There have been a lot of claims recently about the impact of redistricting on the 2012 congressional elections. Progressives are alarmed that Democrats won a majority of the House vote—roughly 51%—while falling a full 17 seats short of a majority. Such a discrepancy between the winner by votes and the winner by seats is rare, so it’s natural to assume that Republican gerrymandering—the process of drawing districts to advantage one interest over others—might be the culprit.
Neuroscientist and election forecaster Sam Wang recently added fuel to the fire, calling the 2012 outcome “The Great Gerrymander.” He identified 10 states, most of them controlled by Republicans, as notable and egregious deviations from a fair outcome, suggesting that gerrymandering cost the Democrats 15 seats in the current House of Representatives and calling for redistricting reform to fix the problem. Wang’s conclusion resembles that of political scientist Nicholas Goedert, who suggests that the 2012 maps cost the Democrats 9 seats.
Is this right? Has gerrymandering allowed Republicans to defy the will of the people? The crucial question to ask when deciding whether redistricting “mattered” is: compared to what? What is the alternative set of districts—the “counterfactual”—to which you’re comparing the current districts? Once we consider some other alternatives, these claims about gerrymandering aren’t as strong as they first appear.
2012 compared to the 2010 Districts
What if we “re-run” the 2012 House election, but using the old districts? We have done that simulation, using the 2008 presidential vote in both the old and new districts to capture how the redistricting might have moved partisans around. If we assume that nothing else affects House election outcomes but the partisanship of the districts—in other words, if we allow redistricting to have its maximum possible effect—we find that the 2011 redistricting cost Democrats 7 seats in 2012. This is not nothing, but it’s far less than what the Democrats needed to take back the House and about half what Wang estimated.
The effect is even smaller if we incorporate other important factors. Incumbency is the most important of these: lots of Republicans who were running as challengers or in open seats in 2010—and then won—ran as incumbents for the first time in 2012. We know that incumbency is a powerful factor in House elections, bringing candidates greater visibility, adding to their campaign coffers, and deterring quality challengers from running. On average, an incumbent in 2012 ran five percentage points ahead of a non-incumbent candidate from the same party in a similar seat. Sixty-one seats were were decided by less than this margin.
More important, once we took incumbency into account, the apparent effect of gerrymandering vanished. That is, the ability of Republicans to retain the House majority may have been due to incumbency advantage, not new and more favorable districts.
We went a step further and subtracted out our estimate of the incumbency advantage to simulate what 2012 would have looked like if this advantage had not existed. In this simulation, the Democrats won 219 seats—virtually eliminating the discrepancy between votes and seats in this election.
To be sure, separating out the effects of new districts and the effects of incumbency is not easy. The new district boundaries could also have factored into incumbency advantage—deterring potential opponents by making some districts more hostile territory in ways not captured by the district’s presidential vote. So we cannot conclude from this analysis that gerrymandering had no effect whatsoever.
But estimates that at least wrestle with this incumbency effect are going to be more accurate than ones that don’t. And we are struck by how rarely some analysts—including Wang—explore alternative explanations in this way.
2012 compared to history
Perhaps the pre-2011 maps are not the right standard. In fact, there is evidence they were already biased toward Republicans. The question is whether that advantage is a product of redistricting. In turns out when we go back further in time across multiple redistricting cycles, House elections have tended to favor Republicans for at least a couple decades. Once we put 2012 in this historical context, it does not stand out as a “great gerrymander” at all.
There are various ways to capture bias, but the one perhaps best suited to this discussion is what political scientists have called “symmetry.” Symmetry is a hypothetical: What if the two parties both received the same share of the vote? Would they get the same share of seats? If not, which party would have an advantage? Because in 2012 the Democrats received a majority of the vote and a minority of the seats, the outcome was clearly not symmetric. The question is, why, and is that common?
The solid line on the graph below shows symmetry for House elections since 1952. The line can be read as the percentage of seats a party gets beyond what would be expected under a symmetric plan. The graph shows that Republican bias has been around since at least the early 1990s. Moreover, changes in bias do not coincide with years when most districts were redrawn (indicated with vertical dotted lines). Instead, the party with more incumbents tends to have an advantage in symmetry bias.
This incumbency advantage is even more evident if incumbency is statistically removed from elections during this period, in the same way as described above. See the dashed trend line in the graph. The Republican advantage now extends almost without interruption to the 1950s. (Andrew Gelman and Gary King first noted this phenomenon about twenty years ago.) The consistent bias since 1991 is especially striking since Republicans now control the redistricting process in far more states than they did then. Note also that the large pro-Republican shift that occurs between 2010 and 2012 disappears once incumbency is accounted for.
