Sep 13

Tim O'Reilly

Tim O'Reilly

Fair Redistricting

automatic redistricting map of California

I wasn't able to attend the Boston Ignite last week, but via the great list of talks on the site, I discovered Brian Olson's provocative proposal to End Gerrymandering Through Automatic Redistricting. The talk description says:

Lots of states have crazy congressional districts drawn to the benefit of one party. Let a computer do it fairly!

It's an interesting proposal, not only in and of itself, but because of the way it highlights how computer power can be applied in new ways to many common problems. It seems to me, in fact, that the ongoing development of Web 2.0 will be driven by "aha" moments in which a combination of data mining, collective intelligence, and a little bit of social engineering can provide us with new options as a society. (I mention collective intelligence here because there's no need for a redistricting algorithm to be based purely on population. Much as search engines use many factors to give the best results, so too could redistricting programs. For example, they could use precinct voting records to adjust for actual propensity to vote in each of the proposed districts.

Of course, it's fairly naive to think that our representatives want the process to be fair. Whether manual or computer aided, redistricting will tend to be rigged to favor the incumbents. But we as a society can demand otherwise, and over time, perhaps make a difference. As baseball statistician Bill James said: "There will always be people who are ahead of the curve, and people who are behind the curve. But knowledge moves the curve."

tags: web 2.0  | comments: 13   | Sphere It

Previous  |  Next

0 TrackBacks

TrackBack URL for this entry:

Comments: 13

  Charlie Park [09.13.07 07:31 AM]

His writeup was really good. The kind of logic you'd hope someone in a seat of power would apply, although, as you said, that's a naïve hope. I'm sure his presentation was well done.

Another part of his notes that I thought particularly interesting were his comments on Proportional Representation. I'm not sure what the best way to give disenfranchised groups representation in an equanimous way is, but the proposal is thought-provoking. Wikipedia has a more in-depth study of the issue. Especially interesting was his suggestion of a PR Senate and a geographically-based House.

  David Megginson [09.13.07 08:08 AM]

If there's any way to eliminate democratic abuses in the U.S., it will have to start with establishing fair, non-partisan institutions and systems that people trust. Technology, properly used, can make those institutions and systems more efficient, but it can't make them fair and non-partisan in the first place.

In Canada, we manage redistricting fine with humans, through an independent electoral commission. MPs sometimes grumble a bit when their ridings change or disappear, but no one suggests that the system is in any way partisan, and all political parties trust it. Similarly, we still use paper ballots for most elections, but manage to have the initial election results up within minutes, and national election results within a couple of hours of the last polls closing. Candidates sometimes request local recounts when the results are close, but no one questions the fairness of the system itself.

  Jerome McDonough [09.13.07 09:38 AM]

I was working as tech. support for the U.C. Berkeley Institute of Governmental Studies back in the 1990's, when some of the researchers there were collaborating with the City of Oakland in their redistricting process and using a GIS system loaded with census data to try to redraw the lines for the City Council districts. It was actually a relatively simple process to feed a set of parameters to the computer and ask it to do exactly what Brian proposed, create districts with even numbers of voters and with as little geographic "stretch" as possible. We even created those maps and showed them to City Council members and some community groups.

You are absolutely right that the City Council members hated the new district lines. The surprising thing was the degree to which the community and neighborhood groups hated them as well. The 'fair' lines were denounced as anything but almost universally.

Two notable examples: The Piedmont district (in City Council District 1) and Chinatown (District 2) were divided up. Piedmont was essentially cut in half, split between district 1 and district 3; Chinatown ended up being part of three different districts. Residents of both areas saw themselves as being part of a single neighborhood, and saw the splits as almost certain to result in any neighborhood concerns they might have being easy to ignore after the redistricting. Since the residents ended up being very small minorities within the resulting districts, council members wouldn't have had to pay as much attention to their concerns as under the current redistricting.

I think the problem with proposals like Olson's is the assumption that an algorithmic process is somehow inherently "fair." Redrawing district lines is inevitably a redistribution of political power. Determining what constitutes a "fair" redistribution is itself a political decision, as evidenced by the arguments in California now as to whether to change the distribution of California's electoral college votes from its current winner-take-all system. While I agree with Brian that the current system has some serious flaws, trying to pretend that this *isn't* a political discussion and hide that behind a computer algorithm seems to me no great improvement.

  Thomas Lord [09.13.07 10:29 AM]

Just as a point of fact, computers have played a role in districting for something like 30 or 40 years now. Here's a lead from a library on the history of Texas redistricting.

It is not just naive to assume all representatives want a fair process, it is naive to assume that the current process is unfair or that a new kind of computer-generated map would be more fair.

