Princeton Election Consortium

A first draft of electoral history. Since 2004

What would it take for the House to flip?

August 12th, 2016, 8:00am by Sam Wang


As I wrote recently, the Speaker of the House, Paul Ryan (R), has raised the possibility that his party could lose its majority in November. This fear can account for a variety of recent political events, including the entry into the presidential race of a total unknown who just happens to be the policy director of the House Republican caucus.

To take control of the House, Democrats need to gain at least 30 seats over their current 188 seats. This is a tall order, but not impossible. A change of this size has happened in two out of the last five Congressional elections: a 31-seat gain for Democrats in 2006, and a 63-seat gain by Republicans in 2010. In the modern era of polarization, 1994-2014, a change of that size has happened in 3 out of 11 elections. Since 1946, it has happened in 10 out of 35 elections.

What would it take for the House to change hands?

One of the best ways to answer this question is not yet fully available. That best way is to look at district-by-district ratings by prognosticators such as David Wasserman of the Cook Political Report or Kyle Kondik at Sabato’s Crystal Ball, and ask what the average seat gain will be. I have written about this kind of information in the past (for example, it was dead-on in 2008). Those sites both currently give maximum plausible Democratic outcomes (i.e. certain/probable Democratic plus toss-ups) as 209 seats, short of the necessary 218 seats. However, the ratings take some time to be updated in response to events in the districts; after all, there are 435 of them, and it takes time for Wasserman/Kondik/their collaborators to collect and interpret incoming information. As accurate as this approach is, it gives a slow-moving picture that will surely change in the coming three months.

In the meantime, it is possible to get some feeling for what would have to happen at a national level. I will start off with an approximate question: what would the total popular vote for Congress need to be for Democrats to end up in control of 218 seats? And where is national sentiment relative to that threshold?

My tool for asking this quesiton is the general Congressional ballot. In the last six cycles, the measure at the end of October has come within 3 percentage points of the actual national vote. In addition, we have a second validation. As part of modern polarization, from 1992 to 2012 the generic Congressional has come within an average of 2.9% of the Presidential popular vote. The Clinton-over-Trump margin is currently Clinton +6.0+/-0.8% (median+/-estimated SEM, n=8) and the generic Congressional ballot is at Democrats +6%. So everything is aligned, up and down the ticket.

Now let’s take a look at the relationship between vote share and seat share since World War II:This relationship follows a broad relationship that is steeper than proportional. It contains many factors that you may have read about: incumbency and the relative packing of Democrats into cities, to name two. Those combined effects influence the location of the gray polygon, which circumscribes all elections from 1946 to 2010.

However, note that the most recent two elections in 2012 and 2014 fall outside the polygon. In these elections, which were unusually favorable to Republicans, something new happened: an organized wave of partisan gerrymandering. As I have analyzed in detail, the added advantage from post-2010 partisan redistricting amounts to about a dozen more Republican seats than what would occur under neutral districting principles. For those who are interested, the full story is detailed in a breezy but informative book, Ratf***ed by David Daley.

Yesterday, I used the generic Congressional preference poll and national Presidential polls to estimate that if the election were held today, House Democratic candidates would win the popular vote by 5-8%, indicated by the curly bracket in the graph. Judging from the last few cycles, that level of public opinion appears to be right on the edge of being enough to give Democrats control of the House.

Note that on the generic Congressional graph, I have displayed a threshold of Democrats +4.5% 7%. Revised – I may revise that threshold in the weeks ahead – informed comments are welcome.

Kyle Kondik takes a more extreme view, and estimates that Democrats need a +10% win. If true, that would be a serious deviation from 1946-2012 trends. There is a research finding from the 1990s asserting that the effects of gerrymandering can fade after a few cycles. Whether Kyle is right or I am, it is probably time to revisit that old finding – for instance with newer statistical tools.

So even though the general Congressional ballot is not as good as a district-by-district analysis, for now it is an up-to-the-minute tracker, and maybe the best measure we will get until October.

Tags: 2016 Election · House

45 Comments so far ↓

  • Steve Garcia

    I love this stuff. I am not so sure that this angle is correct, though, Sam:
    “I will start off with an approximate question: what would the total popular vote for Congress need to be for Democrats to end up in control of 218 seats?”
    Forgive me for framing it another way. I see the encouraged Dems voting in numbers comparable to 2012 and the discouraged GOP voting more in terms of off-year 2014 turnout. The GOP dropoff in votes for the House was 18 million from 2012 to 2014. Only 39 million, vs 57 million. THAT may be the discouraged level of GOP fall-off. The Dems will get close to or exceeding the 69 million Obama got in 2008, no less than the 65 million of 2012.

