How Feasible Is Creating a Gerrymander-Free District-Mapping Program?

[Feeble Wonk]

I've got a challenge for you programming guys, if this wouldn't be more appropriate for a different forum. In the current environment of pathologically polarized politics, an obvious positive move to address this problem would be to undo the congressional redistricting through gerrymandering that fortifies the polarization.
So, my question is this: How difficult (or easy) would it be to develop a program that would have the opposite effect... establishing regional congressional districts of equal populations that "balance" the ideological poles toward the middle?

 

[jedishrfu]

I wouldn't pose this as a challenge. If you are trying to do it then you should show us what you've done. This is a difficult problem to solve.

Here's a map of electoral voting for states and the power that some states have over others:

http://www.slate.com/articles/news_..._showing_the_vote_power_of_all_50_states.html

 

[Feeble Wonk]

I wouldn't pose this as a challenge. If you are trying to do it then you should show us what you've done. This is a difficult problem to solve.

Here's a map of electoral voting for states and the power that some states have over others:

http://www.slate.com/articles/news_..._showing_the_vote_power_of_all_50_states.html

Thanks for the feedback Jedishrfu. Sadly, this effort is well beyond my skill set.
My question is not really about the electoral college per se, but rather a redistricting of existing congressional districts within the individual states. As I'm sure your aware, political strategy to date has been for sitting congress people to attempt to redraw the congressional district maps to facilitate their re-election. This effort tends to produce voting populations within those gerrymandered districts that become increasingly ideological, with the net effect being magnified polarization of the elected representatives toward the political polar extremes.
So, I was wondering how difficult it would be for skilled programmers to develop a system that would redraw the congressional district borders that would maximize ideological balance toward the middle. The congressional districts are, by definition, a reflection of electoral population within a specific state, so that population parameter would have to be maintained.

 

[anorlunda]

maximize ideological balance toward the middle.

Like it or not, that too is a form of political bias.

The neutral way to do is is to draw district boundaries such that each district has the same population and such that the sums of the lengths of the perimeters is minimized. Look to mathematicians to ask whether such a solution is unique, or if there are many such solutions.

Neutrality is a result of being purposely blind to the beliefs of the people in the district. The problem is that hardly anybody, left, right or center really wants a neutral redistricting. We, like you, want to bias it to some belief-based criterion or another.

 

[jedishrfu]

Heres another article on it:

https://www.quantamagazine.org/20170404-gerrymandering-math-standard/

 

[jack action]

What would be the "objective" criteria for the computer program? I think the programmer's opinion will influence those criteria, voluntarily or not.

 

[jedishrfu]

As a longtime programmer, we have unlimited power. We control the horizontal, we control the vertical. Do not attempt to stop readiing this post otherwise...

Time to go to school Jedi. Okay Mom!

 

[anorlunda]

What would be the "objective" criteria for the computer program? I think the programmer's opinion will influence those criteria, voluntarily or not.

From #

The neutral way to do is is to draw district boundaries such that each district has the same population and such that the sums of the lengths of the perimeters is minimized.

 

[Feeble Wonk]

Like it or not, that too is a form of political bias.

I absolutely agree. The system I am asking about would definitely be a biased effort to empower more moderate "centrist" elected officials. But, that said, as an academic question... how difficult would it be to do?

The neutral way to do is is to draw district boundaries such that each district has the same population and such that the sums of the lengths of the perimeters is minimized. Look to mathematicians to ask whether such a solution is unique, or if there are many such solutions.

Again, I fully agree that this might be the most politically "neutral" solution, in that ideology would be irrelevant to the process. I suspect, as you questioned, that there might be many such solutions, and the choice of those solutions might be politicized somewhat. However, that discussion takes us toward a philosophical debate.
I'm really just asking, as a matter of academics... How difficult would it be to formulate a program that would automatically, without human intervention, create congressional district borders utilizing the same population perimeters currently established that would maximize ideological balance?

 

[jedishrfu]

You could phrase it so that most constituents have minimal drive times to get to the elected officials primary office with some accepted maximal drive time.

 

[anorlunda]

maximize ideological balance

For starters, you need a psych report on each citizen to determine their "real" ideological leaning. People lie, even to themselves.

