FL: Rereredistricting

[muon2]

I was looking at the stats on the linked map and continue to be amused at the strange but not uncommon counting that is done for city and county splits. The count all districts that are in counties that are split. There is no benefit for a district entirely within a county and a double penalty for any chop of a county or city too small for a single district. The double count occurs since counties entirely within a district are not counted, but the moment the county is chopped it counts for two. Subsequent chops only increase the count by one each, so it is distinctly non-linear penalizing the first chop.

So in theory there is no penalty for a chopped county that miraculously does not also chop some internal subdivision? I think the assumption is that if a county is chopped, then a subdivision will be chopped, and thus the double count of the first internal subdivision count, is really counting one point for the county chop, and a second point for the internal subdivision chop. That seems sort of reasonable I guess. Your/our system is to just allow one internal subdivision chop, with no additional penalty, and test where it is by an erosity test (if I understand the system, which is always a big if). They have their own erosity test, which I must say does tend to make for pretty CD's. I would not reject it out of hand actually. It seems to work well.

They treat counties and munis as independent variables since munis can overlap counties. The score a re tabulated for each type separately. I can use counties to illustrate the effect of their system which counts fragments. I will use the newly linked map as an illustration with the fragment and chop counts.

Marion 3:2
Citrus 0:0
fragments 3, chops 2

If CD 2 chopped into Citrus instead of Marion and CD 3 and 11 split Marion between them we would have the following.

Marion 2:1
Citrus 2:1
fragments 4, chops 2

The fragment count increased without changing the chop count. This is the double penalty I spoke about: the first chop into a county counts 2 points. It forces a plan to favor multichopping a county when possible. We had a lengthy thread debate on this a few years ago leading to our current definition of chops. The consensus was that each additional chop should count the same whether it was in an already chopped county or in an unchopped county.

The FL plans include a number of separate standard measures of compactness. They each have known weaknesses, so I assume that they list multiple measures so that a plan that might exploit one measure's weakness would show poorly on other measures. The result is that there is no rule, but guidance to hopefully unmask gerrymandering.

[muon2]

The first Senate map is the cleanest map, and also gives the Hispanics in FL-9 the best representation.   

The House Map isn't all that much different though.    I really believe FL-9 will actually end up being a safe Dem seat regardless since there is such a massive influx of Puerto Ricans into the area, so it's really moot which map is chosen in the end.   Going off the 2010 Census does have it's limitations.

The second senate map has better metrics.

As in county chops?   What would the metrics be?

County and muni chops and erosity (compactness) measurements. The second Senate map had one less county chop and one more muni chop than the House map, and its compactness score was about equal.

Though keep in mind their metric measures fragments not chops. As I noted earlier that can create some unexpected choices in map-making, particularly favoring placing as many chops as one can in one county, rather than spreading them out among multiple counties.

[muon2]

"Their complaint on 26/27 is clearly partisan motivated, but they somehow have to convince the judge that the legislature had done the pairing for partisan reasons."

What gave you the impression that the judge thought the pairing was for partisan reasons? In my view, it clearly wasn't. In fact there is no evidence that staff knew the residences of the incumbent Congresspersons.

A FL-05 that kept Leon County whole would be performing in my view. A clear majority of the Dem primary voters would still be black. The Judge would show a lot of chutzpah however in making Leon whole, given that the district was not contested in court. And given that it was not contested, nobody pointed out that there is a SCOTUS decision that racial gerrymandering plus erosity not done for partisan reasons is not Constitutional, and that clearly is what FL-05 is in my view. Maybe the Judge will find that decision on his own if he gets motivated enough.

I didn't see that there was evidence presented to the judge that a sub 50% BVAP district would be performing. I agree that it would be, but absent any evidence as to the floor to maintain performance, I'm not sure how the judge goes that direction.

You must be logged in to read this quote.

To me this highlights a problem with all commissions given a mandate to ignore partisan factors. Constructing minority districts inherently uses partisan data since the VRA is based on the ability to elect, which usually requires a partisan primary. CA avoids this in principle with their top-two system, but I won't be surprised if some group at some point does a statistical analysis based of the party label there and ends up showing that partisan factors matter for minority performance. Even non-minority districts can be impacted by the contradiction in partisan data usage if they are in close proximity to minority districts.

[muon2]

But once one drops below 50% BVAP one has to show where the floor for performance is in FL-05. It looks like that was where some of the questioning about FL-26 was directed - where is the floor for Hispanic performance? The answer was that the LWV map dropped below the necessary level, and because of citizenship and different minority groups, some analysis was needed. The FL-05 district is over 50% BVAP so there is a presumption of satisfaction without providing partisan analysis.

[muon2]

My bad. I thought they had pushed the BVAP over 50 in FL-5 but I see they set it at 45.11%. I still don't know how low one could push BVAP and still get minority performance. I think that is a factor in determining the constitutionality.