Gerrymandering has the effect, generally, of strengthening incumbant majorities. As a result, you wind up with situations where the percentage of representatives from a given party is not the same, jurisdiction-wide, as the percentage of people who voted for that party. The unasked question is: so what?

If we want a general-election parlimentary system then press for that, not for different districts. Computer redistricting doesn't help much here because different people will define "fair district" different ways and different, equally reasonable computer programs will yield different results: gerrymandering is unavoidable; it's part of the game.

One thing that computers are good for is analyzing districts and detecting systematic vote surpression. Districting-based vote supression is not proven by percentages of parties in office but by the mathematical logic of the vote as it applies to geographically concentrated populations. For example, in one famous case, a city (or perhaps it was a county) whose population was geographically segregated was consistently carved up for years in such a way that white neighborhoods formed districts, but the large black neighborhoods were split up among those. The black community could organize to vote as a block but only to no avail. The court ultimately stepped in and ordered a court-supervised redistricting. As I recall, ending in the election of the first black council member.

That was a specific complaint, about a specific set of districts and a specific population whose votes were being diluted in violation of voting rights laws. Does that mean that all redistricting, even party-motivated redistricting, even when it creates weird-shaped districts -- does it mean that all of that is unfair? Not at all: it doesn't follow that just because a districting favors one party over another that, therefore, votes are being systematically surpressed. Every districting will favor one party over another and almost everywhere you can count on being able to find alternative districtings, all equally "intuitively fair", yet lining up on both sides of party favoritism.

If you're looking for a political strategy for, say, the Texas legislature, two come to mind: One possibility is that more Texans could join the republican party and work from the inside. Another possibility is that the Democrats could just come up with some better platforms.

Meanwhile, those who draw district maps already use computers. Those who challenge them on civil rights grounds use computers. We just haven't, and never should take the insanely hubristic step of automating districting by some legislatively endorsed algorithm. (My god, what if we did automate that way? Can you even begin to imagine the distorting effects this would have on property values?)

Reminder: cheerleading is dangerous,

  Mark [09.13.07 12:41 PM]

I have to echo the well stated comments of Thomas Lord - there is really *nothing* new in Brian Olson's proposal to "End Gerrymandering Through Automatic Redistricting". It appears he has failed to review the academic literature on the topic (not uncommon these days since a lot of this material is considered "old" and doesn't show up on Google). I remember attending a presentation by the geographer Peter Haggett about 30 years ago on the topic of using computers for redistricting (with the example of eliminating Gerrymandering). It was apparent back then that computers were not going to be able to create "fair and unbiased" electoral districts as someone still needed to define the actual rules for restricting (by population proportion? by compact geographic area? by ethnic diversity? by age group? by past political party voting or affiliation?). Redistricting by various geographic criteria alone turns out to be very inadequate (e.g. highly "biased") but what additional population criteria should also be included? It is the combination of these numerous allocation rules, not the use of computers per se, that define districts and there is no way everyone is going to agree what specific rules to apply. The fact that some individuals still think that computers are "fair" will somehow replace human decision-making in such cases is a bit disturbing. Has it reached a point that the history of computing is being forgotten and is now repeating itself?

  Tim O'Reilly [09.13.07 01:06 PM]

Good comments, Jerome and Tom. Thanks for the perspective.

  Chris Zambito [09.13.07 03:24 PM]

I think the thing that will become more common along these lines (as indicated, this use of districting already occurs on local and regional levels for many purposes) is that the general populace will become more aware of the effect of districting and aggregating data at different scales. Like all things, there will be good and bad uses of GIS and other spatial tools, but visualization and spatial analytics are quickly coming into a more digestible form for the general public. Community planners, academics, and other with the tools are used to looking at information this way, but others may not be. It also has the danger to induce paralysis by analysis as the amount of variables used can quickly cause a dizzying array of results.

  California for Barack Obama [09.14.07 01:23 AM]

Like you said, it's very unlikely that we'll ever have the political lines drawn fairly. Hell, we even depend on this tactic to get presidents elected (aka '2004). Just looking at Florida's mess last time makes me hope the government gets their act together this time.. or at least whatever it takes for Obama to get elected.. :} Maybe we should redraw the lines.

  Chris Kennedy [09.14.07 12:39 PM]

To reiterate the obvious: Nothing new here... move along. His "proposal" for automatic district creation by maximizing compactness is trite to the extent of absurdity.

Redistricting minimalists, and computer science theologists, have proposed automated, "objective" districting procedures since redistricting was invented in the 60s. We know this. Readers: do a Google Scholar search on "redistricting compactness". First link: "Nonpartisan Political Redistricting by Computer" (Hess, Weaver, et al., 1965). Criteria? Equal population, contiguity (joy), and compactness (via population moment of inertia). Thank you Bolson for a proposal 42 years too late.

