Ungerrymandering(and unskewing) squad!

[jimrtex]

Let me take these one at a time since how you react to each one tells me how you would react to subsequent metrics. I'll start with the VRA and provide 3 options. The two principals in this exercise should come to agreement on an option, perhaps with simple amendments or clarifications.

Option 1: Count the current number of minority-performing districts and keep the same number. If possible the districts should have at least 50% VAP (60% for HVAP) to be acceptable. This is sort of the pre-2010 understanding of how to satisfy the VRA and the simplest to draw and check.

Option 2: Count the current number of minority-performing districts and provide for no fewer districts than the current number. Districts should have at least 40% VAP (50% HVAP) and a sufficient Dem fraction (usually PVI D+6) so that the minority can control the primary and then use crossover white Dems to win the general election. Think of this in terms of the new VA districts this decade and probably more closely reflects current court thinking.

Option 3: In addition to the metric in option 2, maintain contiguous clusters of rural counties that are at least 40% VAP (50% HVAP) as a community of interest where there is less minority population than needed to form multiple districts (otherwise one could have illegal packing). For example one should maintain the Black Belt counties in AL or northeast NC in one CD without cracking them when possible given other constraints. I have lists for the states where this is a factor. This can be applied to municipalities in an urban area, too.

Remember that districts in North Carolina, Virginia, and Alabama have been thrown out because they were drawn to targets.

[jimrtex]

the problem is that map undermines the goals of our map. We intentionally cut out Waterbury from the Litchfield Based district. As for the issue of large deviation, how about this?

An alternative interpretation is that you placed Waterbury in a a New Haven-dominated district.

While you can draw an eastern district that avoids cities other than smallish New London, I don't think you can do this in the northwest. My preference would be to minimize the intrusion on Hartford, Bridgeport, New Haven.

[jimrtex]

Alright, this is my final product, thanks for everyone's feedback. This is pretty much the best it gets, not to sound overly braggy



0 towns split, minimal population deviation,

Much nicer, but I wouldn't call it minimal population deviation as there are many plans that put all deviations under 1000. What I would say is that it stays under a 0.5% maximum deviation. So I would suggest that your inequality metric not be about lowest deviation, just a low enough deviation.

So to justify a larger deviation than necessary we need to identify the metrics that are more important than population inequality. Metro areas don't seem to be super important here since there are metro area chops that didn't need to happen, such as for New Haven and Enfield (Springfield MA). That suggests that shape is highly important, would you agree?

Ideally I could have better population deviation, but I feel like my map achieved my goals the best they possibly could. The general groupings I wanted were...

1-eastern counties
2.Litchfield+Surrounding areas
3. Most of Fairfield
4. Most of New Haven
5. Most of Hartford

 I know you said counties don't matter, but I really do like these types of groupings, and I definitely would sacrifice near-perfect equality for an otherwise bad map.

Here are the stats on your CDs (it's hard to read them in the screen shot):
New London (+2435) D+3.6
Litchfield (-1945) R+2.4
Fairfield (+1500) D+8.5
New Haven (-778) D+8.9
Hartford (-1210) D+14
The population range is 4380 or 0.61% of the quota. The average deviation is 1573.6 or 0.22%.

Your goals are fine, but it is hard to measure how well you met them. WV succeeded in their case with a large deviation because the court could compare the challengers' maps to the state's map based on specific numeric goals. Here's something like what I think you have so far:

1. Towns are not chopped.
2. Counties should only be chopped if they are larger than one district.
3. Counties should have a whole district within them whenever possible.

These don't exactly match your map, since you don't have a district entirely within New Haven county. Is there a reason why? If you did stick to a district entirely within New Haven and kept deviations down it would probably be defensible as the best plan.

Here's a modification of a map I posted earlier that would meet those goals. Not only does it meet the goals above, but it has a lower range and average deviation than your plan. If you don't want the court to replace your plan with this one with then there needs to be a measurable reason to keep yours.

New London (+2435) D+3.6
Litchfield (-1588) R+1.5
Fairfield (-227) D+8.1
New Haven (+708) D+8.1
Hartford (-1326) D+15
The population range is 4023 or 0.56% of the quota. The average deviation is 1256.8 or 0.18%.

This was my map. The southwestern districts are the same. Standard deviation is 0.17%, average deviation is 0.15%. Your competitive districts are marginally more competitive 0.3% and 0.5%).

Your two competitive districts are marginally more competitive.

[jimrtex]

For population would you prefer to measure the range of the average deviation? The range is more straightforward and basically says that the districts with the greatest deviations above and below the quota set the scale for the map and other districts don't play a role for the inequality.

Use of the range is pandering to innumeracy, and has led to map drawers targeting the limits when applied to legislative districts.

We can think of drawing districts as analogous to shooting arrows at a target. We're supposed to be aiming toward the center of the target. Setting a range is like having an inner ring count as 10 points.

But gerryarchers have discovered that you can drive down to the target in a golf cart and put all the blue arrows on one side touching the ring, and all the red arrows on the other. Or even worse they move the target to the left or right so that the ring is still X inches across but is displaced to one side.

Using average deviation, you can have two districts both with 0.3% deviation, and then swap population between the districts and have zero effect on the average deviation. 0.0% and 0.6% is just as good as 0.3% and 0.3%. But if you move one additional person, suddenly the average deviation starts increasing. You have a discontinuity in the first derivative.

