Here’s a video explainer on how to detect partisan gerrymandering:

To learn more, see gerrymander.princeton.edu. Help us with a donation!

Update, September 2018: I have expanded the analysis below to include three independent tests to detect gerrymandering – you can read more about them in the Stanford Law Review. The ideas are also described in the New York Times [February 2013] [December 2015] [August 2017]. Over at Election Law Blog and the American Prospect, we show that the trickle of extreme gerrymanders became a flood only after 2000.

A common objection to this analysis is that population density can account for asymmetry in outcomes. This belief is outdated. As of today, partisan gerrymandering effects have a larger effect on representation than population density.

The Washington Post Second Annual Wonky Awards are out. Wow, there I am for Best Election Modeler. Thank you!

(Mental note: If I ever meet the Worst Modeler recipient, Dick Morris, remember to avoid shaking hands. The whole particle-antiparticle thing. We’d annihilate or get stuck together. Either outcome is bad.)

>>>

As current negotiations over the [fiscal cliff] / [austerity bomb] make clear, rank-and-file Republicans in the House of Representatives are not receptive to the policy implications  of November’s election. As correctly pointed out by Nate Silver,  members of Congress are increasingly insulated by the increasing polarization of their districts. Ever-larger victory margins reflect ever-safer re-election races.

However, Silver has also restated a common belief. He states that partisan gerrymandering is a symmetric problem, i.e. both Democrats and Republicans do it. Although both sides are potentially motivated, only one side has taken redistricting to extremes. Recent changes in partisan gerrymandering constitute one of the major crises facing our system of government (link to Mann/Ornstein book, a fellow Wonky winner).

Today I give an analysis that pinpoints some exceptional – and asymmetric – aspects of this year’s Congressional redistricting. I base this on criteria I have developed for identifying when a political party has been disenfranchised in a particular state. I conclude that the antidemocratic balance of power in the incoming Congress is driven by just a handful of states.

In this and coming posts I will address the following topics:

• Part 1: Developing a “tell” for partisan gerrymandering, and evidence for partisan asymmetry.
• Part 2: An estimate of how many people have been disenfranchised.
• Part 2/3: Steps that would re-enfranchise voters by 2020 or sooner.

Let’s start with some simple but telling examples. A minimum condition for a “representative” outcome is that within a Congressional delegation, the party receiving more votes should end up with more Congressional seats.

As an example, consider Colorado. There, 51.4% of the two-party vote went to Republican candidates, and 4 out of 7 representatives will be Republicans. Colorado’s delegation therefore represents its partisans fairly. (As an aside, it is not required that vote-share and seat-share follow the same proportions. Typically, popular margins translate to proportionally larger seat margins. It’s like the Electoral College.)

In the November election, the following groups failed to meet this minimum representativeness criterion:

	D%vote	R%vote	D seats	R seats
Entire House	50.4%	49.6%	201	234
Arizona	45.6%	54.4%	5	4
Michigan	52.7%	47.3%	5	9
North Carolina	50.9%	49.1%	4	9
Pennsylvania	50.7%	49.3%	5	13
Wisconsin	50.8%	49.2%	3	5

In these five states – and in the nation as a whole – the partisan interests of voters are not being represented fairly. Details can be found in a piece by Griff Palmer and Michael Cooper in the New York Times.

Now I will show a way to generalize this point to all states, even in cases when the seat majority and popular-vote majority are out of whack even though they belong to the same party.

Among political scientists, it is often suggested that imbalances like this are not caused by partisan redistricting, but by other “structural” factors such as concentration of Democrats in urban areas. (This is not true. Let’s come back to that later.)

Instead, let’s ask a simple question. If a given state’s popular House vote were split into differently selected districts, what would its Congressional delegation look like?

We can do this by using all 435 House race outcomes. For a state X with N districts, calculate the total popular vote across all N districts. Now pick N races from around the country at random and add up their vote totals. If their vote total matches X’s actual popular vote within 0.5%, score it as a comparable simulation. Because this approach uses existing districts, it uses as a baseline the “structural” advantages that are present nationwide*. In other words, it’s a measure for distortions in representativeness that are specific to state X.