  • DougH

    Assuming the Republican gerrymandering after the 2010 census produces a constant offset below the historical polygon is fine at the low end of Democratic margin. But due to the “dummymander” tipping point, this effect should be reversed at the high end of the vote margin-that is we should expect the Democratic seat share to be above the gray polygon for vote margins in the high 50′s (except that DWS did the opposite of HD at the DNC and did not challenge enough incumbents).

    The effect of gerrymandering should produce a curve similar to a hysteresis loop, with a steep slope in the middle (the back side of the hysteresis loop would represent the reverse swing of the pendulum when the Democrats overplay their hand). So it’s difficult to predict where this steep slope would cross the magic 50% threshold to flip control of the House, but given Trump’s implosion and it’s unknown effect on down-ballot races, it’s certainly possible that threshold will be crossed this cycle, especially given the dilution of the Republican base due to demographic headwinds amplified by Trump’s narrow appeal (i.e. actuarial attrition not being made up with new immigrant and younger voters).

    Since gerrymandering is a state-by-state thing, the tipping point will be different in each state. It will depend on how aggressive the gerrymandering effort was (in protecting too many incumbents and/or spreading the margin too thin) and on the demographic shifts in each state since the 2010 census. It will also be affected by how well Trump does in each state, how that affects turnout, and how the voters in each state split their ballots.

  • Ravilyn Sanders

    Gerrymandering is based on detailed results of previous elections at precinct level, assuming the general pattern holds. It reduces the winning margin per seat, while increasing the number of winning seats.

    If a large demographic change fundamentally, lot more seats would flip. Republicans had 20 point advantage among white voters. Trump seems to have lost it among college educated whites (60 % of whites) and women (51% of total). Underperforming on such large groups would create a wave election. Republicans also had enthusiasm gap in their favor, that also was baked into the Red Maps.

    Gerrymandering raises the barrier, but once the levee is overtopped, you are looking at New Orelans after Katrina.

    • counsellorben

      Love the levee analogy. Especially since I used it, too. Props to you for using it first. PEC commenters seem to be on the same wavelength this morning.

  • Olav Grinde

    Sam, but could you explain why you increased the threshold of your Generic Congressional Poll from 4.5 % to 7.0 %? That’s quite a jump! (Apologies if I missed it.)

    Is this threshold equivalent to the Republican gerrymandering advantage – or should the two not be confused?

    Does your adjustment of the threshold mean you have revisited (corrected) you calculation of the gerrymandering advantage?

    • Josh

      Sam addressed this to a degree either in this thread on in another. Think about the distance between the 2012 and 2014 dots on the graph and the edge of the polygon; that gives you a rough estimate of the effects of gerrymandering.

      Then look at where, if you subtract that distance from the polygon’s edge, you first cross the D = 50% of seats mark–that’s basically the vote share D’s would need to get to flip the house. On the graph it looks to be about 53-54%, or a margin of 6-8%. Split the difference and you have 7%.

      I’m sure Sam has a much more mathematically sound explanation for how he arrived at his answer but to me this is a good quick and dirty way of thinking about it.

  • Amitabh Lath

    Wow that scatter plot is eye opening. I wouldn’t have expected such an obvious correlation given gerrymandering, ticket splitting, etc.

    I have been scanning local news to check the Congressional races. Scott Garrett (NJ-5th), Leonard Lance (NJ-7th), Rodney Frelinghuysen (NJ-11th) are all being asked to defend Trump. Generally these are safe red seats but this year there seems to be some doubt in the NJ media. NJ-5th even had some polling showing Garrett in trouble (but partisan polls).

    A study of a dozen or so randomly chosen safe or semi-safe Republican seats should give us some idea if there is something new this year or it’s business as usual.

  • Michael Levinsohn

    It seems Trump, at best, has a “puncher’s chance.” Is there historical precedent for a Septemer-October opening for a late round knock-out that may come from a Clunton email or a Clinton Foundation revelation?

    • Ed Wittens Cat

      Trump is betting on more terror– its one reason why hes adamant abt welding Obama FP to ISIS
      He may be right– odds of more terror in the next 90 days is pretty high.
      http://www.nytimes.com/2016/07/27/science/mass-killings-contagion-copycat.html?smprod=nytcore-iphone&smid=nytcore-iphone-share&_r=2

    • Ed Wittens Cat

      In a jury trial, this is called catalepis, meaning “to be seized”.
      Lawyers introduce “sticky” accusations, knowing they will be overruled, but will remain in the jurors memory.
      My pet scenario for an October surprise wud be a dirty bomb.
      http://www.theatlantic.com/magazine/archive/2016/09/are-we-any-safer/492761/

    • Ed Wittens Cat

      and sadly…or not…the emails dont have much punch.
      witness the horserace media desperately trying to drag the viewers “into the weeds” of the State dept. emails.
      Wikileaks has published humdreds of thou– maybe millions of docs– no one reads them– Assange understands this — thus he now tries to make a sensationalist story out of leaks pre-release.
      Trumps core eumemes atm are Obama/Clinton created ISIS and “if-i-lose-it-means-the-election-was-stolen”.

