MI maps - muon2 scoring

[muon2]

One last test was to use the deviation-based inequality and apply the UCC cover and pack penalties (without single county UCC penalties). Six plans make the list.
train D: I+C 20, E 124
muon2 B: I+C 21, E 116
Torie A: I+C 23, E 106
Torie D: I+C 25, E 99
Torie C: I+C 26, E 95
Torie B: I+C 27, E 94
The regression gives a marginal value for I+C of 4.3 in erosity with a standard error of 0.4 and an R-squared of 0.969. It's consistent with the results for the simpler I+C calculation. Interestingly the theoretical marginal value of erosity in terms of chops controlled for inequality is 3.6 (square root of one less than the number of districts), so these results are consistent with theory as well.

For this data I used the following table. The older range basis for inequality is replaced by the average deviation method. Pareto optimality is found by comparing the sum of INEQUALITY and CHOP to the EROSITY.

Plan	ave dev INEQUALITY	CHOP (UP)	I+C	EROSITY
train A3	9	12	21	136
train B	11	11	22	126
train C	11	10	21	125
train D	11	9	20	124
Torie A	9	14	23	106
Torie B	12	15	27	94
Torie C	12	14	26	95
Torie D	11	14	25	99
Torie E	11	13	24	115
Torie F	10	12	22	122
jimrtex A	9	16	25	112
muon2 A	13	12	25	119
muon2 B	11	10	21	116

Is this a reasonable selection from the initial set of plans? In principle these would be the plans that would go to a vote by a legislature, commission or other body. train D would be eliminated since it violates a basic rule for connections. Inequality has a cap on the maximum deviation.

Should there be upper limits for chops or erosity or just let Pareto play out? For example, without UCC penalties or subunit chops no plan should need more than 13 chops (one less than the number of districts), and that could be a rule.

The minimum number of chops any plan in MI could theoretically have is 5: 1 Macomb, 1 Oakland, 2 Wayne, and 1 Detroit. In practice there has to be at least one additional chop or UCC pack penalty in GR and a chop or UCC penalty around Detroit. The nominal ideal average deviation for the minimal chop plan in MI is about 1200 or a score of 9. Together they add to 14, and based on the regression for erosity, a I+C score of 9 would be expected to have erosity of 146. Does that suggest a maximum erosity cap on a plan?

Finally there is the VRA issue. Many of the plans in the list allowed one or both VRA districts to drop below 50% BVAP. If a plan had a Pub skew, I would expect it to be challenged if it used the 47% standard instead of 50%. OTOH a Dem-skewed plan might survive, at least based on challenges in the most recent rounds of redistricting. Should that be a scoring factor, or just up to the commission on final review?

[muon2]

After all that technical analysis here's a little amuse-bouche. I took my Detroit UCC splintering muon2 A and gave it a reworking. I decided to see how far I could push down the combined inequality and chop scores and bite the bullet on erosity. I also wanted to keep both 50%+ BVAP districts. I used train's pack of GR to eliminate any penalty there but I left in place the excess cover and one excess pack in Detroit at a cost of 3 points there. Here's what came out.

MI muon2 2015C



SKEW 2 (R) (3D, 2d, 5e, 3r, 1R)
POLARIZATION 13
INEQUALITY 7 (ave dev 788, range 4217)
CHOP 12 (9 raw, 2 cover, 1 pack)
EROSITY 135

There are a number of interesting features here from a public policy perspective.
1. CDs 2 and 3 could have been maintained as whole counties with a pack penalty traded for the chop count. A whole county version would have reduced erosity, but average deviation was lower the way it's shown here.
2. The chop in Washtenaw is just enough to keep Milan in one CD. In the OH competition that would have been rewarded with no chop counted in that county.
3. The number of highly competitive districts (PVI=0 or 1) is up to 5 and the polarization is down to 13. In AZ increasing the number of competitive districts is a specific goal, though here it is only a consideration after the main scoring.
4. The shape of CD 14 is particularly erose, but linking the Grosse Pointes to Grosse Ile is a riverfront district and could be construed as a community of interest. There were districts in CA that seemed to use this type of logic.

[muon2]

Does Pareto optimization require a discrete domain, or can it be continuous.  I suppose you would have to figure out a way that could compare plans that improved one variable, while not significantly degrading another.   That was the impetus between my question about hyperbolic curve fitting, as a way of defining the direction of improvement.

Pareto works fine with continuous variables as well. The rule is that if one plan improves one or more variables compared to another plan, but makes no variable worse, then it is Pareto preferred. I assume that your two variables are the population shift and the perimeter for the regions. Do you have additional variables to score the districts in regions larger than one?



train - I'll look at the score for your new offering when I get a chance. I'm curious if you can put up a better version that keeps both VRA CDs over 50% than my latest. With my three UCC penalty points in Detroit that should provide some room to work.

