2020 Census and Redistricting Thread: Michigan

[BoiseBoy]

Why is it so hard for them to draw a good map?

[Gass3268]

Why is it so hard for them to draw a good map?

Because they made the stupid decision to disregard partisan data until the end, even though it is constitutionally more important that boundaries and compactness.

[lfromnj]

The best thing to do regarding partisan data is to use a Markov Chain Monte Carlo simulation and just make sure the map lies within the middle 33% of all maps IMO. That way no communities have to be broken up while you can still follow the law. Colorado did at least listen to this idea late in the commission. Probably too late for that by now though. I guess you would also have to program the simulation to draw 2 45% black seats.

[President Punxsutawney Phil]

Hello, and welcome to Amateur Hour with the Michigan Independent Commission!

[Adam Griffin]

Looks like you've only got one black VRA district, which is illegal.

Oh well, trade portions of 12 and 13 a bit (which actually makes their respective boundaries neater) and you can easily get 2 45% BVAP districts. Should be plenty enough in a state like MI to be considered a black-performing district. GOP obviously packed the hell out of the 2 black districts in 2011.

[Gass3268]

https://twitter.com/DavidEggert00/status/1443958427965923331?s=20

[lfromnj]

https://twitter.com/DavidEggert00/status/1443958427965923331?s=20

Reminder that this map already is gerrymandered to split lansing and grand rapids in 2 to shoe the geographical bias.

[Gass3268]

https://twitter.com/DavidEggert00/status/1443958427965923331?s=20

Reminder that this map already is gerrymandered to split lansing and grand rapids in 2 to shoe the geographical bias.

Gonna have to split some more, partisan fairness is constitutionally more important than municipal/county lines or compactness.

[Antonio the Sixth]

https://twitter.com/DavidEggert00/status/1443958427965923331?s=20

Reminder that this map already is gerrymandered to split lansing and grand rapids in 2 to shoe the geographical bias.

Gonna have to split some more, partisan fairness is constitutionally more important than municipal/county lines or compactness.

Tbh I really don't like that as a matter of policy, but I guess it's gonna make up for the hideous gerrymanders the GOP is about to do across the South.

[Gass3268]

Handley strongly emphasizing that Michigan is not a 50-50 state but rather a 52-48 Democratic leaning state.

[lfromnj]

https://twitter.com/DavidEggert00/status/1443958427965923331?s=20

Reminder that this map already is gerrymandered to split lansing and grand rapids in 2 to shoe the geographical bias.

Gonna have to split some more, partisan fairness is constitutionally more important than municipal/county lines or compactness.

Tbh I really don't like that as a matter of policy, but I guess it's gonna make up for the hideous gerrymanders the GOP is about to do across the South.

Technically just worth a note this is for a state senate map although congressional will go through the same cycle. The best solution is just use Markov Chain simulations for a partisan fairness test.

[Virginiá]

Handley strongly emphasizing that Michigan is not a 50-50 state but rather a 52-48 Democratic leaning state.

That's a mostly fair statement to make, given election results over the past 10+ years...



Presidential

2012: 54.21% Obama  -  44.71% Romney
2016: 47.25% Trump  -  47.03% Clinton
2020: 50.62% Biden  -  47.84% Trump

[It’s so Joever]

Handley strongly emphasizing that Michigan is not a 50-50 state but rather a 52-48 Democratic leaning state.

BASED HANDLEY

[palandio]

The best thing to do regarding partisan data is to use a Markov Chain Monte Carlo simulation and just make sure the map lies within the middle 33% of all maps IMO. That way no communities have to be broken up while you can still follow the law. Colorado did at least listen to this idea late in the commission. Probably too late for that by now though. I guess you would also have to program the simulation to draw 2 45% black seats.

How would that MCMC work? You start with an initial map, possibly the one you want to evaluate and then step by step you propose random incremental changes that are accepted as long as certain measures (regarding e.g. compactness, integrity of administrative divisions, agglomerations, etc.) remain good enough (i.e. at least as good as the initial map). Then after some time you get a distribution of maps.

That all sounds good, but how do you choose the incremental changes? How do you make sure that the space of all admissible plans is contiguous?

