Princeton Election Consortium

Here’s the new Congressional map for Pennsylvania, drawn by advisor to the court Nate Persily. I think it’s a landmark contribution to the understanding of partisan fairness, in one of the most politically heterogeneous states in the Union – “Pennsatucky.”

For anyone who’s drawn such maps, this particular one is a pleasure to look at. It does a great job at compliance. It splits 13 counties, less than any plan offered to the court. The only plan that splits fewer counties or municipalities is the “compact D gerrymander” that the Princeton Gerrymandering Project posted over the weekend. So in terms of basic compliance, it wins.

But more importantly, this map undoes the effects of population clustering. As we showed over the weekend, it is just possible within the Pennsylvania Supreme Court’s rules to allow a plan that reaches partisan balance, i.e. a 9 D, 9 R outcome for a 50-50 statewide vote. However, those rules bias the range of possibilities in favor of Republicans, so that a Republican gerrymander – one of the proposals on the table – was also a possibility.

It appears as if the advisor was either trying to achieve partisan balance or build competitive races (as opposed to, say, picking a plan that was in the middle of the range of possible maps). His map – and the one we gave over the weekend – show that even under the constraint of compactness and not splitting political jurisdictions, it is still possible to build a map that treats both parties approximately equally. Notably, he did so while keeping most population centers together. This map passes all three of our tests of partisan asymmetry.

This outcome fits with an unspoken principle in the court’s decision of partisan symmetry: if the two sides swapped vote totals, then their seat totals should be swapped as well. This fundamental principle of fairness underlies our proposed standards for detecting gerrymandering, as described in our amicus brief in the Whitford case.

It is commonly claimed that partisan bias arises naturally from the way that voters distribute themselves. Certainly such a bias could be visible if one drew thousands (or millions) of maps at random. But districting is not a random process. It is done by human beings, and reflects real-life political and social considerations. In that respect, we shouldn’t care about millions of maps; we should only care about one map, the one that is drawn. Persily has shown that it is possible to maintain all the old principles of geographic contiguity, and still create an overall map that does not give either party an undue advantage. In my view, this is a big deal because it shows that the ideal of partisan symmetry can be a primary criterion for drawing a map.

It seems clear that the Pennsylvania GOP made a serious tactical error. The governor and Democratic legislators had offered a plan that would have retained some GOP advantage. The new plan erases that advantage entirely. In our analysis, based on the 2016 vote the new map produces 5 Democratic districts, 7 Republican districts, and 6 tossup districts. The maximum likely performance by Democrats in 2018 would be 11 seats, a gain of 6 seats over their current representation. That gain would be one-fourth of the 24 seats they need to win control of the House of Representatives. Of course, some of those gains could also go away in a later year that was good for Republicans. That’s the point of electoral competition.

The following analysis is a team effort by the Princeton Gerrymandering Project. John O’Neill drew maps which he, Brian Remlinger, Will Adler, and Madeleine Parker analyzed. For other people’s plans, we cited the work of Brian Amos.

In a previous post, we evaluated the Pennsylvania GOP leaders’ proposal for redrawing the congressional map and found strong signs of partisanship. We got curious: under the limits set by the state Supreme Court, what is the maximum advantage in the other direction, favoring Democrats?

We did this in order to put all the proposals into context. If we know the extreme limit on how favorably a map could treat Democrats while following the Court’s directives, then we can evaluate the partisanship of the many Democratic proposals.

We find that the Court’s simple rules (compactness, equal population, and minimization of county/municipality splitting) may still allow Republicans to commit mischief, but they effectively limit Democratic opportunities to gerrymander. Interestingly, Democratic players didn’t go nearly as far as they could have, but instead drew plans that were at the middle of the range of possibilities.

Here’s a map that is impressively compact:
It is the most Democrat-favoring map we could draw under the court’s criteria. On average, it would elect 9 Democrats, 9 Republicans for an equally split statewide vote (using 2012/2016 presidential vote data), with 7 safe seats for each party and 4 tossup seats where the expected margin between parties would be 10 points or less. Democrats can get to 11 seats if they win all the tossups. That’s still 1-2 seats less lopsided than the Republican proposal in the other direction – and requires a wave election.

As far as literal compliance with “floor criteria” goes, this map splits 12 counties, 14 municipalities, and 11 precincts, equal to or fewer than all of the plans we examined. It also performs as well as or better than the other proposed maps on compactness metrics. There could well be a better map out there by these criteria, but we haven’t seen it.

This table contains an analysis of the map above (“Princeton D compact”) and other maps. Click to enlarge:

Democrats gain the biggest advantage by splitting Philadelphia and Pittsburgh. This is the converse of what we saw in the Republican plan, which packed those cities tightly while splitting other communities such as Reading, Montgomery County, and greater Harrisburg.

It’s possible to squeeze out a little more performance by being less attentive to the compactness metrics. The following map (“Princeton D extreme” in the table above) still makes fewer splits than all the submitted plans. Compared with the top map, it converts one safe Republican seat to a toss-up. It’s still a 9-9 map, but its ceiling performance is 12 Democrats, 6 Republicans.

Assuming that the Republican legislative leaders’ proposed map of February 9 was as favorable to their party as possible, this means that under any nominally Court-compliant map, Republicans will always win at least 6 seats and Democrats will always win at least 6 seats. The remaining 6 seats — a full one-third of them — are limited not by geography, but by the whims of redistricters.

