Non Partisan Redistricting.

[JerryArkansas]

Is non partisan redistricting really non partisan.  You decide.

[Vazdul (Formerly Chairman of the Communist Party of Ontario)]

Well, it's certainly better than simply giving whichever side has a majority in the state legislature carte blanche to gerrymander to their hearts' content. But, all too often, "non-partisan" redistricting is in practice bipartisan gerrymandering, or (as happened in Arizona and New Jersey this past cycle), partisan gerrymandering where an "independent" sides with either the Republicans or the Democrats. Rather than give any group, partisan or "not," full discretionary power when it comes to redistricting, it would be better to impose strict restraints on the process such as respect for municipal and county boundaries and communities of interest.

[jimrtex]

Well, it's certainly better than simply giving whichever side has a majority in the state legislature carte blanche to gerrymander to their hearts' content. But, all too often, "non-partisan" redistricting is in practice bipartisan gerrymandering, or (as happened in Arizona and New Jersey this past cycle), partisan gerrymandering where an "independent" sides with either the Republicans or the Democrats. Rather than give any group, partisan or "not," full discretionary power when it comes to redistricting, it would be better to impose strict restraints on the process such as respect for municipal and county boundaries and communities of interest.

The problem with commissions is that they are supposed to be wise amateurs.  This makes them manipulable by lawyers.

In California the selection process was particularly biased towards producing racially gerrymandered maps.   Two of the non-Democratic members were probably not Democrats because the party was too moderate, while the supposition was that those members would be somewhat muddled towards the middle.

Potential members were rejected for saying quite sensible things.  One candidate said they really didn't believe in that race thing, so they guessed they were European-American.  Another had been quite involved in redistricting supervisor districts in their county.  They made the comment that the people didn't understand that the districts had to have the same population, and was said to be insensitive to public input.  You then get the lawyers for the commission explaining what they think the law means, and they end up going along because it sounds reasonable.

[Bleeding heart conservative, HTMLdon]

I'm curious as to how much support there would be on this forum for algorithm-generated redistricting.

[Smid]

I'm curious as to how much support there would be on this forum for algorithm-generated redistricting.

I'm opposed to that. I'm in favour of Muon just redistricting the whole country, at both a state and federal level. If he is concerned that this committee might be looking too partisan, I'm happy for him to choose a suitable Democrat to help him draw sensible boundaries. That's the solution to ending gerrymandering.

[jimrtex]

I'm curious as to how much support there would be on this forum for algorithm-generated redistricting.

It doesn't have to be done by algorithm.  Ants define their territories, why can't humans?

Give and Take

Start by selecting a random sample of the electorate that is stratified by some small area such as an election precinct.  So a small panel represents each precinct and more generally represent larger areas such as cities and counties, and at least their region of the district.

The redistricting process is conducted in a series of rounds.  In each round, a district gets one turn, in which it may either give territory and population to its neighboring districts, or may take territory and population from its neighboring districts.

Each round is divided into two phases, a taking phase where smaller districts take territory from adjoining larger districts; and a giving phase where larger districts give territory to adjoining smaller districts.

The order of turns within a round may change during the round.  After each district has its turn the next district is chosen based on the district population after the turn.

Taking Phase

The district with the least population takes the first turn.  Districts with less than the ideal population take the subsequent turn based on their population after the previous district's turn.

The amount to be taken is based on the deficit of the district's population below the ideal population divided by ten (10).  On subsequent rounds, the divisor is 9, 8, 7, ....  The intent is to achieve convergence quickly, without oscillation.

A district may take population from adjacent districts that have a larger population.  They may not take population from a district that took territory from it in earlier in the same round (no tug-of-wars).   If there are no such districts, the taking district passes for that round.

Each voter on the redistricting commission who is in the taking district designates an area that they would like to be added to their district.   For example, if a taking district has a population of 630,000 and the ideal population is 700,000, the deficit is 70,000.  Divided by 10, the amount to be added is 7,000.   So each voter designates precincts with a total population of 7,000 (zero or more whole precincts, and a fractional part on one).  The voter ranks these in preference order, with the constraint they have to be ordered beginning at the current border.

The votes are weighted based on the population of the voter's precinct and the number of commissioners from that precinct.   So if a precinct has 1,500 persons, the precinct should control 1500/630000 of the total taking (0.238%).  If there are 3 commissioners from the precinct, each controls 1/3 of that (0.079%).  Since 7,000 persons are to be added, multiply 0.079% times 7000, so that the commissioner controls 5.56 votes.

The actual precincts that are taken are determined using STV.  The quota for a precinct to be taken is simply its population.  The precinct with the largest relative surplus will be "elected" and added to the taking district.   Its surplus votes will be distributed among the next-ranked preferences of the precinct's whose votes contributed to the surplus.  Once, there are no precincts remaining with a surplus, we eliminate all precincts that have not received any votes at that point.  We then beginning eliminating precincts, based on which has the smallest votes to population ratio.  If the votes from a taking precinct become exhausted, they are added to the vote total of the precinct with the largest ratio of votes to population total.

Giving Phase

The district with the greatest population above the ideal population takes the first turn.  Districts with greater than the ideal population take the subsequent turn based on their population after the previous district's turn.

The amount to be given is based on the deficit of the district's population above the ideal population divided by ten (10).  On subsequent rounds, the divisor is 9, 8, 7, ....  The intent is to achieve convergence quickly, without oscillation.

A district may give population from adjacent districts that have a smaller population.  They may not give population to a district that gave territory to it in earlier in the same round (no tug-of-wars), but they may give additional territory to a district that took from it earlier in the round.   If there are no such districts, the giving district passes for that round.

Each voter on the redistricting commission who is in the giving district designates an area that they would like to be given from their district to neighboring districts.   For example, if a giving district has a population of 770,000 and the ideal population is 700,000, the surplus is 70,000.  Divided by 10, the amount to be subtracted is 7,000.   So each voter designates precincts with a total population of 7,000 (zero or more whole precincts, and a fractional part on one).  The voter ranks these in preference order, with the constraint they have to be ordered beginning at the current border.  A commissioner may vote to give his precinct (that is, he wants to move to a neighboring district).

