Hacker News Reader

Comment by mattbrewsbytes | Hacker News Reader

One article I read said that there is a 30% Republican voter base that is covered by 76% of the congressional maps. They redrew the maps so they would have 10 seats and Dems would have 3. In the past I've read the Democrats have done similar things.

As a software engineer, isn't there some way to draw maps which balance out demographics programmatically? While taking into consideration population, rural, suburban, urban areas, etc. Shouldn't there be a national policy/process that is followed?

It would make some sense to abolish recording what party people affiliate with. It really serves no purpose. Let people vote in both parties primaries.

View on HN

Data Scientist with a Master of Public Administration degree here (don't ask). My proposal for solving this has always been to frame it as an optimization problem.

You have a certain number of census tracts (essentially, neighborhoods) with known populations and geographic areas. You know you need to divide the state into X congressional districts.

You should simply assign census tracts to districts such that 1) you end up with X districts, 2) each district's total population is within a Y% tolerance of state population / X, 3) you minimize the average of the lengths of the district borders (to force maximum geographic compactness).

Given available data from the Census, some mucking around in PostGIS or ESRI, and a state-of-the-art commercial solver like Gurobi, there's absolutely no reason this couldn't be done.

It'd also be trivial to add on requirements for balancing demographics or political affiliation within certain tolerances as well.

As it is in most countries...

Is it? I'm admittedly fairly ignorant on how other countries run their election systems, but... geographic representation seems uncommon (lots of parliamentary systems) and sophisticated solutions to drawing districts within countries that have geographic representation would have to be even more rare.

I am familiar with Parliamentary systems of 5 countries (Britain & former colonies) and they are mostly geographic, to my knowledge. Each Members of Parliament is voted in by voters residing within a geographic region/ward.

Interesting... looks like a lot of countries with parliamentary systems use proportional instead of geographic representation... but not the UK or Canada. Today I learned.

https://en.wikipedia.org/wiki/Proportional_representation#Li...

And some countries manage to achieve proportional representation while still giving each district their own choice of representative (e.g. Mixed-Member Proportional Representation in Germany and New Zealand).

That actually sounds like it would balance the competing needs of wanting neighborhoods with an identity to have a voice and not stacking districts to unfairly affect the percentage of representative affiliation compared to the general population quite nicely.

Who would have thought someone that has spent a large chunk of their life studying and working in the areas that are required to solve this problem would come up with a useful solution? Not our representatives, apparently... :/

> As a software engineer, isn't there some way to draw maps which balance out demographics programmatically?

There are mathematicians who study gerrymandering and election maps [1,2].

[1] https://sites.tufts.edu/gerrymandr/ [2] https://phys.org/news/2017-08-math-experts-brainpower-gerrym...

I think the current analytical mechanism to achieve balance is by measuring the “efficiency gap.”

https://chicagounbound.uchicago.edu/cgi/viewcontent.cgi?refe...

The recent coverage of considerations of the efficiency gap by the Supreme Court would be available here. The illustrative graphs provide a bit quicker of an introduction to the measure.

https://www.washingtonpost.com/graphics/2017/politics/courts...

According to http://pulse.ncpolicywatch.org/2017/08/22/analysis-large-eff..., NC's efficiency gap is around -11, while anything in excess of +/- 7 is considered a very large gap, so by the efficiency gap metric NC's map is definitely extreme.

> In the past I've read the Democrats have done similar things.

See https://en.wikipedia.org/wiki/Maryland's_3rd_congressional_d... for an example...

> isn't there some way to draw maps which balance out demographics programmatically

We can't even agree on what "balance" means. In particular, people want both majority-minority districts and party-balanced districts, which is a pretty tough constraint in practice.

The Supreme Court will hear a case precisely about that

https://www.washingtonpost.com/news/monkey-cage/wp/2017/10/0...

Fivethirtyeight.com people have written a lot about it (including why you can't simply write a program and go home, just like you can't for most political questions). They also have an ongoing podcast series that talks about the issue in great detail 'The Gerrymandering Project'.

