What districts would have Dems won back if not for gerrymandering?

[muon2]

I am just assuming the GOP controlled legislature states didn't change anything from 2000 maps. Of course NC-2 was very gerrymandered, a fair NC map would probably have a Wake County based district that takes up most if not all of the county. And I have to keep GOP in PA-12 b/c that seat was being eliminated regardless, although the Democrats would have chopped up Pittsburgh.

Shouldn't you make the same assumption in IL, too? A neutral map would probably have been 10-8 Dem for a pickup of 2 instead of 4 seats.

[muon2]

No, I won't count obviously Democratic states like Illinois and Maryland and GOP ones like Indiana, Missouri, South Carolina where each side would have a better map from a commission. This one is just about partisan gerrymanders in competitive states.

So either you should either state that your hypothesis requires that 2010 had been a year when Dems would have captured a proportional number of legislatures, or you should change the title to add GOP in front of gerrymandering.

The IL gerrymander was every bit as egregious as OH or NC according to most pundits who've analyzed the maps. The presidential vote tendencies wasn't what gave control of the map to the Dems. It was the combination of the 2000 legislative map paying dividends with an unbreakable majority and the failed campaign for Gov by the GOP.

[muon2]

12-6 Dems in Illinois really isn't absurd by any means, much less absurd than 13-5 GOP in Pennsylvania. Even in a neutral Illinois map you would expect Dems to have 11-12 of the 18 seats.

12 Dem seats in IL is a huge stretch with a neutral map, and this map was drawn to be 13-5 but the Dems fielded a weak candidate in IL-13. This is different from quoting the number of seats that Obama would carried on a neutral map, but remember he has a favorite son advantage here. 10-8 is the most likely neutral map, with 5 downstate GOP out of 6 (it votes like IN) and 3 suburban GOP seats out of 12 in Chicagoland.

Counting the straight partisan numbers doesn't really tell the tale of the Dem gerrymander. Many of the 18 seats should be swing seats, and one mark of the level of gerrymandering is the number of swing seats that were moved to D+8 or so to put them out of reach in all but wave years. This is particularly true in the suburbs where the GOP was packed into two safe seats so the other 10 seats could be made safely Dem. Downstate the GOP was packed into three seats leaving three seats for the Dems. To do that they had to expose themselves in two swing seats, one of which the GOP retained in 2012.

[muon2]

http://www.politicspa.com/wp-content/uploads/2011/12/PA-Dems-congressional-map.png
I have just a question: how many seats could the democrats have in Pennsylvania with this map?

I can't understand why the minority Dems in PA would put out such a crazy map. At least the OH Dems filed a neutral proposal as an alternative to the GOP gerrymander.

[muon2]

Returning to Illinois, I wonder how much the favorite son effect was.  Is 5% about right (Illinois trended 5 points to the Dems in 2008, and about 5 points back to the Pubs again in 2012)?  

I ask, because per my "good government" non partisan map below using the rules that I like to use (minimizing chops, with a laser beam like focus on minimizing erosity, and ignoring just about everything else other than the VRA (and in this instance making some effort to create a second Hispanic influence CD, even though there is no second Hispanic CD to draw required by the VRA that hits 50% Hispanic VAP), I count nine potentially competitive CD's (yes nine), if one uses a 5 point PVI adjustment towards the Pubs.  Pity the map was not enacted into law, because if it had, Illinois could have had its economy revived just by all the campaign cash pouring into it given the jury sized number of CD's in play.  

I must say, that the Pub strength in the collar counties has just collapsed since my college days - just a massive implosion. Wow, just wow.

Nice map. The GOP lost their challenge in large part because they couldn't present a plan with two majority Latino districts. I posted this in Mar 2011 right after DRA updated the data. It preserved three black CDs over 50% VAP. CD 3 is 59.4% HVAP and CD 4 is 50.1% HVAP.



The 7th Circuit in 2011 used 59.2% HVAP as the benchmark to elect a candidate of choice since that was the 1991 map percent that successfully elected and maintained Gutierrez. Using block-level mapping the following plan in Chicagoland has both Latino districts over that mark, CD 3 at 59.9% and CD 4 at 59.4% HVAP. The black CDs have 52.5%, 51.6% and 51.0% BVAP for 1, 2, and 7 respectively.



That can be nested in a statewide plan that has exact equality and minimizes county splits while preserving the cores of the previous districts. The PVIs (2004-08) work out to 10D - 8R, but 10 of the 18 CDs have PVIs at 5 or less.



Here's the full list of PVIs and who was resident in 2012.

CD-01   D+28   Rush
CD-02   D+26   Jackson
CD-03   D+21   
CD-04   D+23   Gutierrez
CD-05   D+17   Quigley
CD-06   R+1   Hultgren, Roskam
CD-07   D+37   Davis
CD-08   R+1   Walsh
CD-09   D+19   Dold, Schakowsky
CD-10   D+0   
CD-11   R+9   Shimkus
CD-12   D+2   Costello
CD-13   D+1   Biggert, Lipinsky
CD-14   R+3   
CD-15   R+5   Johnson, Kinzinger
CD-16   R+2   Manzullo
CD-17   R+0   Schilling
CD-18   R+4   Schock

[muon2]

In other news, my IL-02 is 59.3% Hispanic VAP. I sweated that number, because I did not want to violate any of my other rules, unless the VRA risk was too high. I decided not, and would believe that even if the CD were below 59.2% Hispanic VAP, because I don't think the VRA requires grotesque CD's to hit a number, violating every other good districting rule in the book. There simply is no community of interest there, above and beyond that.

