; it was a lot easier in the beginning. I hope that this is fixed.

Urban Texas has pretty crazy socioeconomic segregation. The low-ed district that comprises most of Houston is 14% college educated, needing 1.817 million to hit the quota, while the two Houston arrow districts are 65 and 69% college educated for districts 37 and 38 respectively, needing only 325k to hit the quota.

For the following it's Bush 00, Obama 12

Uber-educated districts:

38- South Houston Arrow, Biden +41, Obama +12, Gore + 1, 69% CE
37- North Houston Arrow- Biden +11, Romney +22, Bush +34, 65% CE
35- Fort Bend, Trump +7, Romney +35, Bush +47, 54% CE
31- The Woodlands area, Trump +27, Romney +53, Bush +59, 49% CE
25- Round Rock area, Biden +6, Romney +20, Bush +43, 49% CE
24- Austin Core, Biden +56, Obama +32, Gore +2, 73% CE
22- West Travis, Biden +28, Obama +1, Bush +28, 64% CE
17- North Bexar, Biden +2, Romney +24, Bush +38, 48% CE
09- Northeast Tarrant, Trump +13, Romney +37, Bush +47, 50% CE
08- Southeast Denton, Trump +2, Romney +30, Bush +46, 51% CE
06- North Collin, Trump +7, Romney +37, Bush +57, 58% CE
05- South Collin, Biden +10, Romney +23, Bush +47, 59% CE
03- North Central Dallas, Biden +20, Romney +12, Bush +24, 66% CE
02- Northwest Dallas, Biden +17, Romney +15, Bush +37, 57% CE

In some favored quarters Romney held up pretty well. In others not so much. Northeast Collin is the area with the most favorable GOP trends. It's also the richest and the lowest ed of these DFW districts.

Overall the partisanship is 21-17 Trump-Biden which is what a fair map with normal sized districts would yield. All of the minority districts combine and instead there are smaller suburban districts. Some of these would've made very interesting congressional races. District 37 could probably still be competitive for congressional Republicans, as could districts 25, 17, 05, 03, and 02. These area are places where Cornyn and Abbott ran most ahead of Trump. District 37 was Hegar +1, district 25 Hegar +1, Cornyn +6, Hegar +2, Hegar +8, Hegar +8.

[kwabbit]

Does anyone know why Inner NOVA trended right in 2020? By this I mean Arlington + Alexandria + Falls Church area + McLean. Splitting the core NOVA counties (Loudoun, Fairfax, PW and all those that lie within) in two, you can yield a half where Trump got a higher percentage of the two party vote in 2020 than he did in 2016. That area is basically Inner Nova + east PWC.

It's not like the two halves are all that much different.

Half one: Clinton +51, Biden +50, 15% Black, 14% Asian, 19% Latino, 61% College Educated, $154k MHI

Half two: Clinton +23, Biden +32, 9% Black, 19% Asian, 16% Latino, 60% College Educated, $175k MHI.

Obviously the first half is more Black/Latino and less Asian, but not hugely. It has higher educational attainment and lower income, which would usually cause a greater D swing.

My thought is that Clinton really embodied a well off professional type and that played well in the rich areas of NOVA plus she did better with non-Whites. This is a place that would've swung basically everywhere else though.

[kwabbit]

Does anyone know why Inner NOVA trended right in 2020? By this I mean Arlington + Alexandria + Falls Church area + McLean. Splitting the core NOVA counties (Loudoun, Fairfax, PW and all those that lie within) in two, you can yield a half where Trump got a higher percentage of the two party vote in 2020 than he did in 2016. That area is basically Inner Nova + east PWC.

It's not like the two halves are all that much different.

Half one: Clinton +51, Biden +50, 15% Black, 14% Asian, 19% Latino, 61% College Educated, $154k MHI

Half two: Clinton +23, Biden +32, 9% Black, 19% Asian, 16% Latino, 60% College Educated, $175k MHI.

Obviously the first half is more Black/Latino and less Asian, but not hugely. It has higher educational attainment and lower income, which would usually cause a greater D swing.