People also shift their views day to day, so you would have to determine the district boundaries on election day.

I ask to imagine how many other people before you have tried to imagine a more perfect political system.

 

[anorlunda]

There are also other imperfections more significant than districts. This wiki article does a pretty good job at pointing out how hard it is to design a fair voting system.

https://en.wikipedia.org/wiki/Electoral_system

 

[jack action]

@anorlunda : The problem is not the size of the grid or its population, it is the "quality" of the population. How is the ideological balance within each district? What portion of the district population actually vote?

create congressional district borders utilizing the same population perimeters currently established that would maximize ideological balance?

What does that mean? What is the "best" ideological balance? Isn't that what voting is for? To determine what is the ideological balance.

I think the problem is this system where you vote for someone who will vote for you.

 

[jedishrfu]

The Quanta article gets into the design issues and supreme court criteria.

 

[Aufbauwerk 2045]

I won't address your idea that there is a "pathological polarization." This assumes polarization is a bad thing. That's a political viewpoint, and the admins have very wisely said we should not discuss politics.

But let's look at this as a legitimate math problem. Can boundaries be drawn so that each district has as close to a 50/50 mix of Republicans vs Democrats as possible?

I think it would be impractical, due to the way party affiliations are distributed geographically. How would you balance congressional districts in California, which votes heavily Democrat, or in West Virginia and Oklahoma, which votes heavily Republican?

Besides, we do not have an overall 50/50 split. The popular vote in the last presidential election was for the Democrat. If demographic trends continue, the next presidential election may have an even larger popular vote advantage for the Democrat. So should every district reflect this by having a mix that favors Democrats? Even if that were possible, which it is not due to the localization of party preference already mentioned, it would mean every district would have a slight Democrat majority. I suppose the Democrats would like this, but the Republicans would not like it, since it would mean 100% of the House of Representatives would be Democrat.

 

[jack action]

Can boundaries be drawn so that each district has as close to a 50/50 mix of Republicans vs Democrats as possible?

But how can we even say that 50/50 is an ideal ratio? Any ratio is valid, as long as it is the will of the people. In fact the best ratio is 100/0, since it would mean that everyone's happy because there is consensus and everyone's candidate would have been elected.

 

[Aufbauwerk 2045]

But how can we even say that 50/50 is an ideal ratio? Any ratio is valid, as long as it is the will of the people. In fact the best ratio is 100/0, since it would mean that everyone's happy because there is consensus and everyone's candidate would have been elected.

The question of what is ideal is a political question. The math problem posed by the OP, as I understand it, is how to balance congressional districts so they are as close to a 50/50 mix as possible, in order to encourage middle of the road candidates.

BTW another obstacle to solving the original problem is that congressional districts are supposed to be limited in population size, and to cover a contiguous area. The point is that the congressman or congresswoman should serve a manageable population size in a specific geographical area. The OP problem is not solvable, given these constraints.

 

[Aufbauwerk 2045]

As a longtime programmer, we have unlimited power. We control the horizontal, we control the vertical. Do not attempt to stop readiing this post otherwise...

Time to go to school Jedi. Okay Mom!

Suppose the programmers could write a program to run the world in an optimal fashion, based on operations research and cybernetics. The lead programmers/rulers could sit in a control room, complete with Star-Trek style chairs similar to those on the Enterprise. In effect the whole planet could be run from one master computer center.

"We believe in a way of life that is logical and beneficial." -- Mr. Spock, speaking to his mother.

https://en.wikipedia.org/wiki/Project_Cybersyn

 

[Feeble Wonk]

But how can we even say that 50/50 is an ideal ratio? Any ratio is valid, as long as it is the will of the people. In fact the best ratio is 100/0, since it would mean that everyone's happy because there is consensus and everyone's candidate would have been elected.

Again, in a good faith effort to avoid philosophical value judgements, I'd like to avoid discussion of what would be "ideal". As a matter of academic construct, I have stipulated the program objective to be formulation of an ideologically balanced congressional district, to the degree possible.
However, some other posters have raised the legitimate concern that the concept of ideological status is too ambiguous and fluid to really serve as an effective criteria. It would seem to be more constructive to select more objective demographic categories that would reasonably be assumed to correlate to political ideology. For example, one might use demographics such as official party affiliation, race, income, etc.