[muon2]

My bad. I thought they had pushed the BVAP over 50 in FL-5 but I see they set it at 45.11%. I still don't know how low one could push BVAP and still get minority performance. I think that is a factor in determining the constitutionality.

If blacks are a majority in a Dem primary, in a heavily Dem district, how could it not be performing?

Normally this is determined through a statistical analysis of precinct voting, or by the evidence that a candidate of choice has consistently won at a certain BVAP. Neither apply here to my knowledge.

[muon2]

My bad. I thought they had pushed the BVAP over 50 in FL-5 but I see they set it at 45.11%. I still don't know how low one could push BVAP and still get minority performance. I think that is a factor in determining the constitutionality.

If blacks are a majority in a Dem primary, in a heavily Dem district, how could it not be performing?

Normally this is determined through a statistical analysis of precinct voting, or by the evidence that a candidate of choice has consistently won at a certain BVAP. Neither apply here to my knowledge.

Is this some assertion that there is some probability a significant percentage of black Dems, would not vote for the black Dem in a Dem primary in N. Florida, and in this case, in an open seat?  On that one, I think one can take almost Judicial Notice of.

No, it's saying that there are standards that are typically followed. If it's based on percentage control in the primary then that evidence is presented to the court and is used in the findings. We can intuit the likely outcome from our knowledge, but I don't expect a judge to do that.

[muon2]

I'm always fair and balanced - except when I'm not.

This is all a legal exercise here, following the cases, that weird Florida law, and that weird scoring system used in Florida, that gives you two penalty points for the first chop of a county, and one point for each additional chop, so more counties being multi chopped, in exchange for more counties not being chopped at all, gets you a better score, e.g., it's better to have one tri-chopped county, that two bi-chopped counties. Their erosity scoring also tends to favor more my kind of "artistic" map, with nice clean lines, and rectangles, more than Muon2's road cut system, and super penalties for macro chops into counties. So I am in my element here.

I don't think Florida has a count system.

"respects political boundaries" (paraphrase) is a subjective criteria. It may be that they are focusing on the number of chopped counties, and it happens that many of the larger counties are multi-chopped.

According to the reports put out on the plans they do have a count system. They identify the number of counties that are split and then sum the number of districts over all split counties - ie they count fragments. Then they repeat the counting exercise for municipalities. The two counts are viewed separately from each other.

They sort of punt on erosity and instead compute something like 21 separate measures of compactness. It has the effect of forcing an "artistic" approach since anything else that heads towards a gerrymander should get flagged by at least one of the measures. Of course it's all fairly opaque to the public.

It is interesting to note that among the compactness measures are some related to driving time and distance. It's the first time I've seen those formally used by a state in their official analysis. It also marks a step towards my view of roads as the best tool to measure erosity.

[muon2]

Torie, as a side note, do you see FL 20 as required by the VRA? Set aside the present case and FL law. Are the black areas sufficiently contiguous by your usual standard?

[muon2]

I'm always fair and balanced - except when I'm not.

This is all a legal exercise here, following the cases, that weird Florida law, and that weird scoring system used in Florida, that gives you two penalty points for the first chop of a county, and one point for each additional chop, so more counties being multi chopped, in exchange for more counties not being chopped at all, gets you a better score, e.g., it's better to have one tri-chopped county, that two bi-chopped counties. Their erosity scoring also tends to favor more my kind of "artistic" map, with nice clean lines, and rectangles, more than Muon2's road cut system, and super penalties for macro chops into counties. So I am in my element here.

I don't think Florida has a count system.

"respects political boundaries" (paraphrase) is a subjective criteria. It may be that they are focusing on the number of chopped counties, and it happens that many of the larger counties are multi-chopped.

According to the reports put out on the plans they do have a count system. They identify the number of counties that are split and then sum the number of districts over all split counties - ie they count fragments. Then they repeat the counting exercise for municipalities. The two counts are viewed separately from each other.

They sort of punt on erosity and instead compute something like 21 separate measures of compactness. It has the effect of forcing an "artistic" approach since anything else that heads towards a gerrymander should get flagged by at least one of the measures. Of course it's all fairly opaque to the public.

It is interesting to note that among the compactness measures are some related to driving time and distance. It's the first time I've seen those formally used by a state in their official analysis. It also marks a step towards my view of roads as the best tool to measure erosity.

That their redistricting software reports certain metrics, does not necessarily mean that they have a scoring system.

Comments that I read about the merits of the House vs Senate plans implied that those scores are their implementation of the vaguer language in the Con Amend.

[muon2]

The map jimrtex created seems like an excellent starting point to find a plan the chops the fewest counties. Following the example of Miami-Dade+Monroe, I appended some counties to the large counties to bring districts within 5% of the quota. The remaining small counties were then balance to provide for all districts within 5% of the quota.