Anyway, it was a balancing test, and I sort of used 59% HVAP as the threshold, and if the number fell below that, the edges of the CD would have been more ragged. As it was, if all things were relatively, even if not absolutely equal, I sought out the higher percentage Hispanic precincts at the margin.

Hearing no objection, I am going to assume that the 5 point favorite son effect is in the ballpark,, which is what makes the map just so much fun.

As I said, I think yours is a good map. The maps I posted were initially drawn prior to the opinion of the 2011 case over the actual map. They were drawn presuming certain reactions from the judicial panel, which generally held to be true. The court was clear that no map would be overturned on VRA grounds without showing an alternative with a second Latino CD that exceeded 59.2% HVAP. As long as the alternative district was all in the same media market, the shape was defensible since there is little else that binds the existing lobes of the historic IL-04. I agree that either of the extra CDs are ugly, but beauty is not a factor for VRA districts when no compact alternative exists.

One reason I quoted 04-08 PVI's was to blunt the home field advantage of Obama in IL. It averages the 2004 numbers which lack the effect.

I must say, that the Pub strength in the collar counties has just collapsed since my college days - just a massive implosion. Wow, just wow.

Pub strength in the Chicago suburbs has been declining since the 1994 election. 2010 provided no relief in legislative races in that region though it did downstate. Like around the other old industrial cities of the north, the nationalization of the party messages though cable TV deters local campaigns from getting traction with a message of their own tailored to suburban issues. The media market is just too expensive, and Dems just link suburban Pubs to national stories to win (think Akin). Other state's GOP have fared better with a strong state organization, but that has been lacking in IL since 2000.

[muon2]

Muon2, you are not contending that the VRA requires that second Hispanic octopus that sends out one tentacle from inner city Chicago to somewhere in the vicinity of beautiful downtown West Chicago are you? I guess maybe I missed that case. Next thing you know, someone will have the quite insane idea of hooking up black neighborhoods in Cleveland to those in Akron.

I did not contend that a second one is required, for if it was the court would have compelled it. What I contend is that the court required that a map with a district of sufficient HVAP needed to be offered by the plaintiffs in order to show that the approved map suffered under the VRA. The court found that the plaintiffs failed to meet the first prong of Gingles. That doesn't mean the plaintiffs would have prevailed with the map I posted, but I believe that something like it would have been necessary to prevail on a VRA claim.

[muon2]

Hm.  What do people think of this proposal as a way to lessen, if not entirely neutralize, the partisan effects of line-drawing (both from gerrymandering, and "natural packing")?

Districts should be drawn in each state so that half of them have a PVI more D than the state as a whole, and half of them have a PVI that is more R than the state as a whole.  Maybe allow wiggle room of a point or two in either direction.  So, in the case of Michigan, you'd need seven districts D+4 or more Dem, and seven districts that were D+4 or more Pub.  Conversely, a state like North Carolina would be mandated to have six districts that are at least R+3, and a seventh right around that number.

This should safeguard against the worst abuses, in both directions. 

I know you want to move towards House representation that more closely reflects the popular vote, but the Pubs are never go to agree to that. In fact, I suspect I, and Muon2 too for that matter, are too partisan to agree to that. What I am willing to do however, if need be, is skew a bit towards an excessive number of swing districts. What the Dems would really want, at a minimum, and I understand that, is to offset the screwing that they get from the VRA. That isn't going to happen either. So we need to do the best we can with what reasonable folks on both sides are willing to do.

My sense is that the best method is to design a plan based on demographic and geographic criteria. I use a bilevel approach that starts with apportionment regions then moves to districts. The criteria are limited to population inequality, chops, and erosity, all of which should be minimized.

After a map is drawn one should test the plan for partisan biases. There are two independent metrics to use. One is polarization which judges how many districts are uncompetitive. Statistically districts with a PVI of 0 or 1 have an equal chance of going to either party so they exhibit no polarization. CDs with a PVI of 2 through 5 are favored to go for the indicated party about 3 to 1 and could be scored with one unit of polarization. PVIs beyond 5 are over 95% likely to go for the indicated party and can be counted with double polarization.

The other political metric is the skew. Skew measures the deviation of a plan from the expected result based on the statewide vote. The district plan is expected to have a distribution that is about twice the PVI of the state. Take the state PVI percent, double it, add 50% and multiply by the number of CDs for the expected partisan lean. The raw skew takes the difference between CDs for the party with the preferred statewide PVI minus those for the other party and divides the difference by two. The difference between the raw skew and the partisan lean is the skew for the plan.

A strong system to judge plans should show off its biceps.

[muon2]

You must be logged in to read this quote.

New Jersey is rather unique in the US with both strong counties and town subdivisions. In the Midwest, many voters do not know their township but certainly know their county. For most the township exists as a basis for the county to track assessments and to group voting precincts. Only in farm areas where the township is the primary provider of road maintenance does it take a more prominent role.

Pennsylvania is the same; New York is a little more complicated with its "villages" but I believe that NY towns all have some level of government that extends well beyond rural road maintenance.  (I don't know to what extent school districts line up with town boundaries, though: in NJ every muni has its own school district at least on paper, in PA school district lines follow muni boundaries but often include multiple towns).

I worry that, if your proposed guidelines for dealing with metro areas would allow the last map you put up, then they are far too toothless for my taste.  Splitting Livingston from the Detroit area, and the tri-chop of Lansing's core, together have the effect of ensuring that there is no real Lansing district, when other maps have shown how it is easy to make a single whole-county Lansing district that substantially keeps the whole of the core together in one district, rather than sundering it three ways.   I understand your concerns about not wanting what you consider to be "subjective" CoI criteria... but a "foolish consistency is the hobgoblin of little minds", as they say.  