[muon2]

Does Pareto optimization require a discrete domain, or can it be continuous.  I suppose you would have to figure out a way that could compare plans that improved one variable, while not significantly degrading another.   That was the impetus between my question about hyperbolic curve fitting, as a way of defining the direction of improvement.

Pareto works fine with continuous variables as well. The rule is that if one plan improves one or more variables compared to another plan, but makes no variable worse, then it is Pareto preferred. I assume that your two variables are the population shift and the perimeter for the regions. Do you have additional variables to score the districts in regions larger than one?

What if it improves one variable, while making another variable slightly worse (for mumbly values of slightly).

Then the plans are considered Pareto equivalent, even if it is a big improvement in one but a slight decline in the other.

You must be logged in to read this quote.

I understand that a regional plan might be chosen first, but there still has to be a way to score the subregional plans once the regional plan is selected. There are lots of ways to carve up the Detroit districts and where the extra population is attached will affect the districts, especially if the additional population is greater than the allowable deviation.

[muon2]

What this map sadly does not do is care much about inequality.  That number remains high and, with this arrangement, is staying high unless you add another chop, to give the northern five districts another 11K people or so.  So if inequality is a fully-equal leg on the Pareto stool, that would mean it's unlikely to knock many other plans out.  I've been thinking about trying to make an inequality-minimizing map (within reason), but would certainly go to a different template for that.  (Note that switching Sylvan Lake from 11 to 9 would help average inequality, but it hurts erosity and does nothing for the in extremis inequality measure currently in place.)

In my tabular analysis of the first 13 plans I found that using the sum of chops + UCC penalty + average inequality was a good variable to contrast with erosity. I'm willing to use it until a reason surfaces to not use it. In my plan that dropped INEQUALITY to 7 the average deviation was used.

BTW I really like your Detroit UCC plan. I have to think about how to avoid the ensuing macrochop of Jackson.

Thought: The inequality problem is that if one takes the Detroit and Lansing UCC and all the counties to their south the population is 9207 over 8 CDs. You can spread that around those CDs and if all have a positive deviation the average is only 1151. However, the remaining 6 CDs have a deficit of that same 9207, but the compact CD 6 has excess of 2424, so the remaining 5 CDs have to split a deficit of 11631 which is an average of -2326. That dooms any attempt at a low INEQUALITY. The only response is to chop into either one of the UCCs for 9K and take the point penalty. My guess is that the total I+C would justify that.

[muon2]

I took my suggestions for train's map (train F) and used them to drive down the average inequality. I kept the Detroit VRA districts with the swap of Wyandotte for Allen Park to better equalize population. I got rid of the macrochop in Jackson and moved the chop of Washtenaw to Monroe.

train, since it is based on your plan, feel free to tweak it towards lower average inequality or erosity. Obviously, we could swap the 47% BVAP threshold version from train E and reduce the chop count.

The chop of Clinton is annoying since it incurs both the chop and UCC cover penalty. As I noted in my previous comments something has to chop into either Detroit or Lansing to get low inequality based on the SE MI regions. Seeing the effect of the chop of Clinton, I'm inclined to think that anything short of a macrochop of a UCC shouldn't get a cover penalty. This would only affect multi-county single-district UCCs, and put them on an even basis with single county UCCs. That is to say the chop of Clinton/Lansing UCC should score the same as the chop of Monroe or Jackson in this map.

MI muon2 2015D

[muon2]

Anyway, for scoring,  "your" map is below (as best I could draw it), avoiding subunit chops, assuming that they are penalized in the chop score (as they should be). I must admit "your" MI-04 achieves absolute perfection.


MI jimrtex 2015A

It wasn't really "my" map.  I found a map on the Internet, and was using it for example.  You've jumped ahead to the next step.  But your going back and forth about whether it is better to split Eaton or Ingham illustrates a weakness of a single comprehensive stage.  It become exceedingly complex when trying to consider where the boundary should be between Grand Rapids and Lansing, when it is somehow tied to the division of Hamtramck.   If your statewide map had been approved, then there could be a simple focused discussion on where to get 13,647 persons, where all the options might be considered.

The switch of Osceola (not Missaukee) was automatic.  When a single county on a boundary can be switched and improve the equality between the two districts, then it is shifted.  The algorithm is simple.   Determine counties in the more populous district that have less population than the difference.   Choose the one that reduces the difference the most, while not breaking contiguity.

I had noticed that the shift of Missaukee would produce a 3rd district within 0.5% bounds.  I'll submit it as a joint effort.

You must be logged in to read this quote.

