Princeton Election Consortium

Innovations in democracy since 2004

Outcome: Biden 306 EV (D+1.2% from toss-up), Senate 50 D (D+1.0%)
Nov 3 polls: Biden 342 EV (D+5.3%), Senate 50-55 D (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

A New Project on Partisan Gerrymandering

January 17th, 2017, 9:09am by Sam Wang

This is a big year for partisan gerrymandering. Recently, star litigator Paul M. Smith has cleared the decks for voting-rights cases in the courts. That’s just one move of many that assures that voting rights will be in the spotlight in the coming Supreme Court term.

The effects of partisan gerrymandering are plain in the graph above. Up until and including the election of 2010, seats the U.S. House were related to the national vote as indicated by the shaded gray zone. The redistricting of 2010 led to a jump of about a dozen seats away from recent historical trends. The suddenness of this change, along with my statistical analysis (Stanford Law Review) reveals how this jump arose from partisan redistricting efforts in a handful of states. The jump comes from the fact that more advantage was gained by one side (NC, PA, OH, MI, VA) than the other (IL, MD). This net change can vary by decade, and depends on who controls the legislative process.

Today, I am pleased to announce that starting in 2017, I will take my work on partisan gerrymandering to a new level. I am now looking for full-time help for the next one to two years.

As many readers know, I have developed simple statistical standards to define partisan gerrymandering. These standards are designed to be consistent with existing Supreme Court precedent, and avoid statutory and Constitutional landmines that other standards may hit. Representationally, this project may lead to a net change of dozens of House and other legislative seats. Cases are now percolating through courts in Maryland (a gerrymander that benefits Democrats) and Wisconsin and North Carolina (benefiting Republicans). Without getting into the details, I will say that our work will be unique, and is highly likely to be deployed as an argument in these cases.

My proposed standards have been published in the Stanford Law Review and the Election Law Journal, and have won a prize from Common Cause. If adopted, the tests will level the playing field between the two parties.

I am now recruiting a Statistical Research Assistant to analyze elections and redistricting. The term of appointment is one year, renewable for a second year.

Together, we will:

  • Apply the statistical analysis to Congress and state legislatures, 1900-now, to identify which states and parties benefited, and to compare this with patterns of legislative control;
  • Compare the results with map-based methods and other newer standards such as the efficiency gap; and
  • Assist in the preparation of reports, in-depth analysis, and possible peer-reviewed publications.

This position is very suitable for someone who is between college and graduate school. More experienced applicants are also welcome. In all cases, the person must be available full-time. Progress on this project will drive practical impacts and original publications. The ideal candidate will have experience in statistics, have some experience with MATLAB and Python, and be able to communicate clearly and accurately with non-statistician audiences. He/she will learn to use Maptitude for Redistricting.

If time and expertise permit, we will also make the site more user-friendly. For this work, I am looking for a second person with HTML, CSS, JavaScript, and Python abilities. This may be a contract or shared-time position.

Soon there will be an official job posting, at which time the goals and requirements may be modified slightly. In advance of that, I invite interested people to send me an email at and describe their qualifications and availability.

Tags: Redistricting

30 Comments so far ↓

  • Tim in CA

    Sam, this is very exciting work, and it is ultimately of much greater importance than forecasting election probabilities based on polling aggregation. I wish you abundant success in your endeavors. Hopefully these cases make it all the way to the SCOTUS, resulting in a uniform standard that prohibits partisan gerrymandering in the states. Best wishes,

    • Sam Wang

      Thank you. You’re right. Poll aggregation was pioneering in 2004, but it is now close to a form of entertainment. It’s only useful to the extent that it focuses activist efforts. And we’re far from the next election.

      In contrast, the gerrymandering project may improve the functioning of U.S. democracy, fairly soon.

    • LondonYoung

      In defense of Sam’s poll aggregation:
      If memory serves, this project had its origins in Sam speculating back in 2000 (or 04?) that PA, OH and FL were the big swingy states, and winning two out of three won the presidency.
      For 2016 Sam’s “Power of One Vote” tab shows MI, PA and FL taking on that role.
      But if we look at the approach of both campaigns going through Sam’s “power” list state by state …
      … the only difference between the campaigns is that Trump hit two states on Sam’s list that Clinton didn’t: MI and WI.
      The states that Clinton hit that Trump didn’t don’t appear on Sam’s list.
      Merely using Sam’s algo’s to allocate campaign effort might have made all the difference.

    • Anthony


      Maybe, maybe not. Clinton hit PA hard and still lost it. I’m dubious as to whether campaigning matters much at all, with as Sam routinely points out with how partisan voters are votes may have been locked up as soon as the campaign started.

    • Anthony


      Why are you so confident in solving partisan gerrymandering using the SCOTUS when Trump is about to get at least 1 appointment over the next 4 years and potentially more? If Trumps gets 2 young partisan conservatives on the court in his term there is no chance either money in politics or partisan gerrymandering gets fixed using the SCOTUS over the next 40 years at minimum.

    • Sam Wang

      Not confident, but there is a fighting chance. Kennedy + the liberal wing. See Kennedy’s opinion in the LULAC case, in which he says that partisan symmetry is an interesting standard, but there’s not a specific test..yet. Also read my SLR article introduction. It’s all laid out.

  • Marc

    I agree, this is one of the top priorities. Are you accepting donations to help fund the work?