Maybe it's time to stop using geography? Geography was an important consideration in the horse/trail era where if you were stumping around trying to get people to vote for you, how far you had to travel mattered.

Plus, before skyscrapers and other sorts high density developments geography had an okay correlation with population density, but that is no longer the case anymore. There are often far more people in a single block in a city than on an entire knob out in the knobs, or an entire mountain in mountainous areas, etc. You can't easily "circle" just a floor or three of a single building on a 2D map with any accuracy, but to balance population density fairly you may have to consider problems like that.

With TV, radio, the internet, skyscrapers, and urban density, maybe we need a whole new solution for representative democracy than where physically you collect your snail mail?

I don't have any good theories on what should replace geography, but I do think geography is a problem we could eliminate by ignoring that variable.

Most problems are constrained by the facts that it costs resources to move things, and that you can't trust people to play fair when they can get ahead by cheating.

The latter makes us establish physical security around election polling sites, and the former requires that we spent money to move people, machines, and consumables into and out of those sites.

As such, those costs require that political districts consider geography as a practical, money-saving measure, especially roads and buildings. You have to print N blank ballots and send them to the polling stations, and then send back any unused blanks, and the local issues on ballots for adjacent districts might be different.

In order to eliminate geography, you would need a system that has good data security without requiring physical security for the whole network. And an integral part of that would be a foolproof and uncheatable online identification system.

Good luck with that.

Additionally, geography does matter for the group for similar reasons. Proximity constrains economics and policy.

You can consider geography as far as logistics without using geography as the determining factor as far as who gets elected.

You could simply let anyone vote for anyone and elect the top N contenders where there is N seats.

Logically the choices would resemble the choices of the population.

You can also use the slow but difficult to default distributed network we have had the longest. The US mail.

> You could simply let anyone vote for anyone and elect the top N contenders where there is N seats

You could, but that would actually be worse the FPTP in single member districts.

> Logically the choices would resemble the choices of the population.

No, they wouldn't, except by chance. Say you have six candidates, A-F, for three seats. 75% of the electorate prefers an ideology generally espoused by A, B, and C; 25% prefers an ideology generally espoused by D and E, and 10% prefers an ideology espoused by F.

In the first group, A is by far the most effective campaigner and gets 70% of the total vote, with the remainder split evenly between B and C; in the second group, support is split equally between D and E. The last, of course, all goes to F.

So, in your top 3 vote, A and F win seats in the clear, and the third vote goes to a tiebreaker between D and E; this does not at all reflect the distribution of views in the population.

That's why methods like Single Transferrable Vote and it's relatives, using preference rather than bullet ballots, exist.

I'm all for innovative methods like single transferable vote and agree they would be more effective than FPTP but I think this is logically distinct from whether we should decouple votes from geographical locations.

They aren't practically distinct, and the benefit you claimed upthread is not realized with the voting system (which is also an innovative, non-FPTP, system, just a bad one) you proposed.

Right, Oregon and Washington figured out an easy way to decouple geography logistics of polling places and voting is to use absentee ballots for everyone and allow the physical security of the postal service to do its job.

It's not an interesting solution to the real-time results obsessed crowd, but it is a solution proven to mostly work.

The point of districting is to have the representation be more connected to a particular demographic than the whole state.

I get what you're saying about geography.

But the alternatives have a lot of pitfalls. An at-large candidate from Texas or California is not likely to be particularly close to any minority demographic that might desire a champion for their concerns. You end up with every candidate from Chicago, campaigning in Chicago and its collar counties, and no one ever visits or gets elected from Carbondale or Bloomington-Normal or the Illinois part of the Quad Cities.

It might be appropriate to segment the population by profession, such as to have all the bus drivers, car drivers, train operators, and pilots in the state nominally represented by the same person, while all the farmers, loggers, and ranchers have their own person. That faces the same problems as geographical gerrymandering, and more.

If you segmented the population alphabetically, the representative for surnames starting with the letters A through C does not have a particularly clear mandate. But at least it would likely be a fair distribution.