Standard deviation penalizes more extreme deviations, like range, but doesn't encourage targeting the limits and doesn't limits the measurement to two values.

[jimrtex]

For population would you prefer to measure the range of the average deviation? The range is more straightforward and basically says that the districts with the greatest deviations above and below the quota set the scale for the map and other districts don't play a role for the inequality.

Use of the range is pandering to innumeracy, and has led to map drawers targeting the limits when applied to legislative districts.

We can think of drawing districts as analogous to shooting arrows at a target. We're supposed to be aiming toward the center of the target. Setting a range is like having an inner ring count as 10 points.

But gerryarchers have discovered that you can drive down to the target in a golf cart and put all the blue arrows on one side touching the ring, and all the red arrows on the other. Or even worse they move the target to the left or right so that the ring is still X inches across but is displaced to one side.

Using average deviation, you can have two districts both with 0.3% deviation, and then swap population between the districts and have zero effect on the average deviation. 0.0% and 0.6% is just as good as 0.3% and 0.3%. But if you move one additional person, suddenly the average deviation starts increasing. You have a discontinuity in the first derivative.

Standard deviation penalizes more extreme deviations, like range, but doesn't encourage targeting the limits and doesn't limits the measurement to two values.

Numeracy starts with the basics. Standard deviation is a fine measure, but if the mapper isn't comfortable calculating square roots, it's likely to get ignored as the plan is drawn. Average deviation requires nothing beyond division and range only requires subtraction. I'd rather see someone apply subtraction than use no math at all, and if using range gets people to think about more math as they draw maps, that's a plus for numeracy.

You may not have experienced it because Illinois requires near perfect equality between legislative districts, but it is explained to legislators that they can be within 10% range if the smallest district is 90.5% and all the others are less than 100.5%. They then use the limits as targets.  The main reason that the SCOTUS has refused to set a de minimis standard for congressional districts is because they know that lawyers and legislators will cheat.

In practice, one simply makes sure that the districts are within a 95% to 105% band and don't even bother with a range calculation unless one needs to shift the band slightly to conform to some other rule.

Range is a measure of the two most extreme districts, and provides zero information about the practicable equality of the other districts.

Average absolute deviation is better, but it has the problem that when adjusting population between two districts, the average may not change. Standard deviation has the property that if any two districts are made closer in population the standard deviation decreases.

I don't understand your concern about being able to hand calculate the values. I certainly wouldn't. If hand computing range, I would likely miss the maximum or minimum value and thus compute the wrong range. I would not compute average deviation by hand. I doubt that I have ever calculated a square root by hand. I'd use a slide rule or log tables, or if available a calculator. Any polynomial expansions would be a beginning problem in programming perhaps with recursion.

Today, I would use a spreadsheet, or the calculation would be built into the mapping program, just as measurements of compactness are.

[jimrtex]

I thought the standard for legislative districts was 90% to 110%.

The standard was based on a Texas case where the districts ranged from 105.1% to 95.1%, The SCOTUS later observed that they seemed to have accepted a 10% range.

What they have said is that a State can go outside the 10% range, but has to justify it.

Below 10% the plaintiffs have to show that they can reduce the deviation AND achieve the State's goals.

[jimrtex]

Gerrymandering is the process of drawing districts to get a particular political outcome. Drawing a plan to insure one seat with a Dem PVI is therefore a form of gerrymandering, even if it is for a good public purpose. That's why I pointed out that the plan was more designed to unskew AR rather than ungerrymander it.

Hey, notice how I chose my words carefully and I did not say gerrymander. Your map is a shameless GOP rig just like the current one. I said a GOP rig, and imo an Arkansas map without a tossup to highly competitive seat is a GOP rig. I point out issues on both sides of the aisle and think the maps in MA, CT, MD, and OR are reprehensible and actually drew a Massachusetts with a McCain seat. Shame that people on all sides pull out excuses to allow this. I have no idea why a 3-1 GOP majority delegation would be so controversial unless one just loves being unfair and unrepresentative.

Who drew the current Arkansas map?

[jimrtex]

I wasn't going by the muon rules when I made this,

5-8 is a pretty good split for a Lean R state. 

The ideal map in terms of the expected partisan split is 7R-5D-1 even. The muon rules can generate that type of map as I posted last month.

Good news. I found the DRA file and can load it to read the current stats.



CD 1: D+3.1 (BVAP 40.3%)
CD 2: R+2.6
CD 3: R+8.1
CD 4: D+14
CD 5: R+8.8
CD 6: D+2.2
CD 7: R+8.7
CD 8: D+1.0
CD 9: R+19
CD 10: R+19
CD 11: R+8.9
CD 12: D+5.0
CD 13: D+4.8

The districts are 1 uncompetitive D, 4 competitive D, 1 highly competitive, 1 competitive R, 6 uncompetitive R.  The current NC PVI is R+3, so with these PVI's the plan has a skew of 0, as fair as one can get by that measure.

Note that there are only 4 county chops, and only 2 beyond the 2 that are required (Wake and Mecklenburg). The maximum deviation is 0.337% of the quota.

What is racial composition of the two parts of Mecklenburg? (i.e. have you racially cracked the black population). In Wake, most of the black population is in your NC-13. Though, you may have split Raliegh (and Cary?) NC-5 seems kind of odd, wrapping around from west of Greensboro to almost Durham and Raleigh.