Another way to partisan advantage from redistricting is with the balance of “wasted votes” between the two parties. House elections are “first past the post,” which means a candidate need only win one more vote than the next guy to claim a seat in Congress. Any votes beyond that threshold are technically “wasted” in the sense that they don’t help anyone win anything. By the same logic, all the votes for the losing party are wasted because they didn’t produce a win. If these wasted voters are strong partisans, they can be moved to a different district to help the party win that seat. Therefore, a gerrymandering party usually tries to win its seats by a small but comfortable margin and force the opposition to win its seats by a landslide. The gerrymanderers win more seats with the same number of votes because they are winning more efficiently.
Like symmetry, wasted votes can be represented as the extra share of seats a party gets beyond a “fair” outcome—in this case, equal numbers of wasted votes for each party. Here is what that looks like for the same time period:
The pattern is very similar, except Democrats never quite get a consistent advantage from the 1960s through the 1980s, as the symmetry measure suggests. There has been a Republican advantage since the mid-1990s, and the change in 2011, though favoring Republicans, was modest by historical standards.
Why do Democrats have a somewhat chronic disadvantage in these graphs, especially in the last 20 years? Part of the reason is that Democratic votes are increasingly concentrated in urban areas where they are more likely to waste votes with large majorities. Jowei Chen and Jonathan Rodden have simulated thousands of redistricting plans in a handful of states and found Democrats generally do worse when districts are constrained to be compact (that is, as close to simple shapes like circles and squares as possible).
The Limits of Mapmakers
Of course, there are many other maps that could be drawn besides the current ones. Computers can automate this process, both in real-life redistricting and in simulations like that of Wang or Rodden and Chen. The important point to remember when looking at any such plan is this: redistricting doesn’t guarantee results.
The erratic swings in the two graphs above make clear that partisan advantage in House elections is a moving target no matter how it’s measured. Consider the graph below, which shows the seat share advantage from wasted votes under maps drawn by Republicans, by Democrats, and by bipartisan agreement or nonpartisan commissions. (Information about control of the redistricting process comes from Michael McDonald and Justin Levitt.)
The pattern in 2012 is pretty consistent with what Wang and others have seen: the Republican-drawn plans favor Republicans, the nonpartisan plans also favor Republicans but not as much, and the Democratic-drawn plans are largely fair. But two years earlier, it looks like Democrats had the egregious gerrymander, while Republican plans looked pretty good. Then in 2008 all three sets of plans favored Republicans. Of course, saying that the plans favored Republicans is not the same as saying that 2008 was a good year for them. It was obviously a great year for Democrats, but they didn’t win as many seats as they probably should have given that broader climate.
In short, the answer to the question, “What effect did partisan control have on partisan advantage?” depends on when you’re asking it.
We’ve written cautionary notes about redistricting several times in the past months. Simply raising the possibility that redistricting isn’t always as powerful or pernicious as its critics suggest sometimes leads people to conclude that we are “gerrymandering deniers” who think redistricting has no partisan consequences whatsoever.
That is not the case. The analysis above does not confirm the worst fears about the “great gerrymander” of 2012. But given the challenge of answering “compared to what?”, we would not argue that the 2011 redistricting gave the GOP no advantage whatsoever. Political science research on redistricting has confirmed that control of the line-drawing process does yield some benefits. The challenge is in estimating what those benefits are. We have tried to show that the answer is far more complicated, and that the magnitude of the redistricting effect is probably smaller than many have assumed.
Moreover, our argument should not be construed as a defense of how redistricting is currently done. Reforming redistricting by handing responsibility to independent commissions might produce more even-handed results. In fact, the evidence from California suggests the state’s new redistricting commission produced a fair plan that met the mandate of the law. We’d support using these commissions in more states.
Plus, the need for such commissions is arguably increasing. Gerrymanders will likely be more effective in the future because the partisanship of a district, independent of incumbency or any other factors, is a more important predictor of House elections than it was several decades ago. That makes redistricting a more powerful tool than before.
All the same, it’s important to calibrate our expectations about what redistricting can accomplish. There is a lot more than redistricting to House elections. Democrats are concentrated in urban areas. Incumbents still outperform their party’s presidential candidate. And the electorate can still change its mind, as the turmoil of the last decade has made clear. A party can want to gerrymander, but the best laid plans can still go awry.
Finally, we are struck by how rarely people acknowledge the important trade-offs in redistricting. David Butler and Bruce Cain’s book, Congressional Redistricting, makes this point: you can’t have it all. It is very difficult to achieve equal district populations, respect compactness and contiguity, respect communities of interest, avoid diluting minority voting strength, and create perfectly proportional representation or at least minimize seats-votes discrepancies. As Butler and Cain write:
…almost all the generally accepted principles of redistricting can come into conflict with each other.
So claims about partisan bias in the current set of districts, or claims about what would be true under various alternative maps, also depend on (often unstated) priorities about which standards are most important and what trade-offs we should make.
Correction: "The post originally misstated Nicholas Goedert's finding, saying that the 2012 maps cost the Democrats 14 seats. The correct figure is 9.