 

[Feeble Wonk]

...another obstacle to solving the original problem is that congressional districts are supposed to be limited in population size, and to cover a contiguous area. The point is that the congressman or congresswoman should serve a manageable population size in a specific geographical area. The OP problem is not solvable, given these constraints.

I completely recognize that this would be a fundamental factor in the formation of the confessional district. The size and locations of districts would be largely defined by the distribution of the populations in question. Within any given state, the number of districts (and congressional representatives) are determined by the population of that state. As a starting point, it would seem appropriate to break the state down into geographically contiguous regions that have roughly equal populations that normalizes geographic area as much as possible (part of the program algorithm). Then allow some agreed upon area of potential geographic overlap between the these intrastate regions (consistently maintained as a conventional standard) that would be subject to boundary modification to maximize the demographic diversity while still providing population equity. In this way, the "general" size and location of congressional districts would be determined by population distribution within a given state. Yet, the precise boundaries of individual districts would be "tweaked" by the program formula, again, to maximize demographic diversity.
Fluidity of demographic information would be impossible to avoid altogether. However, the distribution of demographic information could be periodically adjusted at an agreed upon frequency... perhaps after every presidential election, or maybe after each federal census.
A 50/50 distribution of republican/democrat electoral districts would obviously not be achieved, because it would be impossible in most states due to prevailing political party majorities. However, the resultant congressional district boundaries would do away with the current partisan gerrymandered constructions, and be moderated to the mainstream as much as possible while still respecting state sovereignty at the same time that it facilitates more generalized ideological representation of the electorate "as a whole", within any given state.
So... within these constraints, would such a program algorithm, that might act as a federal standard, be difficult to produce?

 

[RogueOne]

I absolutely agree. The system I am asking about would definitely be a biased effort to empower more moderate "centrist" elected officials. But, that said, as an academic question... how difficult would it be to do?
.
.
.
... How difficult would it be to formulate a program that would automatically, without human intervention, create congressional district borders utilizing the same population perimeters currently established that would maximize ideological balance?

If the program's parameters are determined by humans, it will be inherently impossible to do that. You also would need an unbiased way to define "moderate 'centrist' elected officials", which is not measurable objectively.
You'd also first need to determine whether or not politics is "pathologically polarized", and whether or not polarization is actually undesirable. What separates our system of government from a monarchy is the fact that we have opposing viewpoints as well as other checks and balances. This entire idea is fundamentally disqualified from being considered objective.

 

[jack action]

So... within these constraints, would such a program algorithm, that might act as a federal standard, be difficult to produce?

Maybe I'm not following clearly what you are talking about, but I cannot see how you can define a set of rules, a standard, better than it is right now.

To avoid US politics, I will talk about such a problem that happened recently in my province. The electoral map was revised and in its first draft, one district was supposed to be eliminated and merged between 3 other adjacent districts (click on "Second report" on this map). The math works (done by humans obliviously), both about area and population. But one thing they did not consider (and I can't imagine how you could define a rule about it) is that the population of the district is this new-age-kind-of-people (using bicycles, small apartments, very art oriented, left-wing) and part of this population was mixed with Wesmount, which is where all the richest people of Montreal are (and we're talking old money, business oriented, right-wing people). Upon revision, following the input from the population, the committee agreed that it wasn't a good idea to split the district and made other changes instead.

How could you take care of such a process with an unbiased algorithm and with no revision allowed? Emotions are a very important part of this process.

 

[Feeble Wonk]

If the program's parameters are determined by humans, it will be inherently impossible to do that. You also would need an unbiased way to define "moderate 'centrist' elected officials", which is not measurable objectively.

I've apparently not made myself clear. The potential redistricting program that I've inquired about would only remove the existing gerrymandered borders, and replace them by the one designated by the computer program, which would simply attempt to prioritize demographic diversity as reasonably possible within the geographical and population parameters that exist within a given state. Because the algorithm would utilize objective demographic information, it should remove the "human element" as much as possible, other than the initial stipulation that diversity of electorate would be prioritized. My presumption is that this type of "redistricting" would advantage more moderate/centrist candidates, but the voters of the given district would still determine the outcome of any election.