[muon2]

The Orange opportunity SD 12 in 9072 is 35.9% BVAP and 21.2% HVAP+HBVAP. In my version shifting it SE in Orange gets 32.1% BVAP and 27.7% HVAP+OVAP. The minority performance does rise slightly and the chop count goes down for the Lake SD, but at the cost of compactness and greater inequality.

The Tampa-St Pete opportunity SD 12 in 9072 is 34.5% BVAP and 26.5% HVAP+HBVAP. In my version putting it entirely in Hillsborough gets 21.7% BVAP and 32.3% HVAP+OVAP. There is a definite drop in minority performance and again 9072 has lower inequality, but my plan would be substantially more compact and it avoids the traveling chop of Pinellas and Hillsborough.

[muon2]

The Orange opportunity SD 12 in 9072 is 35.9% BVAP and 21.2% HVAP+HBVAP. In my version shifting it SE in Orange gets 32.1% BVAP and 27.7% HVAP+OVAP. The minority performance does rise slightly and the chop count goes down for the Lake SD, but at the cost of compactness and greater inequality.

The Tampa-St Pete opportunity SD 12 in 9072 is 34.5% BVAP and 26.5% HVAP+HBVAP. In my version putting it entirely in Hillsborough gets 21.7% BVAP and 32.3% HVAP+OVAP. There is a definite drop in minority performance and again 9072 has lower inequality, but my plan would be substantially more compact and it avoids the traveling chop of Pinellas and Hillsborough.

Have you submitted your map to the state of Florida, or were you using DRA?

I didn't know they were still taking submissions. I think I still have an account on their website with the software. I would also have to know where the 4% tolerance comes from before submitting anything.

[muon2]

Is this better? I increased the chop count by 1 to decrease the range from 7.2% to 5.3%. With 2 more chops in Hillsborough and Broward I could reduce it to 3.8%.

[muon2]

That staff description is consistent with what I have seen. The people given the task haven't typically thought about how to make a good map, and they stumble through the process. I have looked at making a county-level app to build regions, and think it would be a great tool for the next round of redistricting.

BTW, the link to the google doc spreadsheet isn't working for me.

[muon2]

That staff description is consistent with what I have seen. The people given the task haven't typically thought about how to make a good map, and they stumble through the process. I have looked at making a county-level app to build regions, and think it would be a great tool for the next round of redistricting.

BTW, the link to the google doc spreadsheet isn't working for me.

Does this work any better?

Florida Redistricting Spreadsheet

Looks good.

I assume the use of 20 regions in the spreadsheet is the maximum given the number of large counties with more than 1 district. The existence of 18 regions then implies that there must be 2 county chops beyond the number needed to accommodate chops in large counties. The total chop count should then be the number of districts in large counties + the difference between the ideal and actual number of regions. If it is more than that then there are unneeded double chops. Did I read that right from the spreadsheet?

The FL interpretation would then require looking at where the chops occurred, since they count fragments, not chops. Is that also correct?

[muon2]

"The existence of 18 regions then implies that there must be 2 county chops beyond the number needed to accommodate chops in large counties."

I can't reason my way to understanding why the above is true. I know it has something to do with there being 20 seats associated with surplus population, and only 18 regions, but I can't quite connect the dots.

First, a region is a cluster of whole counties whose population is equal to a whole number of districts that are within the required variance of the quota.

Let's assume that in a large county there are as many whole districts as possible. Each large county can then be viewed as a small county with a population equal to the remainder after all the wholly contained districts are subtracted from the original population. When a large county is used in a region, the number of districts in that region increases by the number of whole districts in the large county.

So, let's make the reduction as described above. The FL map would now be represented by 67 counties all of which have population less than a whole district. There are initially 40 seats and enough large counties to hold 20 of those seats with excess population to spare. That leaves 20 seats to place on our map of 67 counties with suitably reduced populations.

The best we could do is find a configuration of county clusters such that all are within the required variance of the quota (taking into account extra seats for large counties). That configuration would be 20 regions, with multiple districts only for those clusters that include large counties. This configuration would only require chops in the large counties, so the chop count is minimized. That number is the ideal number of regions.

The variance requirements can force some clusters on the reduced map to have the population of two or more districts. The Bradenton-Sarasota region on our map above is an example of that. By lumping those counties together as a two district region it insures that one of those counties must be chopped. The same thing happens with the Ocala-Gainesville region. No other region requires a chop in a small county. Hence my statement that the 18 regions on the map represent two chops more than the minimum required by the large counties alone.

Each wholly contained district requires a chop of a large county and each region less than the ideal represents the chop of another county. Thus the minimum chop count for a plan is equal to the number of districts that can be placed wholly within large counties plus the difference between the ideal number of regions and the actual number of regions.

