For starters, Trende acknowledges that gerrymandering is a real factor, and that it helped House Republicans to win more seats than one would expect based on popular vote, as the median district moved further to the right in the latest round of redistricting. However, Trende points out that there may be bigger factors at work, and that they originate with voter behavior, not with the cartographer's pen. For starters, there's the matter of increased geographical self-selection, with likely Democratic voters packing themselves more and more into urban areas; they aren't distributed efficiently, which makes it difficult to draw a Democratic-friendly map.
And perhaps most significantly, there's the problem of the decline in ticket-splitting. No doubt you've noticed how few House Dems are left in districts that Romney won, and vice versa how few GOPers are left in Obama districts, whereas in previous decades it wasn't terribly unusual to find Blue Dogs in red rural districts or (though more rarely) Rockefeller Republicans in blue suburban districts.
Trende demonstrates that gradual change through a variety of interesting tables. The overall number of "Highly Partisan Districts" has increased, while at the same time, the ability of a person from the "wrong" party to get elected in one of the HPDs has dwindled even more dramatically.
Take districts that went for Republican presidential candidates by 20 points more than the national average, for example (so-called "R+20" seats, using the parlance of the Cook Partisan Voting Index): In 1992, 15 percent of those seats were held by Democrats, while today it's only 3 percent. Nine percent of R+15-19 districts were held by Dems in 1992, compared with 2 percent today, while among R+10-14 districts, Dems held 18 percent in 1992 but 2 percent today.
If Democrats were able to win seats across each of these tiers at the same ratio as they did 20 years ago, they would comfortably control the House. Since there are more HPDs that are Republican than Democratic (and because the median seat is now around R+2), you can see how that decline in ticket-splitting hampers Democrats.
More over the fold ...
Data viz whiz Adam Carstens took some of the data from Trende's piece and turned it into a fascinating graph (reproduced above) that brings the disparity into sharp relief, but also may undercut some of Trende's points. For starters, those Dem declines in dark-red seats stand out less when you see how many fewer dark red seats there were in 1992. The same goes for dark blue seats too, which are also much more numerous today. In 1992, around half of all seats fell in the generally competitive range (D+5 to R+5), while now it's only around one third.
Of course, that only serves to emphasize how important the percentage of seats won by Dems in those swing seats is, and that varies greatly depending on the way the national wind is blowing. After a Dem wave, it's high (like 63 percent after 2008, in the center-most D+2-3 cluster), after a GOP wave it's very low (28 percent in that same tier after 2010).
Carstens' graph is kind of a Rorschach test for whether or not you're in the "gerrymandering bad" or "gerrymandering meh" camp (or if you're on the fence, like me). On the first few glances at it, it looked to me like a very gradual decline: With each passing cycle, the number of swing seats ticked down slightly.
That would tend to support the "Big Sort" hypothesis, about not just geographical mobility, but also more conformity among those already there: some Americans increasingly joining their neighbors in voting Republican in rural or exurban areas, while others increasingly join their neighbors in suburban and urban areas in voting Democratic. But on the third or fourth pass, I noticed how much the number of medium-red seats (especially in the R+8-9 range) jumped between 2010 and 2012. What happened between those two years? Redistricting, of course.
There's one more related visualization that you should check out, from Drew Linzer. (I'm not sure if it's in response to Trende's piece, but it's certainly on point.) It focuses purely on the shift from 2010 to 2012, and it shows the Republican percentage of each congressional district's vote:
So Linzer's histograms show basically the same thing as the 2010 to 2012 shift in Carstens' map, where the bulge around R+8-9 (in other words, the upper-50 percent range for Romney) grew much fatter. Now, that growth in medium-red districts, in itself, doesn't cause polarization. But connect the dots: Thanks to the decline in ticket-splitting, a district that went 58 percent Republican still had a good shot at electing a Blue Dog in 1992 (or Boll Weevil, since the term "Blue Dog" hadn't been invented at that point), but today, it's very likely to elect only a fire-breathing Republican.
If you see how increasingly sophisticated computer-aided gerrymandering, self-sorting, and declining ticket-splitting all interact and feed on each other, then you're approaching a full-bodied theory on how polarization is increasing.