Perhaps a balancing test could be added that keeping small MSA cores in one district, or large MSA cores in one apportionment group, or different parts of a CSA in the same district/group, is a priority goal- and that failure to do so should count as a county cut for the purposes of scoring.  (Maybe breaking off parts of a CSA would be worth half a county cut or something.)

The purpose of my scoring is not to find a single winning entry. Mathematically that is not a tractable problem given the large number of census blocks (or even voting districts). The purpose is to find an objective set of plans that can then be made subject to the review by a group of humans who would select from the set. Subjectivity is fine for that review, but if subjectivity gets a hold on the creation of a set all sorts of games become possible. The core rules are thus immune to exceptions, and that is left to the reviewers to apply.

I would prefer not to put up any metro rules because as one goes through the states there are inevitable exceptions. If you don't like my Lansing chop offering that's fine. My point is that if one looks at each district on its own, there may be no reason to exclude it from the set that goes to the reviewers. Neither the shape of that CD 4 not the way it selects some of the whole municipalities to include from Ingham is particularly unusual. It's only when viewed in the context of the the neighboring districts and the global knowledge that other maps don't split the metro that it becomes objectionable. I claim that that is exactly why a human element is part of the process.

[muon2]

Ladies and gentlemen, the perfect Pennsylvania.  (Using a max 0.5% deviation, that is.)



Only six counties have any chops at all, and five districts are whole-county.  Only Philly is split, and it is only split along ward boundaries.  District 2 is 59% black, District 1 is minority-plurality by total population (though white majority by VAP, as a result of how the ward math worked out; would've rather had it min-maj total, which is almost certainly possible if you get ugly and abandon that nice Broad St. boundary).



In addition, metro areas are kept as damn near pristine as possible.  In SEPA, the five-county Delaware Valley plus Reading (part of the Philly CSA) plus Lancaster (its own metro, not part of any CSA) is one apportionment district.  In addition, the three inner suburban counties, Reading, and Lancaster, are all min split- there is an all-Montgomery district, and Delaware/Bucks/Lancaster/Reading are all kept whole.  The extra chops (all of which I'm pretty sure are necessary) are confined to Chester and Philly.

The 12 (old 19)-17-15 apportionment district in Central PA and the Lehigh Valley does the worst in terms of metro contiguity.  The three counties of the Allentown metro are just a wee bit too large, so a chop of about 5K from Carbon is given to 17.  Not technically a microchop, but it's in that spirit.  The Harrisburg area fares worst, as there are three districts in the MSA, and Cumberland is chopped.  But at least the core-Dauphin- is whole, and Perry County is outlying and rural, and the York-Adams-Carlisle district can't avoid a chop.

In SWPA, the Pittsburgh-New Castle CSA is almost perfectly in 2 districts, except for the addition of Greene in the far southwest corner (it's surrounded) and the subtraction of Westmoreland.

The only other instance of MSAs being broken up is that Wyoming County is separated from the rest of the Scranton-Wilkes-Barre MSA. But, again, it's an outlying county, and Monroe-Lackawanna-Luzerne is just too perfect.  You'd have to split something there, and if it's not what I did, it would be a different MSA plus a county.  Worth it, easy.  As for CSAs, the Lock Haven micro is split from the Williamsport metro; an earlier draft had them together in 10 but splitting them allowed 9 and 12  to no longer be underpopulated, and to be whole-county.  So it saved two cuts, again.

I think this is a map that would be hard to improve upon.  You could arguably clean up things between 10 and 17 at the expense of a county cut if you wanted to, that's about it.  And of course the lines within Allegheny County will always remain gameable for partisan advantage.

As they say - great minds think alike.

Here's a version incorporating Verily's suggestions for western PA. I've also made some adjustments to SE PA.  I still like to rely on whole counties, since that is one of the defensible criteria to allow deviations in excess of one person. CoI is a nebulous criteria and exact population equality would generally be needed.

Here were my criteria and their impact on the map:

Districts are drawn to use whole counties to the extent possible and counties larger than one district have as many whole districts within as possible. The map divides three counties other than the ones that have whole districts within. Within counties no city or township is divided. Within Philly no ward is divided.

Instead of limiting the deviation, I limited the range from the smallest to largest district to be less than 1%. This is from SCOTUS decisions, and note that a 0.5% deviation limit results in a 1% range limit. The population range here is less than 1% (-0.7% to +0.3%) and the mean deviation is 884 persons.

CD 2 is designed to comply with the VRA and is 61.8% BVAP. CD 1 keeps the Hispanic wards together and is 18.8% HVAP and 18.2% BVAP. CD 1 also includes Chinatown and the Asian areas of S Philly with 7.6% AVAP.

Other than the changes in central PA, which have the effect of relocating my chop in Cambria to one in Cumberland, and the shape of the Pittsburgh CD, I can't see much to complain about.

[muon2]

The two ends of the state look solid, central PA does not to me. You guys are working too  hard to find micro-chops at the cost of erosity in my opinion. That pink CD is particularly unfortunate to my eyes.

I agree, but NE PA is particularly challenging. The Allentown-Bethlehem and Scranton-Wilkes Barre areas each very nicely make CDs that are compact with low erosity. Add a CD for Harrisburg and one is left with a lot of relatively lower population counties that make up the rest of NE and central PA. The effect is one on making perfect holes in a Swiss cheese but then there's the odd shape of the cheese itself. Since the erosity for the plan is based on the sum of all the individual district erosities, it is easy to get a good score by making many low erosity districts then throwing away a high erosity district with the remainder.

On the other hand, I drew up this plan before we had most of our erosity discussions, so it is dominated by chop considerations. But chops and erosity must balance each other. If erosity dominates then I can start drawing some quite partisan versions because I lose the power of the chop/microchop constraint. I'll revisit my numbers to see if there are modest balance points to assuage your eyes.