The votes are weighted based on the population of the voter's precinct and the number of commissioners from that precinct.   So if a precinct has 1,500 persons, the precinct should control 1500/630000 of the total giving (0.238%).  If there are 3 commissioners from the precinct, each controls 1/3 of that (0.079%).  Since 7,000 persons are to be subtracted, multiply 0.079% times 7000, so that the commissioner controls 5.56 votes.

The actual precincts that are given are determined using STV.  The quota for a precinct to be taken is simply its population.  The precinct with the largest relative surplus will be "elected" and given to the neighboring district.   Its surplus votes will be distributed among the next-ranked preferences of the precinct's whose votes contributed to the surplus.  Once, there are no precincts remaining with a surplus, we eliminate all precincts that have not received any votes at that point.  We then beginning eliminating precincts, based on which has the smallest votes to population ratio.  If the votes from a giving precinct become exhausted, they are added to the vote total of the precinct with the largest ratio of votes to population total.

This completes one round.

[Skill and Chance]

Well, it's certainly better than simply giving whichever side has a majority in the state legislature carte blanche to gerrymander to their hearts' content. But, all too often, "non-partisan" redistricting is in practice bipartisan gerrymandering, or (as happened in Arizona and New Jersey this past cycle), partisan gerrymandering where an "independent" sides with either the Republicans or the Democrats. Rather than give any group, partisan or "not," full discretionary power when it comes to redistricting, it would be better to impose strict restraints on the process such as respect for municipal and county boundaries and communities of interest.

The problem with commissions is that they are supposed to be wise amateurs.  This makes them manipulable by lawyers.

In California the selection process was particularly biased towards producing racially gerrymandered maps.   Two of the non-Democratic members were probably not Democrats because the party was too moderate, while the supposition was that those members would be somewhat muddled towards the middle.

Potential members were rejected for saying quite sensible things.  One candidate said they really didn't believe in that race thing, so they guessed they were European-American.  Another had been quite involved in redistricting supervisor districts in their county.  They made the comment that the people didn't understand that the districts had to have the same population, and was said to be insensitive to public input.  You then get the lawyers for the commission explaining what they think the law means, and they end up going along because it sounds reasonable.

I don't actually see what the harm is in having greens or tea partiers as the Independents on a California format commission.  Nothing can pass without a majority of R's, D's and I's all agreeing to it.  Even if all 4 of the Independents are far left and they all get on board with some 45D-8R map the Democrats propose, they still can't pass it without support from at least 3 Republicans.  So the commission would just deadlock and the process would go to court. Now the court theoretically could end up adopting the 45D-8R map that looks like MD, but do we really think that's likely?

Furthermore, the interests of far left and far right I's don't necessarily align with those of the D's or R's.  A green or a tea partier is probably more interested in getting the most ideological candidates possible out of the primaries.  So they might actually want a map with lots of R+20 and D+20 districts to elect "true conservatives" or "true liberals."

Now the AZ/NJ system does totally break down with a hard left or hard right I as the tiebreaker, as we arguably saw last year.   

[jimrtex]

The problem with commissions is that they are supposed to be wise amateurs.  This makes them manipulable by lawyers.

In California the selection process was particularly biased towards producing racially gerrymandered maps.   Two of the non-Democratic members were probably not Democrats because the party was too moderate, while the supposition was that those members would be somewhat muddled towards the middle.

Potential members were rejected for saying quite sensible things.  One candidate said they really didn't believe in that race thing, so they guessed they were European-American.  Another had been quite involved in redistricting supervisor districts in their county.  They made the comment that the people didn't understand that the districts had to have the same population, and was said to be insensitive to public input.  You then get the lawyers for the commission explaining what they think the law means, and they end up going along because it sounds reasonable.

I don't actually see what the harm is in having greens or tea partiers as the Independents on a California format commission.  Nothing can pass without a majority of R's, D's and I's all agreeing to it.  Even if all 4 of the Independents are far left and they all get on board with some 45D-8R map the Democrats propose, they still can't pass it without support from at least 3 Republicans.  So the commission would just deadlock and the process would go to court. Now the court theoretically could end up adopting the 45D-8R map that looks like MD, but do we really think that's likely?

Furthermore, the interests of far left and far right I's don't necessarily align with those of the D's or R's.  A green or a tea partier is probably more interested in getting the most ideological candidates possible out of the primaries.  So they might actually want a map with lots of R+20 and D+20 districts to elect "true conservatives" or "true liberals."

Now the AZ/NJ system does totally break down with a hard left or hard right I as the tiebreaker, as we arguably saw last year.   

If a plan does not receive 3 votes in each group, then the Supreme Court appoints special masters to "adjust" the plan.   So there is continuous pressure to play nice and accept the advice of the lawyers.  Even the referendum provision ended up being a farce.

[Skill and Chance]

The problem with commissions is that they are supposed to be wise amateurs.  This makes them manipulable by lawyers.

In California the selection process was particularly biased towards producing racially gerrymandered maps.   Two of the non-Democratic members were probably not Democrats because the party was too moderate, while the supposition was that those members would be somewhat muddled towards the middle.

Potential members were rejected for saying quite sensible things.  One candidate said they really didn't believe in that race thing, so they guessed they were European-American.  Another had been quite involved in redistricting supervisor districts in their county.  They made the comment that the people didn't understand that the districts had to have the same population, and was said to be insensitive to public input.  You then get the lawyers for the commission explaining what they think the law means, and they end up going along because it sounds reasonable.

I don't actually see what the harm is in having greens or tea partiers as the Independents on a California format commission.  Nothing can pass without a majority of R's, D's and I's all agreeing to it.  Even if all 4 of the Independents are far left and they all get on board with some 45D-8R map the Democrats propose, they still can't pass it without support from at least 3 Republicans.  So the commission would just deadlock and the process would go to court. Now the court theoretically could end up adopting the 45D-8R map that looks like MD, but do we really think that's likely?

Furthermore, the interests of far left and far right I's don't necessarily align with those of the D's or R's.  A green or a tea partier is probably more interested in getting the most ideological candidates possible out of the primaries.  So they might actually want a map with lots of R+20 and D+20 districts to elect "true conservatives" or "true liberals."

Now the AZ/NJ system does totally break down with a hard left or hard right I as the tiebreaker, as we arguably saw last year.   

If a plan does not receive 3 votes in each group, then the Supreme Court appoints special masters to "adjust" the plan.   So there is continuous pressure to play nice and accept the advice of the lawyers.  Even the referendum provision ended up being a farce.

