Tuesday, November 06, 2012

Redistricting: A Geek Proposal

One of the peculiarities of the U.S. political system is that it is possible in principle for one party to win a majority of both houses of Congress and the presidential election with slightly over 25% of the votes—properly distributed. 50% +1 of the votes in 50%+1 of the congressional districts elect a majority of the house, 50%+1 in the twenty-six smallest population states elect a majority of the senate, although it may take multiple elections, and similarly, with more states, for winning the electoral college.

That uneven a distribution is unlikely to happen by chance—but it is not always necessary to leave it to chance. Back in 1812 Elbridge Gerry, governor of Massachusetts, signed a bill redrawing state senatorial districts in a way that favored his (Democratic-Republican) party. One of the districts looked rather like a dragon, leading critics to dub it a "gerrymander." The term survives, mostly as a verb describing the practice of redrawing electoral districts to favor those who redraw them. It can be used by one party to try to increase the number of its candidates who are elected by concentrating as many as possible of the opposition voters in as few districts as possible and (a somewhat riskier strategy) creating as many districts as possible where their voters are a majority, but not a large majority. It can be done on a bipartisan basis to guarantee incumbents of both parties safe seats by concentrating voters of one party in some districts, of the other in others. California recently redrew districts, using an at least nominally non-partisan group to do it; when I voted this morning, there was a proposition on the ballot to cancel the result. Which reminded me of an old idea of mine for solving the problem.

A state gives some body—the legislature, the Supreme Court, a non-partisan group—the authority to decide among redistricting proposals. There is, however, one restriction. Every proposal must take the form not of a map but of a computer program. Inputs include potentially relevant criteria such as town and county boundaries, but may not include any information on past voting or proxies for voting such racial, educational, or professional characteristics of the population. There is an upper limit to how big the program may be.

That restriction will not prevent people from trying to create programs whose output favors their side, or the body choosing from favoring ones that they think favor theirs. It will, however, sharply restrict their ability to do so, since the information needed to do it well will not be available to the program. They can try various versions, look at the results, and choose the one they like best, but that is all they can do.


Anonymous said...

In before someone makes the entirely sensible suggestion of Nationwide Proportional Representation and then has to listen to all of the spurious arguments against it.

Anonymous said...

I think that's a bit kludgy. Here are two simpler proposals.

1. Mandate a cap on the perimeter to surface area ratio of the districts. Make it low enough to prevent the worst abuses.
2. Allow people or organizations to submit their own redistricting proposals. If they can create one with a perimeter to surface area ratio of 10% lower than the official proposal, then it must be accepted. If multiple proposals qualify, pick the one with the lowest ratio.

Bruce said...

A simpler version of basically the same approach. The state could have a standardized computer program that jointly optimizes a function involving all the relevant variables, like compactness, correspondence with natural boundaries, correspondence with other political boundaries, homogeneity or heterogeneity of population based on income, race, political affiliation, etc. The only thing not specified is the relative weight for all of these factors.

Some fairly large group of people then proposes the weights. For example, there could be a panel of well respected citizen: retired judges, university professors, B-movie celebrities, sports figures, long-time residents, Nobel prize winners, whatever. Each person gets 100 points and allocates them among the factors. These are then averaged, and those weights get put in the computer program and the result is the final result.

I would imagine it would be virtually impossible for a partisan person to try to vote to favor one party or one candidate. Increasing the importance of one factor relative to others would only tend to favor one party or one candidate over when all the other factors are within a certain relative range.

Anonymous said...

Mandate a cap on the perimeter to surface area ratio of the districts.

The problem with that measure is that it's not dimensionless: it's measured in inverse length. Which means that low-population-density, geographically-large districts will automatically get low ratios even if they are gerrymandered, while high-population-density districts will get high ratios even if they're fairly compact.

jdgalt said...

I've been interested in this problem for years, and like many of the big problems as seen by libertarians, I have come up with a whole range of answers rather than just one (since any one answer could easily turn out to be politically impossible).

My #1 solution is to eliminate districts and replace them with a true Proportional Representation scheme such as the Single Transferable Vote. This idea is not really about gerrymandering at all; rather, my purpose in proposing it is to eliminate the incentive trap which leads politicians to believe "my job is to send home pork to my district, and screw everyone else." A legislator who has no district smaller than the whole state or country can't send home "pork" no matter how much he wants to.

However, in any state (or federal government) with a bicameral legislature, I would limit this change to one house, since the last thing I want is to reduce "gridlock" and make it easier to enact laws and spend our money.

(A side benefit of STV is that it will give all small parties representation in proportion to their numbers, if their members vote along party lines. More robust debate AND more gridlock!!)

My #2 solution is to spell out an exact, fixed method of drawing districts in the (appropriate state or federal) constitution, so there is no choice about it on anyone's part. As an example, what follows is a scheme I came up with for dividing California into districts.

* * *

1. All districts shall be composed of whole census tracts.

2. All districts of a given type (for instance, all assembly districts) in the state shall be equal in population, as closely as possible without breaking rule 1.

3. The state is hereby divided into three "strips" from which districts of each type will be allocated. The strips are defined as sets of counties:
(You may find this map helpful)

Strip 1: Counties of Del Norte, Siskiyou, Humboldt, Trinity, Mendocino, Lake, Sonoma, Napa, Solano, Marin, Contra Costa, Alameda, San Francisco, San Mateo, Santa Clara, Santa Cruz, San Benito, Monterey, San Luis Obispo, and Santa Barbara.

Strip 2: Counties of Modoc, Lassen, Plumas, Sierra, Nevada, Placer, El Dorado, Amador, Alpine, Calaveras, Tuolumne, Mono, and Inyo.

Strip 3: Everything else.

Algorithm: Start at the north end of strip 1, and chop that strip into districts by east-west lines (subject to rules 1 and 2) until the remainder of strip 1 has too few people to be a whole district.

Do the same with strip 2.

Then do the same with the remainder of the state (consisting of all of strip 3 plus the leftover end pieces of strips 1 and 2).

Anonymous said...

After posting last night, it occurred to me that there's an easy fix: don't measure the ratio of perimeter to area, but rather the ratio of perimeter^2 to area. This WOULD be dimensionless, and therefore not inherently biased towards low-density or high-density districts. OTOH, it might be harder to push through a political system populated largely by people who take pride in the statement "I never could do math."

Stirling Newberry said...

It took your commenters only two to break the proposition. Namely, hack the rules. One acre one vote will, in most countries, heavily favor the rural party, but it will be easy to pass.

More generally, there are easy ways to constrain the program so that it is essentially a distribution simulation.

Note that Gerry's mander was actually fairly representative of a coastal/river economy, and rivers do not optimize for being a circle, that would be a "lake," but instead for least energy between local minima and maxima of elevation. ("Water seeks its own level.")

The Code is the Law is a solution that is simple, easy, and wrong – to site on old saw which is credited to many in different forms.

Stirling Newberry said...

As a side note, pushing for "existing political divisions" is to say that some previous gerrymander, the one which was done when those divisions were created, should stand for all time. It's rent seeking behavior. Many US divisions were created in a period where districts could have wildly vary numbers of voters. These divisions were drawn up in a cluster around the turn of the 20th century, and often were expressly designed to disenfranchise immigrant populations in the cities. This was struck down by Reynolds v. Sims (1964).

As with one acre one vote, this is a pseudo-principle, one which is actually a quite partisan – in this case anti-urban – assertion in disguise.