[muon2]

The two ends of the state look solid, central PA does not to me. You guys are working too  hard to find micro-chops at the cost of erosity in my opinion. That pink CD is particularly unfortunate to my eyes.

I agree, but NE PA is particularly challenging. The Allentown-Bethlehem and Scranton-Wilkes Barre areas each very nicely make CDs that are compact with low erosity. Add a CD for Harrisburg and one is left with a lot of relatively lower population counties that make up the rest of NE and central PA. The effect is one on making perfect holes in a Swiss cheese but then there's the odd shape of the cheese itself. Since the erosity for the plan is based on the sum of all the individual district erosities, it is easy to get a good score by making many low erosity districts then throwing away a high erosity district with the remainder.

On the other hand, I drew up this plan before we had most of our erosity discussions, so it is dominated by chop considerations. But chops and erosity must balance each other. If erosity dominates then I can start drawing some quite partisan versions because I lose the power of the chop/microchop constraint. I'll revisit my numbers to see if there are modest balance points to assuage your eyes.

As I said above, my PA map was before our great CA discussion that led to microchops and erosity. I revisited the numbers and found that I could maintain a plan with just two full chops (Westmoreland and Monroe) outside of SE PA and the minimum required one in Allegheny. If I use a full chop in Lawrence I can eliminate the three microchops for CD 3 so that gives a sense of the trade off. By splitting Scranton from Wilkes-Barre the erosity can be greatly reduced for CD 10.

[muon2]

Better Muon2, but I would switch counties between the yellow and pink CD's to square both out more, at the cost of another macro-chop. There is no good reason for that much additional erosity, particular for CD's large in geographic size, in order to avoid a macro-chop. I understand your concern about leashing gaming, but that is where the veto mechanisms come into play to mitigate that in my opinion. Here of course, what is in play really does not have any partisan effect, but I understand that it might in another instance.

Unless you want to game a particular formula to suit your eye, flipping parts around between 5 and 10 in my revised offering doesn't change the erosity much whether you measure by border segments or by conventional compactness. I claim that most neutral observers would have no issue with the shapes of those districts. You really need to point at a formula that suits your taste, and I can see what unintended consequences it would have.

By splitting Scranton from Wilkes-Barre the erosity can be greatly reduced for CD 10.

Yeah, this is where I say that "cutting down on erosity above all" can lead to rotten outcomes.  Scranton and Wilkes-Barre need to be in the same district, and if that increases erosity than so be it.  There are worse things than a little erosity.

I'm not necessarily opposed to adding a chop to the western side of 10 to clean it up... but districts 11 and 15 in my map and muon's old map are obviously optimal, and anything that goes away from that is a step backwards.

One observation I would make is that both Lackawanna and Luzerne are large enough to significantly influence the political outcomes. There are folks I know from that region who would rather have them separated so as to control 2 CDs rather than 1.

I think that by proposing the county lines not significantly divide an urban area then that is equivalent to saying that one use the core of a metro area as a boundary. To deal with geographic obstacles like mountains it made sense to disallow connections between counties that were otherwise contiguous. This proposal essentially requires that connections between counties for the same urban area not be broken. The best neutral version to apply this is to use the central counties of the MSA as defined by the Census Bureau. From a map construction view this is equivalent to treating the counties that can't be split as a single "super-county". When all such mandatory bindings are included it can be shown that the super-county is equivalent to the core of the MSA as defined above. Thus one would in effect be using that area as a boundary.

Central counties are associated with the urban area that has the greatest population in the county.  Central counties associated with a single urban area are grouped as a cluster of counties for purposes of measuring commuting from potentially qualifying outlying counties.

Outlying counties are defined based on commuting patterns between a county and the central counties as a group.  In addition, CBSA may be merged if the central counties (as a group) of one CBSA qualify as outlying counties of an other.

You confused me by your use of "core".

My use of core was not the Census definition but merely to describe the set of central counties in an MSA. I still contend that is the best uniform, neutral measure of the metro area when building apportionment regions. The advantage it has over your definition is only one of simplicity in that I can find that info on the Census web site more readily. Also, it doesn't depend on a state and I find that to be critical for the rule set. State-based and local info do have a role in the process, but that goes to the subjective elements that should be the purview of the commission.

Let me put forward two examples that deserve consideration when viewed by this or any other metro area rule. A prominent one has just surfaced with my two PA maps. My original map maintains the central counties of Scranton-Wilkes Barre MSA in the same region, which happens to be a single district. This comes at the cost of considerable erosity for CD 10 which must wrap around the north and east of the MSA-based district. The erosity problem can be fixed, but it splits the MSA. Should Torie or train have their preferred plan eliminated?

The second issue I mentioned in an earlier post. In my offending MI map my Lansing region maintains all the central counties. The problem occurs when the apportionment region is divided. What you are observing is that the Lansing metro could go into 1 CD and that is what causes the consternation when it is split by two districts. That's a different issue than how to build apportionment regions.

You must be logged in to read this quote.

goon redistricting?

In my view, apportionment regions are more appropriate for legislative redistricting, in which counties with multiple districts are more typical, and (in)equality standards are more relaxed and rational.

If you could convince the SCOTUS to permit more variation in congressional districts, than they might be more useful for creating districts, since it would permit better conformance with COI.

[/quote]

I'm not claiming that it is good redistricting, rather that it is neutral and provides sufficient constraints to bar egregious political gerrymanders. The SCOTUS standard is important and colors the level of subdivision that must be used in order to make meaningful partitioning rules. I'm willing to have some modestly unusual clothes in my closet if it means that outfits designed to push a particular brand are excluded.