I think the simplest segmentation would be between urban and rural. Line up voters by the population density of their census tract. Slice the population into a reasonable number of buckets, 2 or 3, and the voters in those buckets elect a number of at-large representatives. You probably end up with liberal representatives for every urban core, moderate representatives for every suburb and commuter artery, and conservative representatives for every hamlet and unincorporated township that can still see the stars at night.

If rural voters are a small minority of the populace why should their distinct regressive views have more than a mathmatically appropriate ammount of weight? When tyranny of the majority is most people agreeing people ought to have more rights than rural hicks think they ought to deserve I fail to see the problem.

Democracy means representing the majority view tempered by personal judgement.

But they are not a small minority. It just costs more in campaign funding to sway their votes. You would have to have a smaller event in every county seat to equal one big political rally in the metropolis. That's a lot of time and distance, and the payoff goes into the basement if other candidates are doing the same thing.

Or you could just make this simple and use STV or a similar preference-based, candidate-centered, proportional system, which effectively segments the electorate by candidate preferences expressed in the election, rather than prior to the election. This essentially minimizes distortions resulting from pre-vote segmentation.

In Oregon we vote by mail. So there's one way.

Geography still matters because your interests are entwined with where you live, at least so long as you continue to venture out the front door.

Most people's social lives are entirely virtual these days. The boundaries are getting fuzzier.

Yes, there will always be direct local interests like fire, police, emergency service, sewage, utility, etc districts, but those services all have natural boundaries themselves, and the political boundaries can and often do match there at city/county/township/x-district levels.

But the state and federal levels, at least in the US, seem increasingly divorced from geography (outside of the Cold Civil War between rural and urban economies/beliefs/interests). The weird shapes of current gerrymandered districts is only possible because the geography is so divorced from the governing. If the shape of most US congressional districts was the shape of fire or sewage districts there would be riots in the streets.

It doesn't even have to be a complex mathematical equation. Gerrymandering is a particularly acute American problem which isn't present in most other industrial democracies, because in most cases, an independent commission is picked to define district boundaries, and they are mostly guided by common sense and fairness, which seems to work well enough. The problem the U.S. has is to leave this task to people who have an inherent conflict of interest in the outcome of the process.

It's certainly not an issue recently, but the UK used to have lots of issues with it, see rotten and pocket boroughs (https://en.wikipedia.org/wiki/Rotten_and_pocket_boroughs).

That's in some sense the opposite problem - constituency boundaries _didn't_ change so became manipulable, rather than the boundaries being deliberately drawn to introduce bias.

Interestingly, a university in North Carolina is working on this exact problem, right now. Guess how they got interested in it :).

https://services.math.duke.edu/projects/gerrymandering/

related: guess why the state legislature radically altered the UNC board of governors

That is related, but the university I was referring to is Duke :).

It's very true that in the past the Dems have behaved similarly.

I used to agree (also a software engineer) that this is a puzzle that ought to be solveable with weighted voronoi diagrams or somesuch but then I read: https://peoplespolicyproject.org/2017/10/17/the-problem-of-d...

> As a software engineer, isn't there some way to draw maps which balance out demographics programmatically?

Once you define which of the many competing priorities for districting (aside from naked partisan advantage and disenfranchising particular races, which we can assume are agreed not to be acceptable priorities) to optimize and how to weight them against each other this becomes easy.

Of course, the people with illegitimate goals can easily game out which combination of the notionally legitimate goals maximizes their actual illegitimate ones, and advocate for that as the standard.

Independent non-partisan redistricting commissions solve this problem easily. But only a few states (Washington, California among the bigger ones) are brave enough to go through with it.

There are very few independent, non-partisan human beings. Simply changing who does the redistricting and the means bu which they are chosen to do so doesn't solve the problem, because the people who are interested in and chosen to do the task are always going to have political views.

Even having bipartisan commisions doesn't solve the problem. Parties will work together to secure incumbents.

It basically comes down to a who watches the watchers issue.