You'd also first need to determine whether or not politics is "pathologically polarized", and whether or not polarization is actually undesirable. What separates our system of government from a monarchy is the fact that we have opposing viewpoints as well as other checks and balances. This entire idea is fundamentally disqualified from being considered objective.

That is exactly my point. The normal "checks and balances" have been subjugated by partisan politics. The current practice of congressional district gerrymandering does NOT provide a fair airing of opposing viewpoints, and equal protection of voting power. Voting power has been functionally manipulated by those in power to sustain re-election. In doing so, opposing viewpoints are marginalized, and left leaning districts/states are moved farther and farther to the left, while right leaning districts/states move farther and farther to the right. It's my opinion that the current congressional dysfunction illustrates that point self evidently. However, that is not the discussion I have been trying to pursue. My question was whether it was realistic, academically, to be able to set up a program algorithm that would create congressional districts utilizing the existing criteria of population, geography and "objective" voter demographics that might prioritize voter diversity within geographically contiguous districts of approximately equal population within a given state.

 

[Feeble Wonk]

Maybe I'm not following clearly what you are talking about, but I cannot see how you can define a set of rules, a standard, better than it is right now.

I guess this is where we disagree. My belief is that congressional districts produced by the program I've suggested would be much more fair than the "set of rules" currently utilized as the "standard". Perhaps the concept of gerrymandering is not as universally understood as I thought.
According to dictionary.com... GERRYMANDER: noun: "U.S. Politics.the dividing of a state, county, etc., into election districts so as to give one political party a majority in many districts while concentrating the voting strength of the other party into as few districts as possible."
This is the "standard" set of rules used to make congressional district borders as of now. The primary motive is to marginalize opposing views and manufacture voting majorities that are not representative of the electorate as a whole.

How could you take care of such a process with an unbiased algorithm and with no revision allowed? Emotions are a very important part of this process.

Actually, that's exactly what I was asking about. Is it even academically possible to do just that. Can an algorithm be devised with only one bias... which is to prioritize diversity of voter demographics to the degree reasonably possible, and otherwise adhering to neutral district construction that normalizes geographic distribution of populations within states that arrives at the same number of congress people already allotted per state.

 

[jack action]

utilizing the existing criteria of population, geography

This is already in place and would be easy to program.

and "objective" voter demographics

This is also already in place (identifying communities) and it could be programmed but it would still be highly subjected to humans that decide what defined those demographics.

that might prioritize voter diversity

This is where I'm loosing you. How do you separate that kind of statement from "gerrymandering"? You basically want to steer the vote towards a predefined outcome. One that you think is more just in your opinion (we still have to all agree what that desirable outcome should be), but sometimes this leads to unpredictable outcomes. What will you do when you'll get an undesirable outcome (compared to your initial objective) and the law says that you must comply to it, no exceptions allowed?

In a democracy, you must have some leeway given by a revision process. No doubt that right now this process is used with bad intentions by a lot of people, but I think it is still essential.

It's like the justice system. It is easy to make laws that criminalize some actions. Still, before condemning someone, there is always a judge that review every case to make sure that the action done by the accused was in fact criminal with respect to the spirit of the law. There are lots of circumstances that change how the law is applied.

 

[Feeble Wonk]

This is already in place and would be easy to program.

Yes, the formula is in place that assigns a congressional representative for a threshold electorate population base, but that has no bearing on where the borders are actually placed... only the number of representatives that a given state is allotted.

This is also already in place (identifying communities) and it could be programmed but it would still be highly subjected to humans that decide what defined those demographics..

Yes, the demographic information is known, but this is actually utilized not to insure representation of the minority population, but rather to strengthen the manufactured majorities in the gerrymandered congressional districts.

This is where I'm loosing you. How do you separate that kind of statement from "gerrymandering"? You basically want to steer the vote towards a predefined outcome.