[muon2]

Back to the OP and the maps of PA, let me answer the question.

PA is a D+0.3 state, almost even. The approved plan is 4D, 1d, 1e, 4r, 8R. That has a polarization of 29 and a skew of 8 (for the GOP).

My revised plan is 5D, 1d, 2e, 3r, 7R for a polarization of 28 and skew of 5. That's 3 points better but hardly close to the expected division. The high polarization shows why a neutral map still favors the GOP. The western burbs don't have the same characteristics as Philly's so they don't create swing districts around Da Burgh. You'd have to gerrymander Allegheny to move the needle there - swapping CD 18 up along the east side of Pittsburgh moves it to R+1, so it would move the skew by 1.

Switching to my original plan helps the Dems get closer to parity. The combo of Scranton and Wilkes-Barre move CD 11 from R+2 to D+5. That moves the skew to 3, but makes CD 10 uber GOP in exchange.

[muon2]

You'd have to gerrymander Allegheny to move the needle there - swapping CD 18 up along the east side of Pittsburgh moves it to R+1, so it would move the skew by 1.

So, basically, this:

?

Though it must be noted that having 18 take all of the Mon Valley is a decision justifiable on more than just skew-reducing grounds, though skew reduction is certainly sufficient reason to draw it that way.

Let it be known that even my PA map was Pub-favoring.  (I guess my district 17 is marginally more competitive as well, but that still doesn't get us to parity.)

This shows the challenge to fair mapping as opposed to neutral mapping. Neutral mapping would only look at geographic criteria like chops and erosity. Most fair map proposals also want to improve competitiveness and maintain an appropriate partisan balance, which I measure with polarization and skew.

In some states the two goals can line up reasonably well. The neutral maps for MI were reasonably low skew. The packing of Dems in Detroit was balanced by plenty of Dem-leaning burbs. With half the CDs in SE MI it isn't hard to strike a competitive balance.

PA does not fall into the category of states like MI. Only SE PA features a lot of Dem-leaning and swingy suburbs and that area is only 7 of 18 CDs. The university town of State College doesn't produce the large population of Dems that are found in Ann Arbor, and the Harrisburg area doesn't match Lansing, so there's no hope for any swing CDs in central PA. In fact the neutral plans for PA produce 6 or 7 CDs that are more Pub than any CD in a neutral MI map.

If the Allegheny river is maintained as the southern limit of CD 12 to keep erosity in check and Pittsburgh is not chopped, this was the best I could do for a competitive CD 18. It has a PVI of R+0.3. Other than the Allegheny chop, CD 12 and 18 are whole county.

[muon2]

Here is my redraw of Muon2's PA map for the CD's with which I was dissatisfied. The color variations for the chops appending the pink CD are micro-chops. The population for the Erie based CD is only off by 122 people, so no chop at all appears. I used Muon2's chops for the CD's with which I am satisfied, and did not otherwise draw. As per usual I seek rectangles and squares, and compactness (the pink and chartreuse CD's together are sort of a rectangle, and the shape of each is based on population, keeping the Scranton-Wilkes Barre MSA together, and finding a micro-chop). I disliked his split of Scranton and Wilkes-Barre, which is one MSA (and to me counts as another chop). They are back together now.

Thanks it's quite nice and it's very helpful to understand your preferences. One question, is there a reason not to use this plan in the NE? The deviations as shown are -1683, -491, -182 for the pink, lime and orange districts. It keeps towns whole in Carbon and it avoids any other microchops for those districts.

[muon2]

Thank you. I was following your lines for the balance of the map with which I was satisfied. Yes, your lines losing the chop are superior. I just took your chops from CD's with which I did not mess as holy writ. I guess that was a mistake.

I'm just trying to strike the right balance between your push to minimize erosity and train's desire to minimize chops. My adjustment is merely to refine the balance. As my optometrist would say "Is it better with lens 1 or 2?"

[muon2]

Thank you. I was following your lines for the balance of the map with which I was satisfied. Yes, your lines losing the chop are superior. I just took your chops from CD's with which I did not mess as holy writ. I guess that was a mistake.

I'm just trying to strike the right balance between your push to minimize erosity and train's desire to minimize chops. My adjustment is merely to refine the balance. As my optometrist would say "Is it better with lens 1 or 2?"

I would say that my hobbyhorse is more MSA/CSA (and muni) chops than county chops; I think that my desire to minimize county chops is not really any more strong than yours or Torie's (though it obviously does exist as well).

Fair enough. Next question (for Torie, too) - is it worth a chop to reduce polarization and/or skew?

[muon2]

Thank you. I was following your lines for the balance of the map with which I was satisfied. Yes, your lines losing the chop are superior. I just took your chops from CD's with which I did not mess as holy writ. I guess that was a mistake.

I'm just trying to strike the right balance between your push to minimize erosity and train's desire to minimize chops. My adjustment is merely to refine the balance. As my optometrist would say "Is it better with lens 1 or 2?"

I would say that my hobbyhorse is more MSA/CSA (and muni) chops than county chops; I think that my desire to minimize county chops is not really any more strong than yours or Torie's (though it obviously does exist as well).

Fair enough. Next question (for Torie, too) - is it worth a chop to reduce polarization and/or skew?

I am not interested in reducing skews via a new regime of gerrymandering. That is where the multiple maps and veto mechanism can mitigate things up to a point. I am more amenable to a chop to reduce polarization (but then there a nice chops and ugly chops), and the elasticity comment of Train has merit, but I am concerned about gaming what is competitive via AZ (although that state had the favorite son issue).  Measuring what is competitive can be tricky as places swing and trend over time (e.g., we all know what is happening to Hamilton County, Ohio, and what is competitive now, probably will not be in 4 years).  On the other hand, my bias is in favor of competitive districts, because I think over time, beyond partisanship, it will lead to a better public policy product. Both parties desperately need more moderates. I really believe that - the idea being to try to give each side something on important public policy issues, to reduce the toxicity of the public square. And boy, it is toxic now.

