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

How much difference does partisan gerrymandering make? North Carolina and across the decades

May 5th, 2017, 2:58pm by Sam Wang

Today in the Los Angeles Times, Brian Remlinger and I explain partisan gerrymandering, and how many seats it may be worth. Based on our analysis, more seats are affected by partisan gerrymandering now than at any point in the last five cycles of redistricting. In 2017, over 70 seats are made uncompetitive, favoring both parties. The net effect is a change in the margin of about 15 Congressional seats, in a direction favoring Republicans. Considering that the outcome of Affordable Care Act repeal yesterday in the House was decided by 4 votes, the advantage from gerrymandering is highly consequential.

We also review what the Supreme Court could do in the coming term to limit partisan gerrymandering. Two cases are coming before them, from Wisconsin and from North Carolina. Here at Princeton we are developing standards and a framework for the Court’s use. Read about in the Stanford Law Review and check out our website, If you’re interested, perhaps join the effort!

Tags: Princeton · Redistricting

14 Comments so far ↓

  • 538 Refugee

    Republicans lost 20 of ‘their own’ on this vote so we can guess a ‘fair’ map would have been consequential if more districts were in play. Rumors of Kennedy’s retirement loom large?

  • James Franklin

    Is there a point where the gerrymandering could backfire? Like a huge wave election with Dems winning by 15 points nationally?

    • Sam Wang

      Well, at some point the levee could break. See this review of my past work, by Paul Krugman. Of course, there is some chance that it won’t happen. It depends on whether current conditions persist, or whether they dissipate or reverse, as they did in 2013-14.

  • Anthony

    Sam, I am again confused at why you are pursuing this in the courts after the election of Trump. After the appointment of Gorsuch and upcoming RBG/Kennedy retirements there will be little that can happen with gerrymandering in the Supreme Court. This needs to get resolved by wave elections (to overcome Republican gerrymandering bias) which will hopefully come 2018 and 2020 along with election reform (in the local/state level) legislatively.

    • Sam Wang

      To clear up your confusion, Ginsburg and Kennedy have not retired yet. They are expected to be on the Court when two key partisan-gerrymandering cases come up in the fall. This is in the LAT piece. For this reason, time is short. Lots to do.

      The alternative to action: sit around blogging, or commenting on blog posts.

  • Colin McAuliffe

    With regard to the mean-median test from the SLR article, I’ve found that one can juice the test up a bit and get more power in states that are not close to 50-50 partisanship if you use (mean-median)/1.4826*median absolute deviation as opposed to the standard deviation in the denominator. The factor of 1.4826 allows one to use the same asymptotic distribution for estimating the p-value as you would use with SD in the denominator. SD seems to overestimate the true dispersion of gerrymandered partisan states, where win percentages of the redistricting party are often tightly clustered (as you also point out in the SLR). This makes smaller differences in mean and median seem less significant.

  • Rand Wilcox

    Dr. Wang,

    Read your OP ED piece in the LA Times. Ordinarily I do not comment on the approach used by researchers, but you are working on such an important problem, was worried that some might dismiss some of your conclusions.

    Student’s t-test is not robust, for reasons pointed out in numerous journal articles. Part of the problem is that the population mean and variance are not robust for reasons summarized in the books listed below. The two-sample t-test is not even asymptotically correct under general conditions. There are other concerns, some of which have to do with skewed distributions. The obvious methods aimed at salvaging the t-test (e.g., test assumptions, transform data)
    are not supported by numerous journal articles. There are many ways of improving on the t-test.

    When I was young, had an interest in Bayesian methods. At the time I thought I understood non-normality and how to deal with it. I was wrong. Simply switching to a prior having a Student’s t distribution, or some other
    heavy-tailed distribution, does not suffice. But I don’t have a quick explanation. If you look at the mathematical foundation of modern robust methods, followed by the properties of robust methods developed in recent years, you will see why I have concerns. We know how to deal with robustness from a frequentist point of view, but it remains unclear how to do this well from a Bayesian point of view.

    I truly hope these comments are useful.

    Hampel, F. R., Ronchetti, E. M., Rousseeuw, P. J. & Stahel, W. A. (1986) Robust Statistics. New York: Wiley.

    Huber, P. J. & Ronchetti, E. (2009). Robust Statistics, 2nd Ed. New York: Wiley.

    Staudte, R. G. & Sheather, S. J. (1990). Robust Estimation and Testing. New York: Wiley.

    Wilcox, R. R. (2017). Introduction to Robust Estimation and Hypothesis Testing 4th Ed. San Diego: Academic Press.


    Rand Wilcox
    University of Southern California

    • Sam Wang

      Thank you for your thoughtful comments on the t-test. I know we came off as frequentists, but we’re not that bad! We didn’t have space to say so in that article, but the full test in Davis v. Bandemer (1986) is that of intents and effects. I see this as examining the legislative record for whether an offense could have occurred (establishing a prior) and then seeing if something cannot be explained by chance (the frequentist test). In that respect, we are not quite as naive as the op-ed made us sound.

      In regard to specific problems of the t-test, we are alert to pitfalls. Because it’s such a specific problem we are trying to solve, I think the proper approach is (a) to look at a ton of actual cases and see what tests work best, and (b) play around to see which tests can be gamed by future bad actors. If you have time, look at my SLR article, which lays out where I was a year ago.

  • David Swenson

    It has been over one hundred years since the membership in the House of Representatives has been permanently increased to 435. The population of the U.S. has more than tripled in that period of time. There should now be over 1,300 U.S. representatives. The original intent was to have one representative per 30,000 population. What has increased in the last hundred years is the number of congressional staff members. There are now over 8,000 members of the congressional staff. See for more detailed information on the size of the House. Increasing the number of congressional representatives would make gerrymandering moot. All this would take is an act of Congress to change the number of Representatives.

    • Sam Wang

      Actually, you have it backwards. Increasing the number of representatives might have merits, but it would make gerrymandering worse, not better.

      Partisan gerrymandering is made possible by having many districts. About 7 in a state is enough to allow it to occur. Increasing the number of districts makes the exercise easier. For example, in Wisconsin, the Assembly has 99 seats, and redistricters have had quite an easy time gaining a partisan advantage there. Basically, the more boundaries there are, the easier it is to commit bad acts.

    • 538 Refugee

      The capital would have to be moved to an area where they could actually have planned growth to accommodate that many more members. Space is already an issue. That would require having to build all new infrastructure. Public works or boondoggle? Would they get to keep their sports teams? Builders would be the winners. Established businesses would be the losers.

      You will always lose me at “The framers of the Constitution and the Bill of Rights intended “. Those 3/5 of a person, non-citizens would agree with me that this point shows the constitution was never a sacred document that we need to revere and avoid changes too.

      Personally I think the government has outgrown the DC area but adding more reps isn’t the reason. If we were to move it to say, Antarctica, maybe only the truly committed to public service would endure?

    • Lorem

      I don’t think having only those truly committed to public service endure is a model for good governance. I bet the Tea Partiers are among the most committed.

  • Chris Burns

    I am working on a ballot committee in Michigan. We are looking to put a nonpartisan redistricting commission on the 2018 ballot.

    Follow our progress or make a donation at:

Leave a Comment