No, it actually works. When California went through with it, the “other side” actually gained seats. It takes the drama out of redistricting, and districts are of a much regular shape. Of course, isn’t perfect, but a sucks way less solution is much better than a sucks way more one.

> It takes the drama out of redistricting,

The California process may produce better results on average by some measure (and is emphatically more transparent) compared to legislative redistricting, but it emphatically vastly increased the drama associated with redistricting.

How so? Californians seem really happy with it, all things considered.

Largely, the transparency, which meant public hearings with lots of involvement from interest groups, with accusations (which turned out to be true) that both major political parties were actively working behind the scenes to coordinate those groups to bend the process to their partisan advantage, and all kinds of related drama that didn't happen with legislative districting.

But, yes, I think people are mostly satisfied with the results, which is a separate issue from drama in the process.

California has a multipartisan redistricting commission consisting of equal numbers of Democrats, Republicans, and members of neither major party; it does not have a nonpartisan redistricting commission.

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

California is listed as "Non-partisan; super-majority (majority of each group) needed", you can argue that Wiki is wrong about this, but thats the source I'm using.

Wiki further clarifies its use of "non partisan":

Non-partisan means that either, a) the partisan makeup of the commission is not specified beforehand, or b) a substantial portion (i.e. more than one) of the membership of the commission is reserved for political independents or members of so-called Third Parties.

That's a kind of odd use of “non-partisan”—both (a) and (b), actually, since (a) would allow a commission entirely of political appointees of the sitting governor, without in-law specification of partisan makeup, to be a “nonpartisan” commission, and (b) allows a commission where appointment is specifically slots by party and most seats assigned to the two major parties to be non-partisan.

But, yes, by that odd definition, the California system is non-partisan under (b).

First define what goals are "fair" or "balanced".

Proportionate representation.

Proportionate along what lines?

Voter parties?

The interest voters express preferences for by voting?

The interests voters idealize (they may vote pragmatically within existing rules)?

Etc., etc.

A purely algorithmic solution needs to encode preferences and how to balance them, so they need to be defined quite precisely.

Proportionate by whatever groups the voters self-select. If there are 13 seats then any group of 1/13th of the voters that votes for the same candidate should get that cantitate election.

That's not a district plan, it's a different way of apportioning representation. Hence my confusion.

Not that I disagree, but at what level do you establish proportional representation? At the state, county, city/town/municipality, somewhere in between? Currently, I think the boundaries are drawn ad-hoc but loosely defined by population, geography, and demographics that benefit who ever is in charge of the pen.

Some fancy application of Voronoi Diagrams[1] seems like it would make sense to me.

[1] https://sites.math.washington.edu/~morrow/mcm/uw_1034.pdf

I used to share your reasoning that a straightforward technical solution must exist but this article really opened my eyes to the various tradeoffs involved in districting and how finding a balance is indeed a political not a technical problem: "Gerrymandering isn't evil" https://www.politico.com/magazine/story/2015/06/could-gerrym...

Software would be relatively easy, enacting the policy difficult. Who would want to relinquish their power to a computer?

And the concerns of bias in the algorithm could still be there. As would the fear of hacking. Neither seem as big of an issue as flat-out intentional gerrymandering and could be mitigated by open review process.

But it's unprecedented which makes it hard to enact without a lot of momentum.

Keep in mind the whole point of these districts is to bias in favor of a demographic. If you want to automate that then you also have to explicitly define the biases. If you don't want biases then get rid of the districts.

Ironically enough, the software seems to already exist, but right now (according to the article) it's being used perversely to draw the districts in the least fair way possible.

How would the software work? You want to put communities together. Ideally, you don't want your community to be split in the middle because of a congressional line.

Not saying that isn't happening now, it is. But software doesn't solve that issue.

So we created this great problem: congressional lines. Then we try and fix it. Representation is hard. Why not do away with the problem entirely? Remove this type of representation.

Radical, yes, but what does congress do anyway? ;)

> you don't want your community to be split in the middle because of a congressional line.

why does this matter? congresspeople vote on national issues, not local issues.

Ok then why have these lines at all? That is my argument.

Reply on news.ycombinator.com