As an aside, note that my approach does not require the drawing of actual districts. Doing that properly requires professional redistricting software and much time. (Update: One disadvantage of my approach is that it includes in its baseline the shift in district partisan bias that happened in 2011 as a consequence of redistricting. So what I am calculating here does not include the across-the-board difference I showed in my pre-election analysis*.)

Here are 1000 “simulated delegations” for Pennsylvania, along with the actual outcome in red.

It is apparent that most possible redistrictings would have resulted in a more equitable Congressional delegation. For outcomes with the same popular-vote split (50.7% D, 49.3% R), 1000 simulations give a median result of 8 Democratic, 10 Republican seats (average, 8.3 D). The actual outcome was 5 Democratic, 13 Republican. In fact, only 1 of the 1000 simulations led to such a lopsided split. And indeed, Pennsylvania legislators are known to have gone to extremes to favor Republicans during redistricting.

How much structural imabalance is there here? In this case, the structural imbalance is 9-8.3=0.7 seats. Partisan gerrymandering added a further imbalance of 8.3-5=3.3 seats. In other words, gerrymandering’s contribution to Pennsylvania’s partisan outcome was about five times as large as the effect of overall structural advantages.

Here is a listing of top offenders for whom the partisan discrepancy was 1.0 seat or greater.

	D %vote	D sim	R sim	D seats	R seats	Discrepancy
Pennsylvania	50.7%	8.4	9.6	5	13	R+3.4
Texas	39.9%	9.4	26.6	12	24	D+2.6
Ohio	47.9%	6.5	9.5	4	12	R+2.5
North Carolina	50.9%	6.2	6.8	4	9	R+2.2
Michigan	52.7%	7.2	6.8	5	9	R+2.2
Arizona	45.6%	3.2	5.8	5	4	D+1.8
Virginia	49.0%	3.7	5.3	2	7	R+1.7
Illinois	55.4%	10.3	7.7	12	6	D+1.7
Indiana	45.8%	3.2	5.8	2	7	R+1.2
Net, all 9 states	48.5%	58.1	83.9	51	91	R+7.1
D-controlled						D+1.7
R-controlled						R+13.2

The left column is coded by which party controlled redistricting. In black are a court-ordered redistricting (TX) and a nonpartisan commission (AZ). Note that California did not make this list, despite the fact that their redistricting was the focus of a loosely-argued ProPublica article. Basically, California votes Democratic and has a Congressional delegation whose party composition reflects the fact accurately and fairly.

There are some simple lessons to take away from this.

• Republican-controlled redistricting led to a swing in margin of at least* 26 seats, almost as large as the 31-seat majority of the new Congress. Those actions created a new power reality in the House – or more accurately, retained the old power reality.
• In the states listed above, the net effect of both parties’ redistricting combined was R+11.5 seats. Putting all of this redistricting into nonpartisan commissions would lead to a swing of at least 23 seats. The resulting seat count would be 213 D, 222 R or even closer. It is possible that in the absence of partisan gerrymandering, control would have been within reach for the Democrats.

I will end today with a graph showing the total effect of Republican-controlled redistricting in six states: PA, OH, NC, MI, VA, and IN.

In the next post I’ll show how to convert these results to an estimate of disenfranchisement, and suggest possibilities for future elections and activism.

Postscript to political scientists: If this topic interests you, by all means be in touch. This is original research. I would welcome collaboration (and priority).

*Update, 7:57pm: By using this year’s vote totals, I am also counting in the baseline the overall shift in partisan voting index (PVI) that took place in this year’s redistricting. In other words, the baseline itself tilts Republican because gerrymandering is still in it – it’s just “scrambled” all over the country. This baseline corresponds to the calculations I did in October and November – that’s why the black points intersect the green line at >50%. The upshot is that the total effects of partisan redistricting are, overall, even larger than what I have highlighted today.