Friday, January 15, 2010

The Naughtful Naughties: Gerrymandering and an Electoral Results Database Plea

No data-based post today. Instead I'm going to have a long generic discussion of the theoretical gerrymandering, followed by a plea asking whether anybody can steer me to a good database or set of online tables of electoral results, especially for US House elections and/or state legislatures.

Why am I asking for such a thing you ask? So far we've looked at a few sets of economic data. But at some point this year I'd like to take a look at another phenomenon whose effect I think worsened during the Naughts: gerrymandering.

As databases and data analysis has grown more sophisticated over the years, it has become easier for redistricting committees to design districts that tilt towards one party or another. In cases when a single party controls the redistricting party, this allows incumbents to draw districts that maximize the number of representatives that there party will have.

The way this is done is to design as many "reasonably safe" districts for the incumbent party, and to crowd as many voters from the other party into a very few "overwhelmingly safe" districts. "Reasonably safe" districts tend to have something like a 55%-45% to 60%-40% advantage for the incumbent party, while "overwhelmingly safe" districts can have an advantage for the out party of anywhere from 75%-25% to 90%-10%.

Let me give you an example. Let's say we have a state with ten million voters and 100 state house seats. For sake of our example, we'll split our voters into five million voters that favor Party A and five million voters that favor Party B.

In an absolutely even mathematical construct, we would have 100 districts, each of which has 50,000 voters from Party A and 50,000 representatives from Party B. In a completely neutral election these districts would split randomly, 50 for Party A and 50 for Party B.

Geography and demography is not a random mathematical construct of course. Because different areas favor different parties, what we might reasonably expect would be a bell-shaped curve in which we have something like five "overwhelmingly safe" seats for each party, ten or fifteen "reasonably safe" seats for each party, and perhaps as many as 50 or 60 seats that could potentially be up for grabs in each election, depending on which way the political winds are blowing.

However, a clever gerrymander will shift the apportionment. Let's say that Party A is in power. They can then draw up districts that look something like this:

Reasonably Safe for Party A - 83 seats
55,000 from Party A
45,000 from Party B

Overwhelmingly Safe for Party B - 17 seats
25,000 from Party A
75,000 from Party B

(One of those "overwhelmingly safe" seats for Party B needs to be a 35,000-65,000 split to make the math work perfectly, but that's still an overwhelmingly safe margin.)

What effect does that sort of districting have on our 100-seat legislative body?

The most obvious, of course, is that the body will vote overwhelmingly for the positions of Party A on each and every issue. And because their advantage is so large, they will have no incentive in considering or including even part of the platform of Party B as they craft legislation.

But what about cases in which redistricting is shared between the parties? Even in these cases the redistricting is being drawn by incumbents. So it shouldn't surprise anybody if the redistricting tends to favor creating safe districts for incumbents. We may end up with a legislature that is split 50 seats to 50 seats; however, what we are likely to have in district makeup is something like this:

Safe for Party A - 50 Seats
60,000 Party A
40,000 Party B

Safe for Party B - 50 Seats
40,000 Party A
60,000 Party B

The popular description of this phenomenon goes something like this:

Instead of a system in which the electorate chooses their representatives, we have a system in which the representatives choose their electorate.

Whether there is a single-party or shared gerrymander, there is very little chance that we will see seats change parties during the decade that follows the redistricting. This means that the real election to choose the person who holds that seat is the party primary, in which that parties nominee for the general election is selected. Independent voters either shun party primaries or are shut out of them altogether, depending on that state's rules. Primaries also see low voter turnout and tend to be dominated by the most active segments of each party's electorate, the "wings" of their party. In the case of Democrats and Republicans, these tend to be the liberal and conservative bases.

This means that in our gerrymandered legislatures, if you want to represent a district, you should have policies that favor the wings of the party for whom that district is safe. In our single-party gerrymander this means that not only do we have a legislature with representatives who favor Party A, we have a legislature that overwhelmingly favors the most extreme factions of Party A with a noisy minority from Party B. In our two-party gerrymander we have a legislature with 50 representives from the wing of Party A and 50 representatives from the wing of Party B. In either case there's no electoral incentive for the representatives to govern from the center of the poltical spectrum.

Thus, does gerrymandering push centrists out of power altogether.

In our theoretical gerrymandered legislature we would expect to see extreme partisan rancor, very few representatives crossing party lines for votes, legislation that favors the most extreme positions of the controlling party, and a growing sense of disenfranchisement among centrists. The more sophisticated the gerrymandering, the more likely we are to see those behaviors.

Does this sound like a legislature near you? You may have gerrymandered districts.

Was gerrymandering any worse during the Naughts than in previous years? That's why I'd like the data in a nice, convenient bundle. So we can see whether reality matched my nice theoretical construct above.

If gerrymandering is afoot, I would expect to see a pattern in which the average margin of victory is greatest in the year following a redistricting (2002, for this decade) and slowly decreases as voters shift and move. And if it's worse in this decade than previous decade, I would expect to see a pattern in which the average margin of victory for US House and state house elections is larger in this decade than in previous decades.

The data's out there, and looking up individual elections is easy enough. But if somebody's already compiled it all in a single place, I'd just as soon not replicate the effort, and would appreciate you steering me in that direction. Muchas gracias!


  1. That is a wonderful explanation. Thanks. I look forward to seeing if the data you find match up with the patterns you expect to see. Wouldn't it be nice if the DCCC had such data compiled! Providing such a database would be a true service and in the case of the DCCC might even make me begin to alter my opinion of that organization...

  2. Nice analysis, as always. As I remember, such data are available if you are willing to pay for them.

  3. @Monique: I'm sure they have the databases, but I wouldn't expect the DCCC (Democratic Congressional Campaign Committee, for those who don't know) to provide that sort of information publicly. Their purpose is electing Democrats to Congress, not providing the general public with election information.

    This may sound a touch more cynical than I mean it to sound, but I'd especially not expect them to hand over that sort of infomation for the purposes of exposing the growth of gerrymandering over time. The practice of gerrymandering works to the advantage of that organization, if not to the advantage of individual citizens who are Democrats.

    @Arsen: My price point for this information is "free", though I fear I may eventually have to double that offer.

    Really, I'd expect that some organization advocating clean politics and reform might have it handy. If not, I can compile it, but it might take a few days work.