In spring I came up with a metric and an optimization algorithm https://talkelections.org/FORUM/index.php?topic=432371.0 based on a simulated annealing-like method that converged by default on a (not necessarily optimal) map. The result could be different between runs. In the case of Michigan the result was very clear: If I lowered the energy slowly enough, the fundamental layout was always the same (The MI maps posted on the old thread are from an earlier version of the program where I didn't use census tract like I used later, but the result is very similar):

1 North (Safe R)
2 West (Safe R)
3 Kent Co. and ca. two smaller counties in its East/SE (Tossup)
4 Tri-Cities and Huron Bay (Safe R)
5 Flint to Port Huron (Likely to Safe R) [pairing of Flint with Saginaw or the Tri-Cities sadly never remained stable]
6 SW, including Kalamazoo, Battle Creek, Berrien Co. (Lean R)
7 Lansing and Jackson (Tossup)
8 De-packed Ann Arbor, including Monroe Co. and most of Livingston Co. (Safe D)
9 Outer Oakland Co. (Tossup, but trending D)
10 Macomb Co. (Nowadays likely R)
11 Outer Wayne Co. (Likely D)
12, 13 Inner Wayne and Inner Oakland (both Safe D), would have to be reordered to be VRA-compliant

[lfromnj]

The best thing to do regarding partisan data is to use a Markov Chain Monte Carlo simulation and just make sure the map lies within the middle 33% of all maps IMO. That way no communities have to be broken up while you can still follow the law. Colorado did at least listen to this idea late in the commission. Probably too late for that by now though. I guess you would also have to program the simulation to draw 2 45% black seats.

How would that MCMC work? You start with an initial map, possibly the one you want to evaluate and then step by step you propose random incremental changes that are accepted as long as certain measures (regarding e.g. compactness, integrity of administrative divisions, agglomerations, etc.) remain good enough (i.e. at least as good as the initial map). Then after some time you get a distribution of maps.

That all sounds good, but how do you choose the incremental changes? How do you make sure that the space of all admissible plans is contiguous?

In spring I came up with a metric and an optimization algorithm https://talkelections.org/FORUM/index.php?topic=432371.0 based on a simulated annealing-like method that converged by default on a (not necessarily optimal) map. The result could be different between runs. In the case of Michigan the result was very clear: If I lowered the energy slowly enough, the fundamental layout was always the same (The MI maps posted on the old thread are from an earlier version of the program where I didn't use census tract like I used later, but the result is very similar):

1 North (Safe R)
2 West (Safe R)
3 Kent Co. and ca. two smaller counties in its East/SE (Tossup)
4 Tri-Cities and Huron Bay (Safe R)
5 Flint to Port Huron (Likely to Safe R) [pairing of Flint with Saginaw or the Tri-Cities sadly never remained stable]
6 SW, including Kalamazoo, Battle Creek, Berrien Co. (Lean R)
7 Lansing and Jackson (Tossup)
8 De-packed Ann Arbor, including Monroe Co. and most of Livingston Co. (Safe D)
9 Outer Oakland Co. (Tossup, but trending D)
10 Macomb Co. (Nowadays likely R)
11 Outer Wayne Co. (Likely D)
12, 13 Inner Wayne and Inner Oakland (both Safe D), would have to be reordered to be VRA-compliant

Colorado asked for a fairly good study of MCMC recently
https://redistricting.colorado.gov/
Go to Meetings, Meeting materials congressional > sept 22> Ensemble. They had 2 versions. One which did not even consider county/township lines and another which did.

The commission didn't fully follow county lines but unlike a computer they still didn't chop every single county just like the 2nd set of data. In the end it shows that the Colorado map really isn't that R leaning.

[The Mikado]

Rather than this stuff, why don't we just all switch to the Iowa system? Put in some parameters and the computer will spit out a map. Legislature votes yes or no. If no, try again and keep trying until you get one the legislature likes. (IA only tries three times before the lege can draw their own, but just make it keep going until you agree with the computer).

I'm growing unsatisfied with human amateurs on redistricting commissions.

[Tintrlvr]

Rather than this stuff, why don't we just all switch to the Iowa system? Put in some parameters and the computer will spit out a map. Legislature votes yes or no. If no, try again and keep trying until you get one the legislature likes. (IA only tries three times before the lege can draw their own, but just make it keep going until you agree with the computer).

I'm growing unsatisfied with human amateurs on redistricting commissions.

The Iowa system works only because Iowa's demography is very simple from a redistricting perspective. It would be much harder to implement the Iowa system in almost any other state, and you might churn out maps that looked insane from a human perspective.

[President Punxsutawney Phil]

Rather than this stuff, why don't we just all switch to the Iowa system? Put in some parameters and the computer will spit out a map. Legislature votes yes or no. If no, try again and keep trying until you get one the legislature likes. (IA only tries three times before the lege can draw their own, but just make it keep going until you agree with the computer).