My thought is that Clinton really embodied a well off professional type and that played well in the rich areas of NOVA plus she did better with non-Whites. This is a place that would've swung basically everywhere else though.

Where are you seeing this area trended right?

Alexandria
Clinton +58.1
Biden +62.7

Arlington
Clinton +59.2
Biden +63.5

Falls Church
Clinton +57.9
Biden +64.1

I didn't look up McLean, but the other three all trended more towards Biden than the national #'s

Of the two party vote. A little misleading, but 76-19 is the same two party as 80-20. Redistricter uses two party.

[kwabbit]

Overall def enjoy Redistricter. Def has more raw data than DRA, and also the option to draw by block groups is rlly nice for states with awful precincts. It's also nice how you can display by density and stuff. And the multi-state mapping has a lot of potential.

My main complaints so far:

1. It's just slower than DRA. This will probably get worked out overtime as they optimize the software and get better servers. Seriously though, trying to county-fill large counties can take over a minute.
2. No ability to color districts by things such as partisan lean, or to custom-color districts. Honestly, if there was a way to insert a csv file to custom color all districts at once, that would be boss.
3. No ability to add overlays onto an existing map. Honestly, this is arguably one of DRA's strong suits. I think if Redistricter created a better system for building a national map while drawing each state individually, that would be really cool.

These are all things I'm sure they're considering and have been or will work on.

Right now, I'd say Redistricter is better if you want to go into a detailed analysis cause there are so many more data sets, but DRA is still better if you just want mess around building maps and see the partisan result and whatnot.

Most DRA map obsessed people have done everything there is to do at this point. Redistricter adds so many new possibilities because of the additional data. DRA is barely adding anything at this point. They are past the point where they would add education or ancestry.

It is quite slow though. The multi state function is difficult when a big state on its own is slow.

[kwabbit]

Overall def enjoy Redistricter. Def has more raw data than DRA, and also the option to draw by block groups is rlly nice for states with awful precincts. It's also nice how you can display by density and stuff. And the multi-state mapping has a lot of potential.

My main complaints so far:

1. It's just slower than DRA. This will probably get worked out overtime as they optimize the software and get better servers. Seriously though, trying to county-fill large counties can take over a minute.
2. No ability to color districts by things such as partisan lean, or to custom-color districts. Honestly, if there was a way to insert a csv file to custom color all districts at once, that would be boss.
3. No ability to add overlays onto an existing map. Honestly, this is arguably one of DRA's strong suits. I think if Redistricter created a better system for building a national map while drawing each state individually, that would be really cool.

These are all things I'm sure they're considering and have been or will work on.

Right now, I'd say Redistricter is better if you want to go into a detailed analysis cause there are so many more data sets, but DRA is still better if you just want mess around building maps and see the partisan result and whatnot.

Most DRA map obsessed people have done everything there is to do at this point. Redistricter adds so many new possibilities because of the additional data. DRA is barely adding anything at this point. They are past the point where they would add education or ancestry.

It is quite slow though. The multi state function is difficult when a big state on its own is slow.

Ye. The only data I could want at this point is just more older (pre-2016) election results. Having 2012 results for every state would be insane.

Also 2010 and if possible 2000 census data would be amazing.

I hope he is able to make it faster as his first priority. Redistricter is more laborious because of the slowness, although I push through because the data is interesting. I wonder if there’s a limit on how well it could work on the web, given that DRA is pretty slow for California as well. Maybe a desktop version could make things faster.

[kwabbit]

1990, 2000, and 2010 census data was added. It has so much more capability than DRA at this point, but I need a better computer to run it lol. Or the founder needs a better server. He's been working on it on a laptop.

[kwabbit]

I subscribed to redistricter a few months back and its worth it for all the data. The only thing annoying is that I have to use a VPN to use the data because my network is blocking the server that Colin stores the data. I was told that he plans to change it to a different server in a few months. The only state thats buggy for me is California.

