### 9 responses to “Political Districting and OR”

1. The flaw here is that “fairness” is defined (by the party in power at the time of redistricting) as “keeps our seats safe for the indefinite future”. So perhaps a more practical model would minimize the geometric goofiness of districts with a side constraint that the incumbent party should retain at least as many “safe” districts as they currently own. (Not that I’m being cynical or anything.)

2. orbythebeach

Paul: Thanks for the comment. I agree that all kinds of cheating strategies are applied, but the (my) hope is that, by using an OR model, the definition of fairness would have to be made explicit and precise, and perhaps we’d get a bit closer to actual fairness.

3. Robert

Here is a link I saw a long time ago about another approach
to redistricting designed by a mathematician.
http://rangevoting.org/SplitLR.html Haven’t looked at in depth but
it seems more geometrically inspired.

4. orbythebeach

Robert: thanks for the link. I’ll check it out.

5. @Tallys: I’m sympathetic to the desire for transparency,
but my strong suspicion is that any redistricting method that
requires approval by the very people who benefit from unfairness.
The one hope would be judicial intervention to require a truly fair
algorithm, but I’m not sure the judiciary would have grounds to
intervene.

6. I think Paul Rubin and I agree on most everything. But I would add that it doesn’t take OR (now called “Business Analytics” by the way) to tackle the problem. It just takes a ruler, either the kind that dictates (like a strong governor) or the measuring stick kind. Draw a square or rectangular boundry no more than x miles around a district and adjust for geography, like rivers. We, in Florida, have the most aggregious district around—it runs from Jacksonville to Orlando, at some places just the width of I-4. It will be interesting to see if it ever gets changed.

7. orbythebeach

Barry: thank you for the comment. The distinction between Business Analytics and OR will vary from person to person, depending on how you define OR. For example, some people would say that Statistics is part of OR, but that’s not a universal belief. For an interesting point of view on the distinction between Analytics and OR, see the comment by Michael Rappa on this blog post. I certainly agree that it’s easy to come up with a feasible solution manually by just looking at the map, but I still believe that using an optimization model can bring you closer to satisfying criteria like population equality with a lot less trial-and-error. Of course, as Paul correctly pointed out, this is one of those problems in which the decision makers are biased against the mathematically ideal optimal solution. Nevertheless, I still enjoy having this kind of discussion because it highlights the pros and cons of the mathematical approach and helps increase the visibility of our field.

8. I have another reading on the morphing from OR to analytics. You can see my blog of a few days ago at http://www.heizerrenderom.wordpress.com. There are many reasons why the change is needed, and you are 100% correct re stat and the differing views people have (which is, of course, one of the reasons we have a problem in our profession).

9. Burçin Bozkaya

I also read this article and I, myself, have academic and practical experience in use of OR in political districting. I agree that the key issues are fairness (or objectivity) and transparency. When I mention to people that we have algorithms for automated districting, I instantly get a feedback that these are suitable only for other types of districting, not for political districting. Being an OR professional myself, I think otherwise, as OR methodologies, if they can fully (or near-fully) capture the governing rules and they are also transparent, can form an objective framework where few can claim that there is deliberate gerrymandering or other manipulations.

In our recent project with the City of Edmonton in Canada for city council election districting, we applied a heuristic multi-criteria algorithm (that originated from my PhD thesis) integrated with ArcGIS. Using this approach, it is possible to quickly (in a matter of seconds) generate good quality maps that appeal to different tastes (using adjustable weights on multiple criteria). The city officials would in fact take this as their first step and then modify the map as they see fit. But in our experience, the number of modifications were so minimal that we were convinced the algorithm can successfully address most, if not all, of the relevant districting requirements. This work will be published in Interfaces soon. An earlier work of ours was published in EJOR.