Interestingly, none of the remedial plans submitted to the Court, even those submitted by Democratic politicians, sought to maximize Democratic seats. All submitted plans had between 6 and 8 Democratic seats, with most plans settling on 7 Democratic seats. Basically, everybody but the Republicans played it down the middle.

Why didn’t Democrats go for broke? Maybe they think toss-up seats are flippable in 2018. Maybe they wanted to protect their five incumbents. Maybe they prioritized “softer” criteria like maintaining certain political or economic communities. Or maybe they are simply good-faith actors. Whether the Court picks one of their plans or draws its own, the outcome is highly likely to be more fair than the existing plan.

One lesson here is that the median neutrally-drawn map appears to be about tilted two seats toward the Republicans. This cycle, partisan gerrymandering added two seats on top of that.

Back to our maps…to be clear, we don’t believe that our maps are good simply because they perform well on the metrics. Looking at our map one way, it’s the extreme endpoint of a range of possibilities. Looking at it another way, it shows that it’s possible to attain representational equity even in a state as heterogeneous as Pennsylvania.

Whichever way you look at it, we have shown that the simple rules and traditional redistricting criteria used here by the Court can only do so much to constrain partisan actors. The court’s floor criteria of compactness and no splitting can be useful to prevent the flagrant geographic offenses we saw on both sides of the aisle nationwide in 2011. However, a truly fair process can only result from a nonpartisan process that examines the many tradeoffs in redistricting.

The fine print:

We used 2016 2-way presidential vote to categorize districts as Democratic or Republican. 2016 was probably the high-water mark for Republicans in Pennsylvania, so our map (and the other maps) may be slightly more favorable to Democrats than the data suggests. We did not examine incumbency, nor did we attempt to keep communities together beyond counties or municipalities. The plans above were drawn by John O’Neill, an expert with real life redistricting experience. We can’t rule out that there is some plan much more advantageous to Democrats than what he drew, though we suspect not. A shapefile of our map is available here, and summary statistics of our plan, as well as some others, are here.

We examined the following maps, which are available at the PA Supreme Court website. The analysis here taken from Brian Amos on Twitter, unless otherwise noted. All analysis uses 2016 presidential results:

As the old political-science saying goes, “all redistricting is gerrymandering.” Inevitably, choices and tradeoffs have to be made. When it comes to representation, the resulting map can end up balanced – unless partisan advantage is made the primary criterion.

The Pennsylvania Supreme Court set down clear “floor criteria” for what was absolutely required in the revised map. They said that counties, cities, and wards should not be divided, except to achieve equal population as required by federal law, and to comply with the Voting Rights Act. However, they also said (see page 124 of the decision):

We recognize…that there exists the possibility that advances in map drawing technology and analytical software can potentially allow mapmakers, in the future, to engineer congressional districting maps, which, although minimally comporting with these neutral “floor” criteria, nevertheless operate to unfairly dilute the power of a particular group’s vote for a congressional representative.

Although compliant with floor criteria, a close look at the Pennsylvania GOP legislative leaders’ proposed map reveals some of the choices they made. All it takes is an overlay of population density. Here are some examples. [Read more →]

Mathematicians have been working hard to create ways to measure #Gerrymandering. 2017 was a big year on this: accumulating evidence showed PA most gerrymandered of all by two measures, among five worst by any measure. We all lose our voice when our votes don't count. pic.twitter.com/5fgCxTc00r

— LWV Pennsylvania (@LWVPA) February 13, 2018

Today, Philly.com highlights multiple measures of partisan gerrymandering, including several developed here at Princeton.

Now, a major disclaimer: I didn’t think of these, exactly. Several tests (mean-median difference, and lopsided-wins) are over a hundred years old. Another (simulated elections) relies on an equally-old technique, Monte Carlo simulations. These tests are so old that they have whiskers.

There’s one lesson, though: there isn’t just one way to evaluate a gerrymander. Think of these as tools that capture different aspects of partisan asymmetry. For example, Monte Carlo simulations actually account for some of the clustering effect that comes from Republicans gravitating toward rural areas and Democrats gravitating toward population centers.

For more on the tests, see our explainer. If you like, support our work!

An image of the Pennsylvania legislature’s proposed Congressional map has been released. It appears to be nominally in compliance with the Pennsylvania Supreme Court’s order, since it splits fewer counties than the current map. However, as we have seen in North Carolina, a prettier map can still conceal ill intent.

Despite the low resolution, enough information is available to conclude that this is still a partisan gerrymander. Of course one can do a detailed partisan calculation, and Dave Wasserman thinks that it’s still 12 R, 6 D, and Nate Cohn thinks it’s 13 R, 5 D. It is absolutely central to remember that even under compactness-based rules, a wide range of partisan outcomes is possible – and this plan is extreme.

Even without making such seat-based calculations, three clues indicating a gerrymander are apparent: [Read more →]

Today, U.S. Supreme Court Justice Samuel Alito denied Republican legislators’ request for a stay of the Pennsylvania Supreme Court’s redistricting decision. That decision overturned the state’s partisan gerrymander, and remains in place. Therefore the drawing of new maps will go forward.

Here’s a quick rundown of what the decision does – and doesn’t – mean. [Read more →]