So your concern is that there would be a ton of pressure on the opposite party to accept only minor concessions from what the far left indies are pushing (i.e. the D's and I's pass a 45D-8R map and the R's settle for 43D-10R)?  Or is it that the court-appointed special master would have to use the very Democratic map  as a baseline if there was a deadlock on the commission?

[muon2]

Well, it's certainly better than simply giving whichever side has a majority in the state legislature carte blanche to gerrymander to their hearts' content. But, all too often, "non-partisan" redistricting is in practice bipartisan gerrymandering, or (as happened in Arizona and New Jersey this past cycle), partisan gerrymandering where an "independent" sides with either the Republicans or the Democrats. Rather than give any group, partisan or "not," full discretionary power when it comes to redistricting, it would be better to impose strict restraints on the process such as respect for municipal and county boundaries and communities of interest.

The problem with commissions is that they are supposed to be wise amateurs.  This makes them manipulable by lawyers.

In California the selection process was particularly biased towards producing racially gerrymandered maps.   Two of the non-Democratic members were probably not Democrats because the party was too moderate, while the supposition was that those members would be somewhat muddled towards the middle.

Potential members were rejected for saying quite sensible things.  One candidate said they really didn't believe in that race thing, so they guessed they were European-American.  Another had been quite involved in redistricting supervisor districts in their county.  They made the comment that the people didn't understand that the districts had to have the same population, and was said to be insensitive to public input.  You then get the lawyers for the commission explaining what they think the law means, and they end up going along because it sounds reasonable.

I don't actually see what the harm is in having greens or tea partiers as the Independents on a California format commission.  Nothing can pass without a majority of R's, D's and I's all agreeing to it.  Even if all 4 of the Independents are far left and they all get on board with some 45D-8R map the Democrats propose, they still can't pass it without support from at least 3 Republicans.  So the commission would just deadlock and the process would go to court. Now the court theoretically could end up adopting the 45D-8R map that looks like MD, but do we really think that's likely?

Furthermore, the interests of far left and far right I's don't necessarily align with those of the D's or R's.  A green or a tea partier is probably more interested in getting the most ideological candidates possible out of the primaries.  So they might actually want a map with lots of R+20 and D+20 districts to elect "true conservatives" or "true liberals."

Now the AZ/NJ system does totally break down with a hard left or hard right I as the tiebreaker, as we arguably saw last year.   

The problem that I see is that any group will have an inherent bias. Part of this comes from the examples given where self-declared independents are actually quite ideological, but not in favor of a party. Their ideology still translates to their choices for a map. Even if one could know all the possible ideological thoughts of member of a small group, there will be other inherent biases about what make a good district, such as the debate we can see on the merits of the splitline algorithm.

I think that a non-partisan committee can be as biased as a partisan one, but in different ways. The best solution for me is to take the pen out of the hands of a small group and hand it to the people. Let a small group act as the judge, but with no power to draw, and constrain their judgement by criteria defined before the data is available. That creates a double-blind process where the criteria is drafted without knowing how it takes advantage of specific data and the commission acts on third party maps, using only the predefined criteria.

[jimrtex]

The problem that I see is that any group will have an inherent bias. Part of this comes from the examples given where self-declared independents are actually quite ideological, but not in favor of a party. Their ideology still translates to their choices for a map. Even if one could know all the possible ideological thoughts of member of a small group, there will be other inherent biases about what make a good district, such as the debate we can see on the merits of the splitline algorithm.

I think that a non-partisan committee can be as biased as a partisan one, but in different ways. The best solution for me is to take the pen out of the hands of a small group and hand it to the people. Let a small group act as the judge, but with no power to draw, and constrain their judgement by criteria defined before the data is available. That creates a double-blind process where the criteria is drafted without knowing how it takes advantage of specific data and the commission acts on third party maps, using only the predefined criteria.

Why does it have to be a small group acting as a judge?  We let the People choose their elected officials, why can't the People choose the districts from which those officials are elected?

Consider legislative districting, where there are a relatively large number of districts, and we typically have either multiple districts per county, or multiple counties per district, or sometimes combinations of the two.

Some definitions:

quota: The ideal district population.  State population divided by number of districts.

entitlement: The number of districts that a county is entitled to.  County population divided by quota.   Entitlement is a rational number, but is best understood as a mixed fraction, where the integer part represents the whole number of districts a county is entitled to, and the fraction is the part of an additional district that is entitled to.   For example, if a county's entitlement is 4-1/3, the county is entitled to 4 districts, and 1/3 of another.

county apportionment area (CAA): Group of one or more contiguous counties whose total entitlement is approximately equal to an integer number of districts.  In particular, if E is the sum of the entitlement of the counties, then CAA should be apportioned D districts, where
D = floor(E + 1/2) districts.   That is D, is simply the total entitlement rounded to the nearest integer.

And further  | E - D | < sqrt(D) * 0.07

That is, the absolute difference between the entitlement and the number of districts apportioned, which can be considered to be an apportionment error, is less than the square root of the number of districts   This is a blend of concepts.   0.05 is considered a traditional safe harbor, particularly for states that treat districting as primarily a matter of apportionment of districts among counties.   But it can result in systemic error when applied to larger counties.  For example, Tarrant County was entitled to 10.4 house districts in 2000.   With 10 districts, the average error was 4%, which effectively constrained variation within Tarrant County to about 1%.   0.07 is based on the Ohio law that permits single-county districts to have a variation of 10% from the ideal.   I have reduced that to 7%, but extended it to CAA with a single district (where one or more counties will form a district).

To prevent systemic bias for against rural areas, the composite of single-district CAA would also have to apply to the same error calculation.   That is, if there were 23 single-district CAA, their total entitlement would have to be in the range 22.664 to 23.336.   (0.07 * sqrt(23) is 0.3357)

The surplus for a multi-district, multi-county CAA is the sum of the individual county surpluses rounded to the nearest integer.  This represents the minimum number of cross-border districts that would need to be drawn within a CAA.  It is highly desirable that this surplus be one.  It may be necessary in some cases to have a larger surplus - but only as a necessity.  There may be a few instances where the surplus is zero.  This would occur when combining counties with small surpluses so that the collective error is within a permittable range (eg combining a 10.1 county with a 1.1 county).   Single county, or single district CAA are considered to have a surplus of zero.