I suspect that just hewing to the algorithms will generate less polarization in most states substantially, and I am satisfied with that. The perfect can be the enemy of the good. Heck if you really got rid of all the skew, that would probably help the Dems, since there is more population I suspect in states that are heavily Dem that Pub, and as polarization gets more pronounced, the "natural" break moves exponentially. Finally when it comes to chops, for me I also chop to reduce erosity, and chops which increase erosity for partisan reasons kind of leaves me cold.

If we ever write a paper of course, the skew and polarization issues can be presented as options, and maps presented which demonstrate how they look (and change) with this factors overlaid, and let the states decide. Then the issue is how much weight to give them. Hopefully the overlay would not generate too much additional erosity, but that is up to the states to decide. Which raises the final issue. What if some states use the overlay, and others do not?  Then you have the unilateral disarmament issue.

This is a good time to push for this though, since at the moment the existing regime favors  the Pubs, and all of what we are doing should reduce that advantage some (particularly respecting  MSA's. which is why Train gets particularly excited about that. ). So the Dems should be satisfied with half a loaf, which is better than none. If they get the whole loaf, Pub opposition would probably kill off the enterprise in a vast swath of states. In the end you have to be practical, rather than "overly" idealistic.

Sorry for the rather discursive nature of this post. Hopefully, it made some sense.

One possibility is to consider polarization and/or skew is determining where chops should go once the number is set. In my original version of SE PA there are 6 chops for the 7 CDs. The interesting PVIs are CD 6 D+0, CD 7 D+8, CD 8 D+1 and CD 16 R+8.



If I maintain chop count, and use erosity as a measure within a chopped county but rearrange the locations of the chops to reduce polarization I get the following map. CD 6 is D+1, CD 7 is D+5, CD 8 is D+1 and CD 16 is R+5. So two additional CDs just drop into the competitive range at +5.



Though the erosity of the town grouping within Chester is actually quite good, one might not like the protuberance of CD 16 along US 30 from an esthetic point of view. One solution would be to eliminate the trichop entirely and create a chop in Lancaster keeping the same overall count. Again if the erosity is minimized within each county subject to population constraints and a goal of lower polarization I get this map. Here CD 6 is R+3, CD 7 is D+5, CD 8 is D+1 and CD 16 is R+2.



So this is what an overlay might provide. I can imagine that some states would want to use it (AZ) and some not (CA) based on their desire to create intentionally competitive districts by using political data in the mapping process.

[muon2]

Train's map of MD adds a lot of additional erosity for no doubt partisan reasons.

What additional erosity?  No, seriously, you're imagining things here.  I don't see it at. all.  

And I would absolutely draw Massachusetts in such a way as to give the Pubs at least one district they'd have a real shot at.

I guess Muon2 can help us on the erosity issue for MD, on which we disagree. We just have a fundamentally different philosophy here Train, and such is life. We will just have to agree to disagree, because we are both stubborn cusses on this, and I don't think it possible to close the gap. That happens sometimes. Moving stuff around for partisan reasons to me is perhaps not quite as fraught with peril as this communities of interest scam, but it has the potential for great mischief, and gaming.

So, you can't defend splitting Cecil from the rest of the Eastern Shore, or explain how my map is unacceptably erose whereas yours is peachy.  Gotcha.

I said let Muon2 opine on the erosity issue, so we don't have to continually bite at each other on that one. I thought about Cecil when I drew the map, but it was just too far north, and added too much erosity, and the bay turns into but a river there anyway (your comment sounds more like a COI issue anyway - the important thing is that bridge connection to Annapoplis). It also makes the NE corner CD compact, along with making the Eastern shore CD more compact. That is why I did it. I didn't even look at the partisan numbers until I was done, and doubt the Cecil issue per se is that important to partisan issues anyway. I guess I just can't persuade you that I don't have ulterior motives, so there is no point anymore in even discussing that I guess. Thanks.

I... still don't see how adding Cecil adds erosity?  Just asserting as such doesn't make it so?  It seems to me that your issue here, perhaps, is that you are overly wedded to lines that run straight north-south and east-west even when the natural geography of a state is geared towards diagonals.  This seems to me the one possible gloss of your horreur at my eminently fair and compact map which does not require ulterior motives to come into play.

And I have been very careful to try and not assume ulterior motives, as well as play along as best I can with your throw-the-baby-out-with-the-bathwater school of thought on CoI.  (Some things, of course, are just too obvious to let slide, like the Eastern Shore.  And I will not stop harping on MSAs.  But in general I have bent over backwards to accept your terms of the debate, even though I could very easily, and possibly should, challenge some of them.)  But, of course, it is hard when you seem to refuse to do the same.  Perhaps maybe you could refrain from assuming that my proposals are made with partisan intent, and then I would be more than happy to do the same with yours.

I'll get to erosity in a different post, but let me clear up what seems an inconsistency on Cecil. Historically Cecil was counted with the Eastern Shore because it sits across the Susquehanna. That history has little to do with the bulk of the population. For over a decade it has been part of the Philly MSA, the Wilmington metropolitan division to be more specific. It's not even outlying, but considered to be a central county in the MSA. Sticking with MSAs, there is no reason to favor its placement on the Eastern Shore any more than putting Anne Arundel there because of the bridge.