I must plead guilty to this, in all honesty. My proposal would likely lead to functionally increasing the voting strength of the minority, but the degree to which this would occur could be mitigated by limiting the "area of overlap" of the computer generated congressional districts solely created by normalizing population and geography within a state. But I confess, prioritization of demographic diversity is a reflection of my belief that it would be beneficial to the country as a whole. However, I readily admit that it definitely imposes a bias on the process. Honestly, I would be satisfied if the program simply set the congressional boundaries based on nothing but the objective population/geography parameters. This, at least, would be essentially bias free.

In a democracy, you must have some leeway given by a revision process. No doubt that right now this process is used with bad intentions by a lot of people, but I think it is still essential.

It is this "revision process" that I am calling for, and I do think that it is essential for honest representation of the electorate in congress, and I do believe that it would result in more civil and statesmanlike behavior by our elected officials.
But again, the discussion has veered into political opinion and values, and that was not my intent. I only wanted to know whether it was logistically possible to create such a program algorithm in the first place.

 

[jack action]

but that has no bearing on where the borders are actually placed.

Isn't there some guidelines? For example, here in the province, there must be between 122 and 125 districts. The number of people per district cannot vary by more than ±25% from the average, but exceptions are allowed if motivated. The main directing guideline is:

15. An electoral division represents a natural community established on the basis of demographical, geographical and sociological considerations, such as the population density, the relative growth rate of the population, the accessibility, area and shape of the region, the natural local boundaries and the territories of local municipalities.

For the federal government, it is similar.

I guess it could be feasible to make a computer program that would take this into consideration, but prioritizing a criterion over another is bound to be subjective (sociological vs natural local boundaries, for example).

 

[Feeble Wonk]

Isn't there some guidelines?

Each congressional district consists of approximately 711,000 people, and is allotted one representative in the congressional House of Representatives. That's really the only statutory requirement for most states. In most cases, a state's district lines--for both state legislative and congressional districts--are redrawn by the state legislature, and the majority party controls the process. These congressional district borders often become perversely distorted in convoluted patterns to achieve the majority party's political objective, which is to maintain power.
It seems to me that it would be much more fair if the process was determined by computer generated district boundaries, utilizing a standardized formula that normalizes population distributions and geographical areas as equally as mathematically possible.
As I've already confessed, I'd also think it would be beneficial if demographic diversity was worked into the formula to some degree, because a candidate for office would be motivated to take positions that would benefit diverse segments of the electorate more equitably rather than simply catering to the majority interests. However, I'll admit that it's a personal opinion, and is definitely debatable.
So, if I remove the social demographic parameter, would it be relatively easy to create an algorithm that divides segments of a state's population into districts of 711,000 citizens (or slightly more) in a way that results in the geographic areas of the districts being as close to equal as possible, with no regard to party affiliation or any other factors?

 

[jack action]

So, if I remove the social demographic parameter, would it be relatively easy to create an algorithm that divides segments of a states population into districts of 711,000 (or slightly more) in a way that results in the geographic areas of the districts being as close to equal as possible, with no regard to party affiliation?

I would answer yes to this question (even with the social demographic parameter), regardless of the use of program algorithms or not. Once you create criteria for your guidelines, it can be translated into an algorithm.

But what you are really looking for is independence in the decision process. Here, there are independent commissioners approved by 2/3 of the members of the National Assembly to take these decisions. So you need a well established majority to be able to name people "on your side"; But they still need to be (officially) independent from any political affiliation. So you don't really need a computer to assure the independence of the decision process. The human factor can even be an advantage to adapt easily to any situation.

 

[lpetrich]

Here is what one needs to do:

1. Acquire fine-grained population and demographic data, like for each zip code. That is what one will work from.
2. Create some initial districts, either from scratch or from some existing or previous districts.
3. Improve those districts, using some target measure of good districting, like population equality or geometrical compactness.

Each step of improvement can be done by transferring a zip code from a district to a neighboring district, and then working out the target-measure value for the new districts. If it improves, then use the new districts. Otherwise, keep the old districts (hill climbing) or only do so with a certain probability (simulated annealing.

BDistricting - About -- Brian Olson has developed some redistricting software that works like what I've described.