A valid proposed plan is one in which every county in the state is assigned to a single CAA, and the total number of districts for the CAA is the number of districts for the legislative chamber.  That is, the CAA must cover the entire state, with no overlap.

Each plan would be scored.   It would receive one point for each CAA; an additional point for each single-county CAA; and one point for each single-district CAA.   A single-county, single-district CAA would account for three points.

To the extent consistent with equal protection, we want to apportion districts to single large counties, and form single districts from whole smaller counties.  The scoring system would highly favor this by giving such CAA a bonus, and also encouraging more CAA.

Process:

(1) Interested individuals could submit a valid proposed plan, which would be published and scored.   

Plans with CAA with an unnecessarily large number of CAA with a surplus of two or more would not be considered valid.  If any submitted plan has no CAA with a surplus of two or more, then all plans that have a CAA with a surplus of two or more would be considered invalid.   If all plans have CAA with a surplus of two or more, only plans with the fewest number of such CAA will be considered valid.

(2) Interested individuals could submit a new valid proposed plan.  If they had submitted a plan in the first round, the new plan must score at least as high.  If there was a significant improvement in the high score, or a plan reduced the number of high-surplus CAA, additional rounds could be held.

Note, the object is to produce the best plan for the State, rather than have a contest for plan-drawers.  However, it might be possible for a private foundation to provide prizes for highest scoring plan.  If someone can submit a plan that improves on another plan, that should be encouraged.

(3) The highest scoring plans would be identified.  These would be submitted to a representative sample of voters across the state.  This panel should be large enough to be representative not only of the state, but also of each county, and areas within each county (since weighting would be used, smaller counties could be oversampled).

The panel members would rank each of the highest scoring plans.  In doing so, they would only be shown the portion of the CAA which their county might be placed in.  They are not judging the entire plan, but only its impact on them.

The best composite plan would be chosen by Condorcet.  Then for each CAA in the best composite plan, it would be determined whether they favored the best composite plan.  Those CAA which were in concurrence would go on to the next step, which would define the districts within the CAA.

Those CAA which disagreed with the composite plan would go through another iteration of the process, beginning with step (1).

(4) Districts would be defined within each CAA.  This might be similar to the county process, but using townships.

(4a) In regions of contiguous single-district CAA, there could be an opportunity to rearrange the districts.


 

[muon2]

jimrtex there are some interesting elements to think through here. A quick reaction is that there really are two parts here - one governing what maps go forward for consideration, and one governing the process to select and perhaps amend the map that came forward. One key result of this separation is how that affects submissions, and they should be whole plans in the initial stage.

I think that a whole plan has to be reviewed by all. The reasoning that places bias in a small map drawing body is due to knowledge beyond the specific task. That bias would also be present in a real body voting on plans. With a public submission process it is unavoidable, and in fact desirable, for the maps to be known as they are validated. It would take a sequestered body to maintain knowledge of only part of a map while casting a decision. That would be unworkable for a real state acting with real time periods for submission and selection.

With some more thought I'll react to more of the specific elements.

[muon2]

Some definitions:

quota: The ideal district population.  State population divided by number of districts.

entitlement: The number of districts that a county is entitled to.  County population divided by quota.   Entitlement is a rational number, but is best understood as a mixed fraction, where the integer part represents the whole number of districts a county is entitled to, and the fraction is the part of an additional district that is entitled to.   For example, if a county's entitlement is 4-1/3, the county is entitled to 4 districts, and 1/3 of another.

county apportionment area (CAA): Group of one or more contiguous counties whose total entitlement is approximately equal to an integer number of districts.  In particular, if E is the sum of the entitlement of the counties, then CAA should be apportioned D districts, where
D = floor(E + 1/2) districts.   That is D, is simply the total entitlement rounded to the nearest integer.

And further  | E - D | < sqrt(D) * 0.07

That is, the absolute difference between the entitlement and the number of districts apportioned, which can be considered to be an apportionment error, is less than the square root of the number of districts   This is a blend of concepts.   0.05 is considered a traditional safe harbor, particularly for states that treat districting as primarily a matter of apportionment of districts among counties.   But it can result in systemic error when applied to larger counties.  For example, Tarrant County was entitled to 10.4 house districts in 2000.   With 10 districts, the average error was 4%, which effectively constrained variation within Tarrant County to about 1%.   0.07 is based on the Ohio law that permits single-county districts to have a variation of 10% from the ideal.   I have reduced that to 7%, but extended it to CAA with a single district (where one or more counties will form a district).

To prevent systemic bias for against rural areas, the composite of single-district CAA would also have to apply to the same error calculation.   That is, if there were 23 single-district CAA, their total entitlement would have to be in the range 22.664 to 23.336.   (0.07 * sqrt(23) is 0.3357)
 

I like the idea of the square root of the number of districts in a CAA times a base number as a measure of the variance permitted. For state legislative districts I think that number has to be 0.05 and for congressional districts it has to be 0.005. In some cases the state legislative number may have to be smaller as well, for example in IL the court has held that 0.5% is the maximum allowable population variance for legislative districts under the constitution. As recently as in the WV case, SCOTUS noted that the use of the 1% maximum range for congressional districts was the point where the burden of proof shifts from the state to the plaintiff. That seems like a strong reason to use 0.005 for congressional districts and 0.05 for other districts except where state law is more restrictive. As to the OH example, the 10% variance is closely tied to the creation of single county districts, and I would not be comfortable with the burden of proof for all parts of a plan using that standard.

I think that some weight also must be given to the population range between CAAs. In the WV case SCOTUS noted that the plaintiff had to show that they could achieve the same goals with a lower population range. From that I would gather that if two plans with the same score based on CAAs differed in range, the court would require the use of the smaller range plan if there was a challenge. One could use the county equivalence table that was empirically derived in the IA style redistricting series to score the population range, here extended shifted to accommodate a 10% maximum range.

definition - range: the difference in population between the largest and smallest population districts expressed as a percentage of the quota.

0: 10%
1: 8%
2: 6%
3: 5%
4: 4%
5: 3%
6: 2.5%
7: 2%
8: 1.5%
9: 1.2%
10: 1.0%
11: 0.8%
12: 0.6%
13: 0.5%
14: 0.4%
15: 0.3%
16: 0.25%
17: 0.2%
18: 0.15%
19: 0.12%
20: 0.10%
21: 0.08%
22: 0.06%
23: 0.05%
24: 0.04%
25: 0.03%
26: 0.025%
27: 0.02%
28: 0.015%
29: 0.012%
30: 0.010%

Using this a plan is improved if either its CAA score or range score increases.