And speaking about that bridge, the Baltimore MSA now includes Queen Anne's county on the ES as a central county! Carroll and Harford are the outlying counties. Commuters are now such a significant portion of the population using the bridge that they have overrun the historical population. Even a decade ago when I spent some time on the ES I could see evidence of the transition, particularly where the people live on the peninsula that includes the bridge. So if MSA preservation is a criteria, one cannot ignore that anymore than one can ignore Lackawanna-Luzerne can one? Redistricting is about current populations not historical ones.

One other connection issue concerns the southern end of the Chesapeake. In discussions about WA it was generally agreed that regular scheduled ferry service could replace a public highway for the purposes of determining a connection. The Smith Island ferry links St Marys and Somerset counties. Erosity doesn't care about the bay when links are present. Yet using that link can provide additional options to preserve the aforementioned Queen Anne's link to the mainland. Then again it may not apply since the schedule doesn't show for the winter. However the boats provide a number to call for service even in the winter, so this one is a bit gray.

[muon2]

Here by the way is a map of Maryland that a drew a couple of weeks ago, which restores to the Pubs the 2 CD's that the Dems ripped off in their gerrymander. I just thought I would throw it on the pile.



And here is an alternative iteration, with perhaps a tad less erosity. It is not applicable here, but this is one instance where which map I would prefer would turn on which reduced polarization the most. In MD, good redistricting principles basically shut out competitive districts, and to reach for them, would just be a bridge too far. So it just doesn't matter here.

Well, at the very least Cecil County is normally considered to be part of the Eastern Shore, so it should really go in 1 instead of Anne Arundel.  And a 5-3 map is, if anything, gerrymandered towards the Pubs given that Maryland is just so Democratic overall.  I haven't done the exact skew calculations but I would expect 6-2 to be the "fair" expected result given the state's partisan lean.

How about something like this:



If anything, this map is still Pub-leaning, since the Dems are still only clearly favored in five of the districts.  District 3, taking in Howard and the vast majority of Anne Arundel, is both a swing district (53.9% Obama, so a PVI of D+1 maybe? ) and entirely consistent with any sort of good redistricting principle you'd care to cite.  So you should be happy with it.

First some comments on how I see erosity. For me an erosity measure should reward compact shapes, but not penalize shapes due to the underlying geography. For example the panhandle can only connect through Frederick, so the panhandle counts as if all that area and population were in Frederick. Similarly if we don't consider the open question of the ferry, all the area and population south of Queen Anne's counts as if it's in Queen Anne's. Political geography also counts so I don't assess a penalty for following the boundary between two political units such as Baltimore county and city.

Chops also factor into erosity. I view each chop as creating a new county for assessing erosity. Chops that fill in a pocket in a district can reduce erosity just like including a county that was surrounded on three sides. Chops that project into a county from an otherwise reasonable boundary will increase erosity. Microchops, though they don't count is the chop count, do factor into erosity in the same way as regular chops. Since microchops don't count as chops this is a way to create a balance between multiple microchops and a single regular chop.

With that background here's how I see the plans. All three of these plans have the same chop count of 7. That's the maximum that would be needed for 8 CDs. For Torie 1 the yellow district dip into Montgomery and the brown CD moving up into Anne Arundel do nothing to make those CDs more compact than they were so the chops hurt erosity. Torie 2 fixes that and moves the erosity down as he expected it would. By my measure it drops from 25 to 23 points. As an aside, I count 3 CDs in Torie 1 that span the central counties of the Washington and Baltimore MSAs. Torie 2 brings that down to 2, one of which is the barely more than a microchop of the yellow CD into Prince George's.

Train's map moves those 7 chops into a significantly different pattern. But since the longer north-south shape of the blue CD 1 is due to the isolated Delmarva population there's no penalty for that. And as I commented earlier the Baltimore wrap around preserves the CD in the county and follows the political boundary so it generates only slightly more erosity than the NE corner CD on Torie's plan. In the end each of the CDs have either a point more or less erosity than their closest equivalents in Torie 2. The net effect is a wash and I score train's plan the same as Torie 2. They both can go forward.

As I noted at the outset, the real issue comes to the existence of chops. One reason I start with apportionment regions is to identify where I can conserve on chops. In this case I looked at the regions with an eye towards MSA preservation, but I will suggest that there must be at least one CD and one region that spans the MSAs. The Washington MSA counties down the western shore but without Frederick are just barely above 3 CDs and with one microchop can be divided nicely within the 0.5% population limit. By putting all of Glen Burnie into the Baltimore city CD I was able to avoid any other chops for that CD or the Anne Arundel-Howard CD. Eliminating those two chops also has the effect of reducing the erosity by 1 point. I can probably clean up the line between CD 4 and 5 with some work, but the precincts aren't the best unit to match city lines. Edited to reflect a better line through PG that avoids city splits and keeps both CDs within 0.5%.



For the record the BVAPs for CD 4, 5, and 7 are 41.2% (plurality), 54.9% and 53.1%. The partisan distribution is 5D, 1e (D+0.1%), 2R for a polarization of 14 and a skew of 0 after accounting for the expected Dem lean.

[muon2]

If the Smith Island ferry crossing counts as a connection, then it is possible to eliminate the microchop so that there are just the five main chops This substantially reduces the population inequality. In the plan below the range drops from 7066 to 3820 which corresponds to an inequality score change from 13 to 8. The increased erosity created by the split of the ES is compensated by reductions elsewhere so that the erosity is the same as for my previous plan adjusting for the ferry.

There is also a noticeable increase in competitiveness with the seats going 3D, 2d (D+1.9, D+5.2), 2e (D+0.1, R+1.2), 1R. That's a 4 point improvement though it does create a 1 point skew for the Dems.