    • Matt McIrvin

      There’s been an assumption that any international terrorist attack helps Trump, but that’s not even clear to me any more. Certainly it would ratchet up fear of foreigners, which plays to Trump’s message, but Clinton now has the edge in appearing qualified to deal with a crisis. When the Orlando mass killing happened, Trump’s first response was to congratulate himself for “being right” about it in a way that mostly disgusted people.

    • Ed Wittens Cat

      Matt, there are distinct kinds of terror events, and inability to distinguish btwn them is a big problem with public perception.
      if I was planning a black swan event that would tip the election I’d go for a dirty bomb as an October Surprise
      discussion here–> http://www.theatlantic.com/magazine/archive/2016/09/are-we-any-safer/492761/

  • The Ohioan

    The problem I see is that the line varies by state, due to gerrymandering and perhaps local political cultures and organization. In Ohio I’d say you need that +7 to flip districts; but in Michigan it can be much lower. I wonder if it’d be possible to model probabilities by using state-level presidential polls, and past voting record in the district.

  • Locke

    For an accurate calculation for this specific cycle we need to pre award all the seats with no challenger running.

    Since Debbie Wasserman Schultz decided to gift republicans a couple dozen districts that were less than +10% romney in 2012, the overall battle is vastly more uphill than 7% since the Republicans have already claimed a lot of winnable districts, due to no Democrat running.

  • Ellen

    We shall see. I am in the NJ 7th Congressional district, typical suburban moderate Republican district represented by an increasingly atypical moderate Republican, Leonard Lance. I think an experienced Democratic candidate could have made an issue of Lance’s support of Trump and been competitive, but the Democrats didn’t even try. Can’t beat something with nothing.

  • Jeremiah

    Eyeballing that chart and imagining a straight line drawn between the 2012 and 2014 result it looks like the Democrats need 55% or more of the national vote to gain a majority.

    • Steve

      It seems more appropriate to consider the distance of the 2012 & 2014 data points below the gray polygon. By that model, the Dems would need about 53% of the votes to flip the house.

    • Martin

      Just doesn’t seem likely, given Clinton’s favorables and the fact that Obama barely scraped over 51% four years ago. Trump will really have to bomb the debates AND be “seen” as bombing the debates by the wider public. Even then, I would not bet on 55%.

    • Sam Wang

      This is not a good approach, with only two data points and hidden nonlinearities. If you want to take a graphical approach, try Steve’s suggestion of looking at the distance underneath the shaded polygon.

    • Jeremiah

      Looking at the bottom of the gray polygon would only be a good approach if we knew all the points moved in that way. The size of that polygon is such that it could be that within a particular decade the curve moved the other way, or was always approximately linear. Of course, it doesn’t take much curvature to make the difference significant in terms of House control. Maybe a good approach would be to re-plot the data with a separate curve for each decade and that would reveal the curvature. Note both 2012 and 2014 have the same seat gaining power of approximately 0.91 which at least indicates the gerrymandering is a consistently linear effect within the 47 to 51 percent vote share range.

    • Sam Wang

      If we move all races uniformly by the same amount, the national vote that would have flipped the House was D+7% in 2012 and D+9% in 2014. Partisan gerrymandering effects should fade by an unknown amount. More R seats means more R-favoring incumbency effects, by an equivalent of +1% I estimate. Thus my estimate of D+6-8%.

      The gray-zone graphical approach probably contains this reasoning implicitly…but it is a noisier approach.

  • Joshua Baker

    If it really only takes a 5% generic ballot advantage to win back the House then I’m pretty sure Democrats are going to do it. I see Trump losing by at least 10% points when it comes down to it. He’s just too crazy and is spiraling into a feedback loop of insanity that won’t end anytime soon.

    With that kind of anchor weighing them down (and with Democrats doing everything they can to tie all GOP House & Senate races to Trump) I think Republicans will end up losing the overall House vote by quite a bit (although of course not by as much as Trump loses by.)

  • Joel

    Sam, feels like there’s less polling on local (congressional) races this year. And, seemingly, less state polling overall compared to 2012 and 2008. Is this true?

  • Carl Nyberg

    Do retirements follow patterns?

    Do Dems tend to retire after big GOP wins? And vice versa?

    Does change of which party holds majority decrease retirements among new majority? Increase among new minority?

    How much does retirement correlate to age? With the increased volatility of election outcomes, is the current Congress younger than Congresses in the past? Or have older members been secure with volatility with younger members of Congress?

  • Carl Nyberg

    Based on my Illinois perspective, the Dem Party didn’t even recruit serious candidates for districts that are officially swing districts (IL-12 & IL-13).