[jimrtex]

jimrtex there are some interesting elements to think through here. A quick reaction is that there really are two parts here - one governing what maps go forward for consideration, and one governing the process to select and perhaps amend the map that came forward. One key result of this separation is how that affects submissions, and they should be whole plans in the initial stage.

I think that a whole plan has to be reviewed by all. The reasoning that places bias in a small map drawing body is due to knowledge beyond the specific task. That bias would also be present in a real body voting on plans. With a public submission process it is unavoidable, and in fact desirable, for the maps to be known as they are validated. It would take a sequestered body to maintain knowledge of only part of a map while casting a decision. That would be unworkable for a real state acting with real time periods for submission and selection.

With some more thought I'll react to more of the specific elements.

Imagine that the federal House of Representatives could have districts apportioned among states, but with limitations on how states were combined.

This is quite similar to how apportionment for the Texas House of Representatives works, and implicitly how it should work for Ohio, but doesn't.

For example, instead of Washington with an entitlement of 9.501 districts having 10, and Oregon with an entitlement of 5.415, having 5 districts, Washington+Oregon with an entitlement of 14.916 would have 15 districts.

Based on a House size of 435, 20 individual states are within bounds for apportionment to a single state only; while 30 states would be required to be grouped with one or more additional states.   It is somewhat likely that a few of the states that are within bounds will have to be combined anyhow.

It happens that Hawaii (1.923) and Alaska (1.015) would not have to be paired with other other states.  If a plan proposed pairing Hawaii and Alaska, for a slightly smaller error, it would score badly, going from 5 points: 2 SAA, 2 single-state SAA, and 1 single-district SAA to 1 point for 1 SAA, which would have multiple states (2) and multiple districts (3).

So why should the people of Alaska and Hawaii have any say in the rest of the plan?  From their perspective, they are all the same.  The people of Hawaii will still have the opportunity to devise the two districts in their state.

It is quite possible that Washington and Oregon would be paired in all plans.  So how each plan treats the rest of the country is immaterial, and they will soon have a chance to decide whether the cross-border seat is at Vancouver-Portland or in the east.

But imagine that some plans paired Washington-Idaho and Oregon-Nevada; while others paired and Washington-Oregon and Idaho-Nevada.   Then Washingtonians would weigh in on whether they preferred to be paired with Oregon or Idaho.  Even in this case, it is pretty marginal whether when considering that Washington-Idaho would force Oregon-Nevada that they would pay attention to the impact on Oregon and Nevada.  But what if two plans had a Washington-Oregon pairing, but differed in how they treated New England.  To someone in Washington, Washington, the plans are the same.

[jimrtex]

county apportionment area (CAA): Group of one or more contiguous counties whose total entitlement is approximately equal to an integer number of districts.  In particular, if E is the sum of the entitlement of the counties, then CAA should be apportioned D districts, where
D = floor(E + 1/2) districts.   That is D, is simply the total entitlement rounded to the nearest integer.

And further  | E - D | < sqrt(D) * 0.07

That is, the absolute difference between the entitlement and the number of districts apportioned, which can be considered to be an apportionment error, is less than the square root of the number of districts   This is a blend of concepts.   0.05 is considered a traditional safe harbor, particularly for states that treat districting as primarily a matter of apportionment of districts among counties.   But it can result in systemic error when applied to larger counties.  For example, Tarrant County was entitled to 10.4 house districts in 2000.   With 10 districts, the average error was 4%, which effectively constrained variation within Tarrant County to about 1%.   0.07 is based on the Ohio law that permits single-county districts to have a variation of 10% from the ideal.   I have reduced that to 7%, but extended it to CAA with a single district (where one or more counties will form a district).

To prevent systemic bias for against rural areas, the composite of single-district CAA would also have to apply to the same error calculation.   That is, if there were 23 single-district CAA, their total entitlement would have to be in the range 22.664 to 23.336.   (0.07 * sqrt(23) is 0.3357)

I like the idea of the square root of the number of districts in a CAA times a base number as a measure of the variance permitted. For state legislative districts I think that number has to be 0.05 and for congressional districts it has to be 0.005. In some cases the state legislative number may have to be smaller as well, for example in IL the court has held that 0.5% is the maximum allowable population variance for legislative districts under the constitution. As recently as in the WV case, SCOTUS noted that the use of the 1% maximum range for congressional districts was the point where the burden of proof shifts from the state to the plaintiff. That seems like a strong reason to use 0.005 for congressional districts and 0.05 for other districts except where state law is more restrictive. As to the OH example, the 10% variance is closely tied to the creation of single county districts, and I would not be comfortable with the burden of proof for all parts of a plan using that standard.

The 0.05 assumed an apportionment model, or a blended apportionment-redistricting model.

The Texas Constitution specifies an apportionment model.   There have been enough court decisions that it is now more of a blended model.   But those decisions also blocked efforts to simply ignore the constitution and create equal population districts without regard to counties.

The Illinois Constitution says: contiguous, compact, and substantially equal in population.

I don't know of any states where an apportionment model would work for congressional districts (the post about apportioning US representatives across State lines was intended as a thought experiment).   The 0.07 was in part intended to make the range for larger magnitudes a little more flexible.  If a 0.05 was considered to be a maximum acceptable range, I would use:

min (0.07*sqrt(N), 0.05).   That is, the single district case would be acceptional.

Ohio's special case for single county, single member districts may be a special case derived from its previous reapportionment provisions.   Ohio at one time had strict whole county apportionment, but with fractional apportionment, to the nearest 1/5 of a representative.  A county would then elect an extra representative in that many 1/5 of the 5 terms over a decade between censuses.   A county entitled to 1.4 members, would elect 2 members in 2 out of 5 terms.  But there were a lot of fudging for small numbers.  For some reason, they decided 1.8 should always be rounded to 2, but 2.8 was OK.  So the special case for counties with between 0.9 and 1.1 may have been derived from trying to preserve part of the old model.

You must be logged in to read this quote.