[muon2]

Is the short cut version of the explanation, that with only 18 regions, but 20 seats outside the multi district counties, there need to be two regions outside the big counties with two districts, thus necessitating two extra chops? If the populations balanced better, you would create more regions to avoid the extra chops, but you can't, so thus you are short the requisite number of regions to avoid the chops?

Works for me.

[muon2]

Have you submitted your plan?

[muon2]

The Florida legislature is in the process of redrawing the state senate map, and are accepting public input. One of the maps was automatically drawn, and the algorithm is described here:

Autoredistricting

This is the map

Autoredistricted senate map (PDF)

The map might not be legal. The Florida Constitution forbids drawing a map with the intent of favoring or disfavoring a political party. Since an explicit constraint is partisan equity, this could be regarded as acting favorably toward a party.

It also ignores county and city boundaries and doesn't create or preserve the opportunity to elect for minority groups.

But it is an interesting approach.

The panhandle looks strange. It looks like there is a water only connection to Pensacola. I note they have an unusual definition of contiguity which makes it not absolute so that it can be optimized, perhaps that led to the connection.

I also agree that the software is not compliant with the FL constitutional provisions.

[muon2]

Can you file an amended map that corrects the errors? What is the deadline?

I completely agree about the subdivision of counties. It was a feature I used in the VA redistricting exercise last year.

It would really help in Florida to have some pre-redistricting organization of counties. If you are trying to build using cities, you get really ugly boundaries with lots of holes, and then you have to go back in and build by VTDs, and then use the city again to make sure the VTD's didn't intrude on another city. The district builder and/or Silverlight seems to have problems with complex geometries with holes, as the holes get filled. You also get slivers between city limits. It appears that some city limits are partially within canals, but don't meet at the center of canal, leaving a narrow strip between cities.

You could have a hierarchy of: County, County Areas, and Neighborhoods.

A county area would be an incorporated city, plus adjacent unincorporated areas, or larger unincorporated areas, which might be similar to CDPs, or rural areas. It would provide 100% coverage of the county, similar to towns in the Northeast and Midwest.

A neighborhood would be an area within a county area.

County areas and neighborhoods would serve not only as building blocks for congressional and legislative districts, but for county commissioner districts and city council districts.

I think that Census geography is released in about December of the '0 year, so that would provide time to build block equivalencies between county areas and neighborhoods.

[muon2]

The coalition plaintiffs have submitted a plan that they said creates a Black opportunity district entirely within Hillsborough County. So I decided to compare:



The green district is mine, the blue district is theirs, with the green district on top. The portion of the green district to the west and south of the black line is in my district. The visible blue portion is in their district, but not mine.

The major difference is that I have the arm off to the northwest, intended to pick up Hispanic population, vs. their map shifting to the east, to pick up areas that have a black presence, but certainly not predominant.

My district: 29.28% BVAP, 31.77% HVAP
Theirs: 29.17% BVAP, 23.35% HVAP.

The plaintiff's district is more compact, but yours provides a better opportunity with both higher BVAP and HVAP. How do city chops compare?

[muon2]

How does one distinguish between a racially-motivated gerrymander and an effort to promote minority interests when considering the cross-bay district?

[muon2]

How does one distinguish between a racially-motivated gerrymander and an effort to promote minority interests when considering the cross-bay district?

After they drew the Hillsborough-district, they scored it, and then asked the legislative counsel. In the part that I have listened to, they haven't heard back yet.

The lawyers for the senate and the house said that blacks and Hispanics were not cohesive in the primary and so the map-drawers should go into another county to maintain a black opportunity-to-elect district.

This could prove interesting in the upcoming trial. In the congressional case, the lawyers for the house and senate failed to cross-examine the plaintiff's expert witness, and whether he had considered primary results in his analysis.

When considering the possibility of a Hispanic opportunity seat in the Orlando area, the lawyers for the legislature were of the opinion that there was not sufficiently racially polarized voting to warrant creation of a district.

In the senate case, the plaintiffs presented their Hillsborough-only district as an alternative, and then formally withdrew it. They must have got push back from somewhere.

It appears that the House has not presented a remedial plan.

There's a lot of advice about performing minority districts from legislative attorneys. Is their advice disputed by any of the other parties?

[muon2]

He was also pushing really hard on metrics. Hillsborough County includes the lower part of Tampa Bay all of the way to the Gulf of Mexico. If you include the tail in a Hillsborough-based district, it blows up the Reock score. You would think you would want to hide how you were gaming the metrics rather than badgering a witness about how more compact is better.

This was also my observation during the 2010 cycle. These tails on counties are not unique to FL. The artificial distortion of the score led to my notion of erosity based on connections not on shape.