INEQUALITY 11 (range), 9 (ave dev) (range 5977, ave dev 1257) [11/9 in Torie D]
CHOP 12 raw (UC 14, UP 16, US 17) [12/13/14/16 Torie D]
EROSITY 112 (changes 1/2:5[6], 1/4:5[5], 2/3:10[2], 2/4:5[5], 3/4:4[2], 3/5:1[0], 3/8:5[3], 4/5:4[4], 4/8:0[1], 5/8:0[1], 5/10:4[3], 5/11:4[3], 8/11:1[1], 10/11:11[10] net +13) [99 in Torie D]

Shifting the chop from Saginaw to Ingham doesn't affect the raw CHOP, but does increase the UCC cover count, though if single county UCCs are counted it's a wash. Note that Kent is now a macrochop so erosity increases there, plus the other shofts tend to hurt erosity as well.

Shifting the chop from Clinton to Eaton increases the ave dev INEQUALITY to 10 and leaves the CHOP the same. The EROSITY drops to 110.

Edit: The chop into Ingham decreases the Detroit UCC pack from 5 to 4 so the UP score goes up an additional 1 beyond the 1 for GR.

When you originally split the Lansing UCC, you went all the way to the Ingham-Eaton line.  There is clear distinction between that, and chopping 5% of the county's population.

I don't see any difference in a policy standpoint from having two districts extending outside the Detroit UCC, and just one.  Keeping whole districts within a county might make sense for the Ohio and Texas houses, where you have many house districts, per county.  And in Ohio, it was the only part of the constitution that they followed faithfully since it was an absolute standard.

The main difference between our scoring systems is that I'm focusing on trying to achieve whole county districts, and stranding the minimum number of people outside their counties.  I think we should compare to West Virginia rather than Iowa.  First and foremost, we want to have whole county districts.   If we can achieve this in multiple ways, we prefer less erose maps.  But if we can't achieve whole county districts, shouldn't we try to make the smallest adjustments, rather than try to drive down to Iowa equality standards, just because we can once we breach county boundaries?

I understand what you are suggesting and that is equivalent to saying that there is a priority among criteria. I agree that there are states that approach it that way. I am suggesting a model that balances criteria with little priority between them. The result would be a small set of balanced alternatives that would go to a commission or legislature for final selection.

For example, Torie's plans use more chops than others, but push erosity quite low. train's plans choose to preserve counties and UCCs, but tend to be more erose. I'm ok with both types of plans going forward. In your model UCC preservation becomes so strongly favored that it is hard to see if there are reasonable trades to make in other parameters that violate UCCs.

[muon2]

Non-partisan redistricting is an interesting concept, one I am open to.  It's interesting an attempt is being made here to do that.
That's of course in the event that a law was passed that required House seats to be drawn with presidential voting in mind.  That's my favorite alternative to the current system - which I am not sure about on whether I want it reformed or not.

California and Florida do not permit use of political data, including locations of incumbents, except when used in conjunction with racial or ethnic minorities.  Arizona requires creation of competitive districts if possible.

Political information will always be a factor, even if it is not explicit on the software used to create the map. People who are likely to be involved in mapmaking have some idea where the D's and R's are and it seems a fools errand to think that knowledge vanishes. Even if a neutral map comes out, my observation is that the first thing the media does is figure out the political dimensions of the map. It is certainly what happened with maps here on Atlas.

I take a different approach, one that assumes that the data is there and will be applied even if the law forbids its explicit use. In this metric the political factors are explicitly scored, but not used to judge the plan on Pareto equivalence. The political scores go forward with the successful plans and are available to the body that makes the final decision. That places all the cards on the table for the final decision, even if that decision doesn't use the data.

[muon2]

You must not understand my model to suggest that I favor UCC's over other "reasonable" trades.

While you claim that your approach is balanced, you disregard the magnitude of chops.  Why, in your opinion, should county splits be disfavored?

In state after state that I see, splits of counties and munis are the things that appear in reports and in the media. It isn't the size of the split that first matters, just that there is a split.

You must be logged in to read this quote.

In terms of counting chops, I don't see how this differs from what I do, other than it starts at the minimum number of districts for a large county. It seems the same for the UCC until one gets to the point of counting the population in the chop as opposed to the chop itself. It also doesn't address the concern about fanning out from a UCC that is addressed by looking at the pack count as well as the cover count.

[muon2]

If there are only 3 districts in Wayne County, it is a matter of necessity.  I don't see the point of assessing what would reasonably perceived as penalty points for doing what is necessary.

There has to be a minimum number of severed links in the erosity score, too. It is easier to count, than to count and subtract the minimum. The raw score for both chop and erosity are just counting exercises.