  • Brian

    This is important work! But, I’m not sure I find the graph above convincing. A few comments/thoughts:

    What does it look like if you just show “Winning two-party national vote share” and keep everything in the upper right quadrant. I’m not convinced 2014 will appear as an outlier. Likewise, why not graph the median curve between vote share and representation. There are points in your gray zone where apparently Democrats got 50% (or 51.2%) of the vote and 54% (or 56%) of the seats. Should these be acknowledged as an inequity in the past? Is there another explanation?
    In short: to be effective, this effort needs to be shown as non-partisan as possible. And, I think there needs to be more evidence that this is a “new” problem to be solved.

  • Jen S

    Hi Sam:

    I’m working on non-partisan redistricting in Michigan, and I just wanted to thank you for all the work you’ve done! It’s been a valuable resource (which is why I’m here now!).

    Am I correct in assuming this position would be closer to Princeton, or could work be done remotely? I’m employed but I’m also spending an incredible amount of time working on this – it’s incredibly important, and a labor of love for me in my own state – so figured I’d ask.

    Thank you again!

    • Sam Wang

      Jen, hi. How great that you are doing that.

      Yes, I have to get somebody close to here. The work will go far faster if I am face-to-face with the person often. There is a lot to do in the coming 1-2 years.

      Good luck with it – Michigan is important.

  • taney

    I want to help crack the gerrymandering nut. I read the SLR article – what can I do next?

  • ArcticStones

    Glad to see you put your important work on gerrymandering on the front page again. Today I am astonished to read this as a guest post on Professor Rick Hasen’s blog:

    Full paper:

    Professor Wang, this seems very much at odds with your own work. I hope you will offer a critique.

    • LondonYoung

      Sam discusses the main difference in section IIB of his SLR paper, mainly with section 2 on page 1303. If you like the assumptions in this section you will not like the Michigan paper.

      To put it a different way, consider Utah and Nebraska. Decide if you think whether each is gerrymandered, then run on them for 2014. This illustrates the problems in gerrymander identification.

      IMHO, if you don’t think one of these two states has been gerrymandered, then you have no soul. But I don’t have a strong opinion on which one it is!

  • bks

    [Completely off-topic.] Trump’s science adviser is likely to be Princeton physicist William Happer

  • Joseph Elfelt

    I am a software developer in the field of online maps. The link below will display a Google + GIS map of all congressional districts with highlighting on Pennsylvania’s 16th congressional district.

    Map link:,-76.291483&z=9&t=h&congress=PA,16

    The map is displayed by Gmap4. I am the developer of this enhanced Google map viewer. You, and everyone else, are welcome to use Gmap4 map links to help educate people about gerrymandering. However, due to a restriction imposed by Google I request that people not embed a ‘live’ map on a website.

    Anyone (no tech savvy required!) can make similar Gmap4 map links to highlight any congressional district when the map opens.

    For more info, including more example map links, please click “Map Tips” in the upper left corner.

  • poorlando

    California has a top two primary that resulted in a few races where you had to choose between two Democrats. This would mean that if you were a Republican who wanted some say in the race, you had to vote for a Democrat, which would lead to misleading results when trying to measure Democrat’s national support by simply counting how many votes they got. How do you account for this in your research?

    • Sam Wang

      The usual approach is to assume those voters would split 75-25 or 80-20 for the majority party. See pages 1292 and 1304 of my SLR article.

  • John Maerz

    I am very serious about the issue of gerrymandering and encouraged by your efforts. Nonetheless, I have a question about your figure at the top of the page and your statement, “Up until and including the election of 2010, seats the U.S. House were related to the national vote as indicated by the shaded gray zone. ”

    What do you mean “were related”. You centered the distribution of points pre 2012 rather than square the figure to have proportionate axes. When I fit a 1:1 line through your graph, it appears that while the last three elections have yield over-representiation by GOP, there has been significant historic over-representation by Dems (most points are well above the 1:1 line]. In fact 24 of your points occur above the 1:1 line (favor Dems), while only 9 points occur below the lien (favor GOP). What accounts for those patterns? Was that the result of gerrymandering by the Dems or some other factors? There is little doubt that the overrepresentation has shifted in favor of GOP since 2010, but I would like to know more about longer historical context. Thank you for your work.

    • LondonYoung

      Perhaps a plot of “majority party seat share” (always at least 50 pct) vs. “majority party vote share” would be more clear. Such a plot would land 2014 smack dab in the middle of the distribution, but 2012 and 2016 would still be interesting.

  • pquant

    I have a probably somewhat naive question: in the graph above, why is the general slope so far from one? is it just the result of the winner-takes-all system or is there more to it?

    • Sam Wang


      To put it technically: it’s a function of the distribution of partisan preference. Imagine a bell-shaped curve indicating the estimated %D vote in each district. The predicted slope of that graph is the integral of that curve – a sigmoid curve, where the steepest part of the curve determines the slope of the graph shown.

    • pquant

      Interesting. I assume the distribution cannot be exactly Gaussian to get this?

  • LondonYoung

    Sam – on your gerrymandering tool – any chance you can add Cook PVI (or, really, just presidential election results) to Step 3?
    And if you were super inclined, add that to Step 1 as well (though your tool does allow this to be done manually).
    This would solve the noncompetitive district problem as well as take out candidate issues in some of the races.

    • LondonYoung

      (I was just looking at the special election coming up in George 6. Price won it 62 to 38, but Trump only won it 48 to 47. I haven’t seen that mentioned often in the news …)

  • Philip Cerruti

    I ran across this interesting article on Vox:

    Is this similar to what you are doing?

Leave a Comment