Algorithms are not the important problem for redistricting (true, heuristic approaches to district optimization have made good progress in recent years)... the problems are social in nature, in particular political feasibility of reform, and, relatedly, agreement on standard criteria.

There is also the small problem of VRA compliance that also takes us back to 1965, and if the Supreme Court can't even create a consistent policy on it for more than 10 years, how is an automated redistricting algorithm going to have a chance? Oh yeah, for the sake of simplicity we are choosing to avoid reality.

Finally, Bolson is incorrect that a constitutional amendment would be needed: redistricting policy can be changed via Congressional statute. Let's not get ahead of ourselves.

Ah, the genius that results from ignoring a dense, 60 year literature across political science, mathematics, law, economics, and geography. I'm sure if we add Ajax support it will revolutionize our political process.

  Chris Kennedy [09.14.07 12:50 PM]

Follow-up: if you can only read one document on redistricting, it should be Micah Altman's 1998 doctoral dissertation. It is available online at and, in particular, Chapters 4 and 5 are "Predicting the Electoral Effects of Mandatory District Compactness on Partisan Gerrymanders" and "Is Automation the Answer? -- The Computational Complexity of Automated Redistricting".

  Dahlgren [09.14.07 02:52 PM]

Computer-aided gerrymandering has indeed transformed that black art into a precise science, but it's merely the tip of a black political iceberg.

The major scandal is that there are only 435 Congressional House Representatives for 300 Million Americans -- there should be many times that number in Congress for fair representation of the citizenry.

The political power of an individual Representative has increased a hundredfold since the 1st Congress in 1789 -- by artificially restricting the total number of Representatives. Gerrymandering is but icing on that huge politically-baked cake. Congress could easily increase its official membership to better represent its 'constituents' -- but politicians never willingly diminish their power.

The 1st U.S. House of Representatives had 65 members for about 4 Million Americans (~ 1 representative per 6,000 persons).
Today's House has 435 members for 300 Million Americans (~ 1 per 690,000 persons).
The original U.S. Constitution demanded that a Congressional Representative 'represent' no more than 30,000 persons.

The geographic shape of Congressional voting districts was unimportant under the U.S. Constitution; however, the total number of elected representatives 'per state' was very important under our system of a constitutional republic of 'federated states'.
Simple district lines around roughly equal numbers of people... worked OK-- until the Supreme Court invented a one-man/one-vote scheme in the 1960s that opened the door to vast gerrymandering across the nation. The Constitutional dictate of 'proportional' representation was that among formally sovereign 'states' of the federated "United States" -- not among politically arbitrary voting 'districts' within a state.

...step back from counting gerrymandered 'trees' -- and contemplate the political 'forest'.

  Brian Olson [09.17.07 11:43 AM]

Ok, I admit it, I'm a bad academic. I didn't fully read up on the literature. Maybe this is why I didn't go to grad school. It was more fun to just hack something up and see what I could make work. Anyway, thanks for the links. I'll definitely read up on some of those now.

I'll still defend the value of having a proof-of-concept around to point to. If someone asks what the result of such a system might be, I can point to results. Having it in a modern code base, still under active development (by me, at least), should keep the discussion going too. When the next Census comes out in 2010 I'll be right there providing alternative district maps.

  Mark Rush [09.26.07 09:28 AM]

Was referred to this post by a colleague. I wrote a lot on redistricting and gerrymandering in the early 1990s. Problem is that there really is NO interest in reform. Too many interests--from incumbents to geographically concentrated interest groups--have a stake in this system.

One can propose to reform electoral districts. But, the fact is, so long as we use geographic districts, some interest or another will be at a disadvantage.

I do think that the "solution" (if there actually is one) to the gerrymandering problem is a move to multimember districts or to statewide PR (where it is feasible. If you draw more multimember districts, you have correspondingly fewer lines to draw.

I think this could actually improve politics because candidates might actually have to band together and coordinate campaigns. This would also rejuvenate party organizations.

MMDs would enhance competitiveness because larger districts would enable challenges to draw upon broader constituencies for support. Similarly, they would make incumbents more attentive because they'd have to campaign harder (instead of looking to win a race in a gerrymandered district that is not competitive).

Changing the electoral system is not a panacea. Many electoral reform advocates act as if reform would do everything from fix politics to cure your ED and stop hair loss. It won't.

But it would change the political dynamic by changing the incentives to run for office.


Post A Comment:

 (please be patient, comments may take awhile to post)

Type the characters you see in the picture above.