 

[lpetrich]

For the actual voting part, there are alternatives to first-past-the-post or plurality voting, alternatives that are much more multicandidate-friendly. A simple one is runoff elections. The top two of the first election then go head-to-head in a second election. Another one is preference voting. One ranks the candidates by one's preferences, and these ballots are then counted up to give the overall winner. There are a variety of algorithms for doing so, and a commonly-used one is Instant Runoff Voting, a sequential-runoff algorithm. In each round of counting, for all the candidates remaining in the race, the top preferences are counted up, and whoever gets a majority wins. But if no candidate gets a majority, then the candidate with the fewest votes gets dropped from the count. DemoChoice Polls has polls that use IRV, and you can see how the counting works in them.

 

[lpetrich]

But why stick with single-member districts? Why not multimember ones?

One can extend IRV to multimember districts with the Single Transferable Vote. In each round, a candidate wins if that candidate gets more than some victory threshold. The candidate's "excess" votes continue in the count, with the victory-threshold number of them being dropped from the count. Alternately, all the candidate's votes could get reweighted by multiplying their weights (initially 1) by (number of excess votes) / (number of total votes).

But a more common system is proportional representation. In its purest form, each party gets a number of seats in proportion to how many votes that it got. Parties usually publish lists of candidates that they want to seat, so this is called the party-list system. There are mixed systems, with both district seats, each member being elected by some district, and list seats, with members selected in party-list fashion. The list-seat members may be selected to make the whole body proportional (mixed-member), or else only those seats proportional (parallel).

I myself have written some vote-counting code, to see what different algorithms produce.

 

[anorlunda]

My state, Vermont, has a much simpler solution. We have fewer than 711,000 people, so the whole state is one district.

I must plead guilty to this, in all honesty. My proposal would likely lead to functionally increasing the voting strength of the minority, but the degree to which this would occur could be mitigated by limiting the "area of overlap" of the computer generated congressional districts solely created by normalizing population and geography within a state. But I confess, prioritization of demographic diversity is a reflection of my belief that it would be beneficial to the country as a whole.

I don't think you realize that demographic diversity works to weaken the strength of the minority. Let's say the population was 60% ying and 40% yang. Perfect demographic diversity with a winner-take-all system produces a result where ying gets 100% of the seats. That is why the US Voting Rights Act, specifically prohibits demographically diverse districting, but instead mandates some districts where the minorities can win.

But a more common system is proportional representation. In its purest form, each party gets a number of seats in proportion to how many votes that it got.

This is a better solution than playing with districts. If we ever have a constitutional convention where we have the opportunity to change everything, that is one of the things to consider.

Even without the districting question, even without elections, the proportional representation principle can be used. For example, the Tea Party minority elected in 2010-2012 might have had more influence if they remained a third party, rather than becoming a minority wing of Republicans (ditto for the black Caucus wing in the Democrats). With a 2 party system, when a compromise is negotiated, the Democratic majority deals with the Republican majority. But in an N party negotiation (N>2), the compromise generally gives each party at the table at least some of what they want. So the key strategy for minorities is to gain a seat at the table.

 

[jack action]

I never understood the need to officially incorporate parties in the election process. IMHO, this goes against the democracy concept (one person, one vote; 50% + 1 majority). It seems that everybody wants to alter - and complicate - the actual party system such that it represents more the 50% + 1 majority. Why not just do the democratic thing and vote directly for the head office candidate. Why do we vote for someone who will vote for the head office candidate? If my "surrogate" of choice doesn't get elected, then my vote doesn't count, which is anything but democracy. Worst, my "surrogate" can decide to change his/her mind and vote differently, decide to change party or become independent (basically removing all of his/her power).

I have looked for an answer as to why parties were introduced into democracy and what were they supposed to accomplished, but never found one. For example, in cities (at least here in Canada), you can have parties or not, but we still elect the mayor AND the local counselor.

I don't mind that parties do exist such that a candidate can clearly show what he/she stands for, but I don't understand why these parties are incorporated into the election system. An independent representative shouldn't have less power than one that is part of a party, like for spending or time available to speak in chamber.

 

[anorlunda]

@jack action , you are questioning the premise of the Republic form of government. Parties add a second level of representation.

That's fine, but you won't find explanations of you don't use the right search term.
https://en.m.wikipedia.org/wiki/Republic