The Texas Constitution says:

"The members of the House of Representatives shall be apportioned among the several counties, according to the number of population in each, as nearly as may be, on a ratio obtained by dividing the population of the State, as ascertained by the most recent United States census, by the number of members of which the House is composed;"

That is to say, what is important is the difference between each counties entitlement and the statewide quota.

[jimrtex]

Imagine that the federal House of Representatives could have districts apportioned among states, but with limitations on how states were combined.

This map scores 49:

31 State Apportionment Areas (SAA)
17 Single-State SAA
1 Single-District SAA

Excessive surplus 3 (NY-NJ-ME-NH-VT); 2 (MN-MT-ND)

4 states that were close enough to be single-state SAA, were used in multi-state SAA (NY, PA, CT, ND).

Overall region shifts, Florida+Georgia lost 1 to Missouri+Oklahoma.  This will mainly help Missouri get its 8th district back after adding in a bit of northeastern Oklahoma.

[muon2]

county apportionment area (CAA): Group of one or more contiguous counties whose total entitlement is approximately equal to an integer number of districts.  In particular, if E is the sum of the entitlement of the counties, then CAA should be apportioned D districts, where
D = floor(E + 1/2) districts.   That is D, is simply the total entitlement rounded to the nearest integer.

And further  | E - D | < sqrt(D) * 0.07

That is, the absolute difference between the entitlement and the number of districts apportioned, which can be considered to be an apportionment error, is less than the square root of the number of districts   This is a blend of concepts.   0.05 is considered a traditional safe harbor, particularly for states that treat districting as primarily a matter of apportionment of districts among counties.   But it can result in systemic error when applied to larger counties.  For example, Tarrant County was entitled to 10.4 house districts in 2000.   With 10 districts, the average error was 4%, which effectively constrained variation within Tarrant County to about 1%.   0.07 is based on the Ohio law that permits single-county districts to have a variation of 10% from the ideal.   I have reduced that to 7%, but extended it to CAA with a single district (where one or more counties will form a district).

To prevent systemic bias for against rural areas, the composite of single-district CAA would also have to apply to the same error calculation.   That is, if there were 23 single-district CAA, their total entitlement would have to be in the range 22.664 to 23.336.   (0.07 * sqrt(23) is 0.3357)

I like the idea of the square root of the number of districts in a CAA times a base number as a measure of the variance permitted. For state legislative districts I think that number has to be 0.05 and for congressional districts it has to be 0.005. In some cases the state legislative number may have to be smaller as well, for example in IL the court has held that 0.5% is the maximum allowable population variance for legislative districts under the constitution. As recently as in the WV case, SCOTUS noted that the use of the 1% maximum range for congressional districts was the point where the burden of proof shifts from the state to the plaintiff. That seems like a strong reason to use 0.005 for congressional districts and 0.05 for other districts except where state law is more restrictive. As to the OH example, the 10% variance is closely tied to the creation of single county districts, and I would not be comfortable with the burden of proof for all parts of a plan using that standard.

The 0.05 assumed an apportionment model, or a blended apportionment-redistricting model.

The Texas Constitution specifies an apportionment model.   There have been enough court decisions that it is now more of a blended model.   But those decisions also blocked efforts to simply ignore the constitution and create equal population districts without regard to counties.

The Illinois Constitution says: contiguous, compact, and substantially equal in population.

I don't know of any states where an apportionment model would work for congressional districts (the post about apportioning US representatives across State lines was intended as a thought experiment).   The 0.07 was in part intended to make the range for larger magnitudes a little more flexible.  If a 0.05 was considered to be a maximum acceptable range, I would use:

min (0.07*sqrt(N), 0.05).   That is, the single district case would be acceptional.

Ohio's special case for single county, single member districts may be a special case derived from its previous reapportionment provisions.   Ohio at one time had strict whole county apportionment, but with fractional apportionment, to the nearest 1/5 of a representative.  A county would then elect an extra representative in that many 1/5 of the 5 terms over a decade between censuses.   A county entitled to 1.4 members, would elect 2 members in 2 out of 5 terms.  But there were a lot of fudging for small numbers.  For some reason, they decided 1.8 should always be rounded to 2, but 2.8 was OK.  So the special case for counties with between 0.9 and 1.1 may have been derived from trying to preserve part of the old model.

You must be logged in to read this quote.

The Texas Constitution says:

"The members of the House of Representatives shall be apportioned among the several counties, according to the number of population in each, as nearly as may be, on a ratio obtained by dividing the population of the State, as ascertained by the most recent United States census, by the number of members of which the House is composed;"

That is to say, what is important is the difference between each counties entitlement and the statewide quota.

The IL constitution has been generally granted wide latitude due its vague redistricting language. However the state supremes set 0.5% as the maximum deviation consistent with the phrase substantially equal for legislative districts.

Besides TX is any other state still based on an apportionment model as opposed to a pure redistricting (sometimes with rules)? I'm more interested on your take of the map selection criteria assuming that for most states it's just about redistricting.

[jimrtex]

The 0.05 assumed an apportionment model, or a blended apportionment-redistricting model.

The Texas Constitution specifies an apportionment model.   There have been enough court decisions that it is now more of a blended model.   But those decisions also blocked efforts to simply ignore the constitution and create equal population districts without regard to counties.

The Illinois Constitution says: contiguous, compact, and substantially equal in population.

The Texas Constitution says:

"The members of the House of Representatives shall be apportioned among the several counties, according to the number of population in each, as nearly as may be, on a ratio obtained by dividing the population of the State, as ascertained by the most recent United States census, by the number of members of which the House is composed;"

That is to say, what is important is the difference between each counties entitlement and the statewide quota.

The IL constitution has been generally granted wide latitude due its vague redistricting language. However the state supremes set 0.5% as the maximum deviation consistent with the phrase substantially equal for legislative districts.

Besides TX is any other state still based on an apportionment model as opposed to a pure redistricting (sometimes with rules)? I'm more interested on your take of the map selection criteria assuming that for most states it's just about redistricting.

Hawaii requires apportionment among the basis island units, forbidding canoe districts.

Ohio is an apportionment model.   Oklahoma is totally apportionment, but may so badly in violation of equal protection that it is ignored.

The Iowa Constitution does not really have anything more substantive in its constitution than Illinois, yet has established standards by statute.  I suspect that they could have set standards such as I have proposed back in 1971 and perhaps had a court approve it.  Iowa actually has two standards for population equality.  They require a mean absolute deviation less than 1%, and a maximum deviation less than 5%.   But it then requires the legislature to justify deviation of more than 1%.  It appears that the Legislative Services Agency believes it is impossible for the legislature to justify such deviation - and essentially says that the legislature can comply with the standard itself set.