    Dems prob lucked out with a decent candidate in IL-12, but IL-13… the candidate seems unlikely to be viable in the best circumstances.

    If Dems were actually trying to win a majority in US House, they needed to be on offense in IL-06, IL-14 and IL-16.

    Dems didn’t even file a candidate in IL-16. I personally like the candidates in IL-06 & IL-14, but… In IL-06 the Dem had over $30K cash on hand on June 30 and in IL-14 he had under $5K.

    Even if things look good for Democrats on a “generic ballot”, are there actual candidates positioned to take advantage of this in actual winnable districts?

  • Carl Nyberg

    I have a hypothesis that seems testable.

    The partisan advantage of gerrymandering decreases over the decade.

    • Sam Wang

      As I said, that is what people think. See Figures 7-9 of Gelman and King 1990. I believe there is a later reference too.

      This topic needs to be revisited with simpler descriptive methods, such as those found in my SLR article.

    • Ravilyn Sanders

      Gerrymandering advantage morphs into incumbency advantage that is all. The effect on the winning party lasts for a long time.

    • counsellorben

      I agree that it is likely that the advantage of gerrymandering fades. I would extend the hypothesis to state that the advantage fades due in large part to the continued geographic expansion of metropolitan areas.

      Gerrymandering tends to act on voting populations the same way levees act to contain rivers. However, when the populations contained in a gerrymandered levee almost inevitably overflow into the surrounding districts, the effect of the gerrymander fades.

      For example, in Pennsylvania after the 2000 census, certain Democratic areas in Montgomery county which border Philadelphia were moved from the 13th CD (a then-competitive seat) to the 2nd CD (a safe Democratic seat covering much of North Philadelphia). Between 2002 to 2010, the 13th CD became a safe Democratic seat due to the continued growth of the Philadelphia metro area. After the 2010 census, these Democratic areas in Montgomery county were moved back into the 13th CD, since there was no longer a partisan advantage to the gerrymander.

      It would be interesting to study CDs which surround packed CDs during the span of a redistricting plan, to see how victory margins in the surrounding districts change over the plan’s lifespan.

  • Rob in CT

    I’m really curious as to whether we’ll see an uptick in ticket-splitting this year. It’s been on the decline for a while now, trending along with polarization. I think it’s plausible to expect an uptick (perhaps enough to keep the HoR Republican even if Trump loses by, say, 8%), but there are usually a ton of ways in which “this time is different” that turn out to be false on election day.

    • Sam Wang

      I agree, it seems more possible this year than the last five Presidential elections – as I said, this is part of modern polarization. However, note that the generic Congressional ballot and the national Clinton-versus-Trump matchup have been tracking together. It is not at all clear that 2016 is that different – yet.

  • Harvey Moseley

    Can gerrymandering be addressed by the courts only when a specific protected minority is denied voting representation? Is it generally considered as politics as usual otherwise?

    • Sam Wang

      Currently, that is the easiest lift. However, note that racial gerrymandering pertains to single districts.

      Partisan gerrymandering is theoretically justiciable, but the Supreme Court has not endorsed a specific standard. That is a live topic, and might be resolved in the next year or so. There are cases percolating upward through lower courts.

  • Scott

    Even if the Ds can’t win back the House outright, Trump may have the effect of forcing R’s to become much more moderate in order to hang onto their seats: http://www.politico.com/story/2016/08/obamacare-gop-campaign-trail-226939

    • Harvey Moseley

      At least during the campaign.

    • Kevin King

      Or it might knock moderate Republicans out of Congress, leaving more right wing congressmen from safer districts.

    • Will Hutchinson

      I doubt it because they would be worried about being primaried in the next election. They may change their tune after the primary during the general election, but come the next congressional session, I would expect them to return to form. So long as Citizens United lets someone drop tons of money into targeted races, incumbents in gerrymandered districts will toe the party line.

  • Randall Cream

    Question: Wouldn’t the most efficient gerrymandering scheme result in creating lots of R~+3% districts, allowing a majority to persist even with the loss of a national race – but become absolutely swamped once the needle moves past some threshold? Wouldn’t gerrymandered maps result in massive seat changes for only a few additional votes (small sigma, large net seat gain)? If this hypothesis is correct, the precise location of the line is very important, since few additional seats are gained as one approaches the line, but many, many are gained once surpassed.

    • Sam Wang

      That is basically true, except that the current standard for safe gerrymandering is a win by 10-20 percentage points. See Figure 4 of my Stanford Law Review article. The question is whether a national popular margin by Democrats of 5-8% would capture enough districts in that range, whether gerrymandered or not.

      The exact threshold is challenging to estimate because of heterogeneity among districts and effects like incumbency.

    • DaveM

      Hi Randall,
      When that happens, it gets called a “dummymander” – the party controlling redistricting spreads itself too thin and ends up being disproportionately _dis_advantaged due to slight electoral shifts.