I'm growing unsatisfied with human amateurs on redistricting commissions.

The Iowa system works only because Iowa's demography is very simple from a redistricting perspective. It would be much harder to implement the Iowa system in almost any other state, and you might churn out maps that looked insane from a human perspective.

I could see an Iowa style system creating a CD taking in all of Detroit, which is a bad idea in Michigan for...obvious reasons.

[BoiseBoy]

MICRC has cranked out another horrible map.

[Born to Slay. Forced to Work.]

MICRC has cranked out another horrible map.

They’re atleast getting closer…

The “let’s making Lansing into a literal belt” idea isn’t great tho

[politicallefty]

It looks like most of these maps are collapsing MI-02 and MI-04 into one district and creating a new Republican seat in Macomb County (the latter being virtually unavoidable). Otherwise, it looks to me like they're just tinkering around with the current lines and attempting to compensate for the partisan imbalance by trying to make the Grand Rapids district a new Democratic-leaning seat. That said, it does appear quite difficult to create three Democratic-leaning seats outside of the Detroit area (assuming four there).

I made an attempt at a map myself, but I wasn't too pleased as to how it turned out overall (not for partisan reasons, just overall). I'd say it's more of a rough draft of intentions. While it does create a new Democratic-leaning Grand Rapids-Muskegon district and a Republican-leaning Macomb district, I didn't like that there are four districts each in both Wayne and Oakland Counties:



https://davesredistricting.org/join/05370be7-89d0-4e07-8229-c1bf513747a4

According to the analysis, 9 counties are split a total of 13 times. Other key metrics:

Proportionality: 74/100 (partisan bias of 52/100)
Competitiveness: 41/100
Minority Representation: 67/100
Compactness: 66/100
Splitting: 50/100

It's a slightly Republican-leaning map. The median district is MI-02, which Trump won by 0.1% (or 636 votes). Making MI-03 a Democratic district (Biden+7.9/Clinton+0.5) pushed MI-05 out of Bay City and into Oakland County, making the new district Biden+0.7/Trump+2.9. The Lansing district (i.e. Elissa Slotkin's seat) is MI-02, which I mentioned above. It moves very slightly to the left. Hillary lost it by 4.3% versus 7% in the current district.

In terms of previous elections:

2020 President: Trump 7-6
2020 Senate: James 7-6
2018 Governor: Whitmer 9-4
2018 Senate: Stabenow 8-5
2018 AG: Nessel 7-6
2016 President: Trump 8-5

[lfromnj]

https://twitter.com/umichvoter/status/1445831326360162305

Anyway although a natural house map would have a GOP advantage one thing I did notice is actually baconstripping Detroit may have hurt Dems slightly. It probably didn't affect anything in Wayne/Oakland but a lot of those districts actually took up moderately Dem leaning areas that would have been with swingy areas further up in Macomb. Instead those swingy areas are now with red areas in central Macomb.

Also Eid is starting have a bit of Tafoya syndrome. However the main thing is that there isn't any real counterweight to Eid that Tafoya had in Schell/Kelly/Leone.

[Virginiá]

Also Eid is starting have a bit of Tafoya syndrome. However the main thing is that there isn't any real counterweight to Eid that Tafoya had in Schell/Kelly/Leone.

Seems kind of subjective, but also as far as I know, Colorado's commission doesn't place a strong priority on partisan fairness, so it kind of limits the damage an ineffectual proponent can do.

[😥]

https://davesredistricting.org/join/abf6a9ee-0c85-479d-bac7-46cbb3b23e78

My attemp on redistricting in Michigan: 1 majority AA seat, one majority-minority seat (both of them can be majority AA if switch few precincts between them), 2 safe Democratic seats, 2 likely D seats, 3 safe Republicans seats and 6 highly-competetive seats

8-5 Trump in 2016
10-3 Biden in 2020

[Tintrlvr]

https://davesredistricting.org/join/abf6a9ee-0c85-479d-bac7-46cbb3b23e78

My attemp on redistricting in Michigan: 1 majority AA seat, one majority-minority seat (both of them can be majority AA if switch few precincts between them), 2 safe Democratic seats, 2 likely D seats, 3 safe Republicans seats and 6 highly-competetive seats

8-5 Trump in 2016
10-3 Biden in 2020

Definitely competitive! I think you could neaten it up a bit if you put Holland and Grand Haven in district 2 while putting (part of) Eaton in district 4 and then pulling district 4 south out of some of its remote northern arm without affect partisanship in the districts much if at all.