Why couldn't Iowa statute set the maximum deviation to 5%, and require apportionment of districts among counties?

[muon2]

The 0.05 assumed an apportionment model, or a blended apportionment-redistricting model.

The Texas Constitution specifies an apportionment model.   There have been enough court decisions that it is now more of a blended model.   But those decisions also blocked efforts to simply ignore the constitution and create equal population districts without regard to counties.

The Illinois Constitution says: contiguous, compact, and substantially equal in population.

The Texas Constitution says:

"The members of the House of Representatives shall be apportioned among the several counties, according to the number of population in each, as nearly as may be, on a ratio obtained by dividing the population of the State, as ascertained by the most recent United States census, by the number of members of which the House is composed;"

That is to say, what is important is the difference between each counties entitlement and the statewide quota.

The IL constitution has been generally granted wide latitude due its vague redistricting language. However the state supremes set 0.5% as the maximum deviation consistent with the phrase substantially equal for legislative districts.

Besides TX is any other state still based on an apportionment model as opposed to a pure redistricting (sometimes with rules)? I'm more interested on your take of the map selection criteria assuming that for most states it's just about redistricting.

Hawaii requires apportionment among the basis island units, forbidding canoe districts.

Ohio is an apportionment model.   Oklahoma is totally apportionment, but may so badly in violation of equal protection that it is ignored.

The Iowa Constitution does not really have anything more substantive in its constitution than Illinois, yet has established standards by statute.  I suspect that they could have set standards such as I have proposed back in 1971 and perhaps had a court approve it.  Iowa actually has two standards for population equality.  They require a mean absolute deviation less than 1%, and a maximum deviation less than 5%.   But it then requires the legislature to justify deviation of more than 1%.  It appears that the Legislative Services Agency believes it is impossible for the legislature to justify such deviation - and essentially says that the legislature can comply with the standard itself set.

Why couldn't Iowa statute set the maximum deviation to 5%, and require apportionment of districts among counties?

Maybe I'm confused on an apportionment model. OH calls the process apportionment, but only applies some principles of apportionment to large counties suitable for one or more whole districts. Pure apportionment would require joining small counties together to make whole districts as well, and OH lacks that element. The result is that small counties get carved up in the legislature. IA could be about apportionment, but their statute is really about criteria for drawing districts. IL may be limited from doing the same as IA because the constitution has no enabling clause in the redistricting section that would permit the legislature to enact supporting statute.

I think what I'm seeking is a first step based on true apportionment, which is also what you describe. I however think that in order to comply with the courts in most states I've seen, a plan has to proceed to draw districts within each apportioned region as a second step before any judging or adjustments. Considering a plan as a whole shows up in a number of court decisions, so it seems that a complete plan should be a requirement of a valid map. That doesn't stop criteria involving the apportionment step, but does require that both steps be present for a plan to be realistically implemented.

I also think that an apportionment first step can work for congressional districts in a state as well. They just need changes to population equality to meet court requirements. Even in legislative maps, some states have also viewed their population equality clauses to favor plans with more precise equality when all other factors are equal (IA and IL are both in this category).

[muon2]

Here's how I would revise the definitions to reflect what I think is an apportionment-based first step for redistricting. I've allowed for state-specific flexibility for the maximum population deviation so that it should work for both congressional and legislative plans. I've also incorporated contiguity definitions from the erosity thread and provided state-specific flexibility for the definition of links.


quota {Q}: The ideal district population.  State population divided by number of districts.

range: The difference between the population of the largest and smallest district in a plan expressed as a percentage of the quota.

maximum deviation {D}: The maximum absolute percentage difference permitted for a district from the quota. This is set by state or federal law, and may be set at half the maximum permitted range if range is the appropriate legal standard.

population node: The most significant population center in a county. If there is a census defined urban area in the county then the municipality with the highest population in the highest population urban area is the node. If there is no urban area in a county then the county seat is the node. If there is neither an urban area nor a county seat then the node is the municipality with the greatest population.

link: A path between directly connected nodes. Two nodes form a link if there is a continuous path between those two nodes that uses only non-seasonal numbered state and federal highways or non-seasonal regular ferry service that does not pass through any county other than the counties of the two linked nodes. A state may add or remove links to reflect unique geographical circumstances.

contiguous: Two counties are contiguous if there is a link between the population nodes of the counties.

region: A single county or group of contiguous counties. The region population {P} is the sum of the population of the counties in the region.

entitlement: The number of districts within a region. The region population divided by the quota, rounded to the nearest whole integer { n = round(P/Q) }.

apportionment deviation: The population deviation for a region. The absolute difference between the region population and the product of the entitlement of the region times the quota. { d = | P - nQ | }

apportionment region: A group of one or more contiguous counties whose total population is nearly equal to an integer number of districts. Nearly equal means the absolute difference between the total population of the region and the product of the entitlement of the region times the quota is less than the maximum deviation times the square root of the entitlement of the region. { d < sqrt(n)*D }

[Torie]

Who draws the districts in your perfect world Muon2?  Anybody can draw them and submit them to a "small group," and the "small group" then picks one the comports with the law?  I understand the idea for tight rules to leash discretion to only that which to disallow creates more problems than it solves, but I am bit confused by your description of the implementation process.

[jimrtex]

Maybe I'm confused on an apportionment model. OH calls the process apportionment, but only applies some principles of apportionment to large counties suitable for one or more whole districts. Pure apportionment would require joining small counties together to make whole districts as well, and OH lacks that element. The result is that small counties get carved up in the legislature. IA could be about apportionment, but their statute is really about criteria for drawing districts. IL may be limited from doing the same as IA because the constitution has no enabling clause in the redistricting section that would permit the legislature to enact supporting statute.

In Ohio it is based in part in its history before the OMOV decisions.   When its current constitution was adopted it provided that legislative districts be semi-permanent, and based on single county or multi-county districts.  Instead of a whole number of representatives, it would be apportioned to the nearest 1/5 of a legislator.  The fractional representation would be provided by electing a member that share of 2-year terms in a decade.