[muon2]

Oh damn, I realize in the second map that the yellow CD raped the Baltimore MSA, so back to another map in my pile, that had the Eastern shore CD fill that bit in, pushing the red CD east. I will put that map up tomorrow. Pity that. The red CD was just so gorgeous.  It will mean a potentially nasty chop of Hartford County, and Cecil rather dangling out there (which sucks, although it will constitute a bone thrown to Train on his Cecil thing). I think maybe my first map will get a higher score now, depending on the chop count.

Of course, I'm not so committed to MSAs. jimrtex points to other measures, and I'm not convinced yet that CD-based use of them is the right direction. Perhaps it makes sense at the level of apportionment regions, and that is worth exploration. As I noted with MI, the problem only really showed up when the MSA was more than one county, yet containable in a single CD - to wit Lansing. As I noted in my MD post, without using the ferry one must have at least one link that carves a piece of both the Washington and Baltimore MSAs. I don't see that as a problem, any more than a single shared split is an issue in the MI rules. That's why I raised the point when there were multiple CDs spanning the same MSAs.

We also need to think about compactness. Yes in the second map, the "Eastern shore" CD has a nice rectangular shape, but it has become rather obscenely huge. Compactness should count for something. We need to spend more time on that. Compactness matters - yes, it really does. Heck even the Dem mole on the AZ commission bearing the false flag of an "independent" yammered on about the compactness issue. If she and I agree on something, that is quite an event actually.

This comment worries me more than any other. Compactness measures are all based on theoretical models of ideal shapes. The papers I have read and occasionally posted show that there are a myriad of competing geometric measures of compactness and all have serious glitches when applied real redistricting which occurs on real natural and political shapes, based on rivers and roads, and populated with a density based on history that may no longer have relevance to electoral districts. The fundamental problem boils down to one that makes a small protuberance in an urban area as important a consideration as a large one in a set of rural counties. Mathematicians would call this a scaling problem.

The suggestion of looking at erosity as opposed to compactness was brilliant and offered what seemed to me to be the only sound path out of decades of unresolved debate. Rolling back to compactness only returns to the arguments and counter-arguments of those same decades of experts with no better conclusion. The model I've spent the much of this year developing and refining is designed to deal with the scaling problem by recognizing that the basic units of redistricting, whether census blocks or VTDs also scale in size with density. Graph theory has tools to deal with this that are lacking in classical geometry.

If you really want to return to compactness as we customarily see it, so be it. I fear though that, as a certain classic song goes, the "mission is a failure, ... prepare the transit beam."

[muon2]

Regarding the matter of using regions to take cognizance of them, I was never clear how using regions worked mechanically to force/influence changes in maps (just as I am not clear exactly how to measure erosity except by eye, even though I keep asking that we work on  that issue, and try to agree on the best approach that makes stuff look appealing to the eye, and  on that one I need your mathematical mind to help me), as opposed to just being a useful tool to find micro-chops, whatever one might want to do with them. Perhaps you might explain that.

On the issue of compactness, maybe there is no good solution, but it is a negative for a CD to just wander all over the state, like that AZ CD that went from Sedona to Snow Low via the Indian reservations. Maybe erosity measures that we have not yet defined, is the best that we can do. Sometimes unfortunately the mind of man is just too inadequate to fashion rules that really work on a global basis. Such is life.

One of my observations from public discussions of gerrymandered maps is that what bothers people are the really gross shapes that have nothing to do with natural geography. They don't mind river bends and modest deviations from a straight line. They do take offense at shapes that create unnatural deviations from regular shapes. In a sense the public has a threshold where a district is shaped well enough, beyond which they'd rather concentrate on effects other than the shape such as geographic integrity or competitive districts. Most mathematical models of compactness continue to reward improving the shape ad infinitum with no natural threshold. One goal is build in a sense of a threshold based on the shape of the underlying geography.

Let's use pentonimoes as counties with uniform population in a perfectly rectangular state. Here's an example I found at random on google search.



Now suppose we have to divide this into 4 districts. Clearly there are a wealth of ways to group these 12 counties into contiguous groups of 3 pieces each. Since the ideal division without counties would be four rectangles that are 3 x 5, most compactness formulas are going to reward whatever mechanism gets you there. If the formula entertains chops then it would force so many chops to get those rectangles that I expect the public would be hugely unhappy. So we recognize that something has to be traded between the shape and the number of chops.

Traditional formulas that try to work with the compactness of an area would generally consider a district made of the magenta, blue, and light blue counties in the NW corner more compact than the pink, purple and orange district in the NE corner. That's because it's more square or circular in shape than the elongated district I describe for the NE. Yet, I contend most ordinary observers would say that the elongated one makes the better district. We would describe it as less erose, and from that I conclude that one has to look more at the perimeter than at the area. In particular the internal perimeter is what matters to the public observer, because they will always forgive erose shapes due to the external border of their state.

For this simple example one could start by finding the division that minimizes the total perimeter of all districts. The outer perimeter is the same for every plan so that can be subtracted from that total. The difference that is left counts for each district on the boundary so it should be divided by two to get the unique amount of perimeter created by the division into districts.

This works well when all the boundaries that are under consideration are built from straight lines. But suppose that the boundaries are sometimes straight and sometimes winding rivers. A pure formula like the one I just described is going to be strongly biased towards avoiding putting winding county edges on the perimeter of the district because it adds to the length without changing the area. Yet my experience is that the public sees no reason not to treat the county river boundary as equal to the straight line segment as long as there's a bridge across the river on that segment.

My solution for this is to count segments instead of the actual distance. Each boundary between two different counties on either side of the district boundary count as a segment. On the average for the straight line pentominoes it is equivalent to the actual length, but it has none of the problem of biasing against naturally wiggly county lines.

In an upcoming post I'll address the connection between this type of erosity and chops.