That's interesting. I am able to use Redistricter on my old, personal laptop but not on my faster work laptop, that might be the reason. DRA works on both.

[kwabbit]

I think I developed a pretty good baseline setting for the new "election shifter" tool, based on the available polling (regressing the shifts among Black voters a tad).

Dem Shift%/Turnout Shift%:

White: 2.5/0
Black: -6.0/-10.0
Asian: -6.0/-5.0
Native: -6.0/-5.0
Pacific: -6.0/-5.0
Multiracial: -6.0/-5.0
Other: -6.0/-5.0

No High School: -7.0/-5.0
High School: -7.0/-5.0
Bachelors: 10.0/0
Graduate: 10.0/0

The result is an election that's probably Biden +1.5 or so nationally and Biden +.25 in the tipping point of WI. WI, PA, MI are all about on the knife's edge for Biden, Sunbelt swing states go for Trump.

Here's a county sample:

Milwaukee: Biden +40 -> Biden +37
Dane: Biden +53 -> Biden +59
Waukesha: Trump +21 -> Trump +16
Ozaukee: Trump +12 -> Trump +5
Washington: Trump +38 -> Trump +37
Kenosha: Trump +3 -> Trump +5

I was also playing around with a scenario closer to the Trump +2 super racedep polls and the results are startlingly bad in the Sunbelt for Biden. Not so bad in the Midwestern swing states but blowouts in GA, TX, AZ, NV, NC, FL.

[kwabbit]

Using the 2008 Data in IL I made a congressional district that was 99.5% Obama 0.4% McCain. 286,684 votes to 1,267. You could make an assembly district within that where it 99.9 to 0.1. Crazy margins, almost unbelievable.

In 2020 it was 95.6-3.7. 205k to 8k. Kind of illustrates the difference the small persuasion gains and turnout drops can have. 285k margin to 197k margin is quite a fall. 7 point swing right is small but also significant.

[kwabbit]

Using the 2008 Data in IL I made a congressional district that was 99.5% Obama 0.4% McCain. 286,684 votes to 1,267. You could make an assembly district within that where it 99.9 to 0.1. Crazy margins, almost unbelievable.

In 2020 it was 95.6-3.7. 205k to 8k. Kind of illustrates the difference the small persuasion gains and turnout drops can have. 285k margin to 197k margin is quite a fall. 7 point swing right is small but also significant.

Must be Chicago's South or West side, right?

South Side, stayed within city limits. Taking in some of the southern Black suburbs or making a tentacle over to the West Side could've brought it to 99.7% but I kept it compact.

[kwabbit]

Is redistricter better optimized/does it run faster? I love it but one of the reasons I didn't renew it was that is was very slow, laggy and prone to crashing on chrome.

That’s definitely the biggest downside vs. DRA. It works better on a fast computer (obviously), but there’s also some tricks. I’ve found that selecting precincts via the box tool causes way less lag than individually clicking or using the brush. The only state where I have crashing is California, but that’s slow af on DRA as well.

[kwabbit]

Is redistricter better optimized/does it run faster? I love it but one of the reasons I didn't renew it was that is was very slow, laggy and prone to crashing on chrome.

That’s definitely the biggest downside vs. DRA. It works better on a fast computer (obviously), but there’s also some tricks. I’ve found that selecting precincts via the box tool causes way less lag than individually clicking or using the brush. The only state where I have crashing is California, but that’s slow af on DRA as well.

California is unfortunately quite massive so making it performant will always be a big challenge.

What's the bottleneck, if I might ask? If I had a $4k computer would it run smoothly, or is it a limitation of the web browser, or a server limitation?

[kwabbit]

Out of curiosity, could you see yourself adding ACS ancestry data? It would be neat if one could create districts based of specific concentrations of ethnic groups.

It has Ancestry Data already, I believe from the ACS. The numbers don't add to 100% though, because it's single ancestry and excludes those who chose multiple options or some people just left that part blank while filling out the ACS. The estimates are also screwy in lower population areas, but that's just how it's going to be with a survey that doesn't intend to reach every person.