Around 1900, Mark Hanna got a constitutional amendment passed that ensured that any county would have at least one representative (previously they would have been combined into a district electing 1.0 or more representatives).   They were also elected at large.  This led to an equal protection as well as minority voting rights challenge, and the current version of Section 11 was implemented.  I don't know if the fractional representation has ever been challenged or not.

I think the revisers thought they were creating an apportionment model consistent with equal protection.

§ 11.07(a) says that "to the extent consisted with Section 3 [the 95% - 105% range]" districts shall be composed of whole counties.  It was entirely consistent with the 95% - 105% range to draw bunches of whole county districts, which were not drawn.

§ 11.07(b) says that when the creation of whole county districts "cannot feasibly be attained" that then whole townships should be used.

I would interpret § 11.08 as implying that the surplus of a multi-county district should be treated as a whole county for creation of a district - the same way that it is treated in Texas (under the current interpretation of the constitution).

In Ohio, they apparently are flexible in the meaning of "feasible" and "consistent"

I think that Iowa could replace their Section 42.  The constitution itself only says that 40% of the population can not elect 50% of the senators.

You must be logged in to read this quote.

I don't think that the iterative steps need to be in the form of a final product.  That would be the equivalent of someone filing a judicial challenge to a filed bill, or a bill that has passed first or second reading.

You must be logged in to read this quote.

As a gross first step, maybe.

Texas has regional planning organizations, that are multi-county - I think these are required by the federal government.  You could let the commissioners courts decided if they wanted to switch regions for congressional representation process, and then go through a process of combining areas until they are large enough that an apportionment can be made.

And then the areas can be refined by switching counties to get to a closer apportionment, followed by a final districting.

[jimrtex]

Here's how I would revise the definitions to reflect what I think is an apportionment-based first step for redistricting. I've allowed for state-specific flexibility for the maximum population deviation so that it should work for both congressional and legislative plans. I've also incorporated contiguity definitions from the erosity thread and provided state-specific flexibility for the definition of links.

You must be logged in to read this quote.

I'd avoid redefining contiguous from a conventional meaning, unless you want to exclude point contiguity or near-point contiguity.

Instead, I'd define "connected" as based on some relationship between nodes in contiguous counties.

I think we may have gotten sidetracked here using transportation connectivity for different reasons.  I think we want an absolute prohibition on districts directly connecting Whatcom and Okanogan counties, and how we are treating the relationships between Garfield, Asotin, and Whitman counties.

You can drive from Pullman to Pomeroy  on numbered roads, but Google and Mapquest will tell you to drive through Lewiston and Clarkston.

How did you use erosity in Iowa?  Was it just so plans weren't penalized for corner cuts?

Maybe there needs to be be three relationships:

Contiguous: touching.

Connected: easy to travel between.   Whatcom is not connected to Okanogan.  Districts may not include unconnected counties without going through other counties.   An outline map would show a gap between the counties, as if it were outside the limits of the state.

Linked: A more significant relationship, the cutting of which increases the erosity count.

So you could include connected counties within a district, but would not reduce erosity.

You must be logged in to read this quote.

You've mixed units.   D is a percentage, but you've calculated d as the deviation in terms of population.   I'd use relative deviation for a region.  { d = | P/Q - n | }

[muon2]

Who draws the districts in your perfect world Muon2?  Anybody can draw them and submit them to a "small group," and the "small group" then picks one the comports with the law?  I understand the idea for tight rules to leash discretion to only that which to disallow creates more problems than it solves, but I am bit confused by your description of the implementation process.

In my perfect world, anyone draws the districts. Instead of the perhaps futile attempt to eliminate bias, I freely admit to the bias of the mapmakers but give them plenty of constraints. I haven't yet delved into my preferred process yet, though jimrtex clearly has.

For me there are two bodies, a group to oversee the submissions and determine which set meets the necessary criteria and a group that chooses from among the cleared maps. With solid constraints I'm rather neutral about how those bodies should be composed, or whether they should be independent of each other. I would expect that the oversight body should be small so as to be able to function in a timely manner, but the deciding body could be as small as a commission or as large as a legislature to get a diverse set of decision makers. In any case there should be a well-defined method to select a map if the deciding body cannot reach a timely choice.

[muon2]

Here's how I would revise the definitions to reflect what I think is an apportionment-based first step for redistricting. I've allowed for state-specific flexibility for the maximum population deviation so that it should work for both congressional and legislative plans. I've also incorporated contiguity definitions from the erosity thread and provided state-specific flexibility for the definition of links.

You must be logged in to read this quote.

I'd avoid redefining contiguous from a conventional meaning, unless you want to exclude point contiguity or near-point contiguity.

Instead, I'd define "connected" as based on some relationship between nodes in contiguous counties.

I think we may have gotten sidetracked here using transportation connectivity for different reasons.  I think we want an absolute prohibition on districts directly connecting Whatcom and Okanogan counties, and how we are treating the relationships between Garfield, Asotin, and Whitman counties.

You can drive from Pullman to Pomeroy  on numbered roads, but Google and Mapquest will tell you to drive through Lewiston and Clarkston.

How did you use erosity in Iowa?  Was it just so plans weren't penalized for corner cuts?

Maybe there needs to be be three relationships:

Contiguous: touching.

Connected: easy to travel between.   Whatcom is not connected to Okanogan.  Districts may not include unconnected counties without going through other counties.   An outline map would show a gap between the counties, as if it were outside the limits of the state.

Linked: A more significant relationship, the cutting of which increases the erosity count.

So you could include connected counties within a district, but would not reduce erosity.

You must be logged in to read this quote.

You've mixed units.   D is a percentage, but you've calculated d as the deviation in terms of population.   I'd use relative deviation for a region.  { d = | P/Q - n | }

Thanks for the math catch. I'll fix that in the next edit. I appreciate your concern about contiguous, and it's clear than linked counties are also contiguous, so there would be no violation of any constitutional language to restrict regions to linked counties.

How does this look:

link: Two nodes form a link if there is a continuous path between those two nodes that uses only non-seasonal numbered state and federal highways or non-seasonal regular ferry service that does not pass through any county other than the counties of the two linked nodes. A state may add or remove links to reflect unique geographical circumstances. (were you suggesting to drop this?)

connected: Two counties are connected if there is a link between the population nodes of the counties. Connected counties are by definition contiguous.

region: A single county or group of connected counties. The region population {P} is the sum of the population of the counties in the region.