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

Synch And Swim: Quantifying The Coattail Effect

October 18th, 2016, 9:00am by Sam Wang

With three weeks to go to the election, Hillary Clinton is headed for an electoral victory comparable to President Obama’s 2008 and 2012 wins. In national polls, Donald Trump’s support is where it has been all year, around 40-42% of voters. If Clinton’s lead increases, the added support will probably come from “undecided” and minor-party voters. Meanwhile, Trump is focusing entirely on core supporters, who tolerate or endorse white nationalism and racially-driven sentiment. His supporters are committed Republicans – but they are considerably outnumbered by other voters.

At their July convention, Clinton, Obama, and other Democrats said that good Republicans can vote for Clinton and still remain good Republicans. This approach implicitly says that voters are free to split their tickets for Clinton and for downticket Republicans.

Despite this permission slip from Democrats, partisan loyalty appears to be as strong as ever. And the fortunes of a Clinton Presidency appear to be highly dependent on that loyalty. That loyalty allows us to estimate that a Clinton win by more than 3% would probably be associated with Democratic control of the Senate. And a Clinton win by more than 8% would favor a Democratic House. Currently, the optimal way to help/hurt Clinton is in Senate races.

The Princeton Election Consortium has converted the questions of the Presidential race, Senate control, and the national House vote to a common “currency” of measurement: percentage points of popular support. In the case of the Presidency and Senate, it’s a Meta-Margin, defined as how much swing in popular vote would be needed to create a toss-up. In the case of the House, it’s just the generic Congressional ballot.

All three of these measures move up and down together. We can compare them to ask the question: when the Presidential Meta-Margin moves by 1.0%, how much do the other margins move?

The answer is: about 1.0% in the House, and 0.7% in the Senate. The Presidential race, House popular vote, and Senate control move in lockstep.

Here is the House generic ballot, plotted against Clinton’s Meta-Margin:

The eight-day delay probably arises from the fact that both House polls and state Presidential polls take 2-3 weeks to turn over, and Presidential polls take a little longer to become fresh across all swing states.

The red diagonal line indicates equality. Most of the data points are above the diagonal. House Democrats have typically been running 0.0-1.0 percentage point stronger than Clinton. That’s true today: the House is at D+5.0%, and the Presidential Meta-Margin is at Clinton +4.4%.

I estimate that Democrats must win the national popular vote by 8% to have any chance at taking control of the House. This large margin is driven by two major factors in equal measure: gerrymandering to pack Democrats into districts, and population patterns which they pack themselves. Therefore the magic number for House Democrats is a Clinton win by 8%. In national polls Clinton is currently ahead by 5% (7 polls starting on October 10th or later), and Obama outperformed his 2012 polls by 3%, so it’s not crazy to imagine. I’d give the House Democrats a 1 in 5 chance of making it over this bar. A long shot…but not a crazy long shot.

Here is the Senate Meta-Margin:

A Clinton Meta-Margin of 2-3% is associated with a Senate Meta-Margin of 0%, which is defined as identifying a perfect toss-up for Senate control. Note that the slope indicated by the data points is about 0.7. 0.7 is less than 1, so changes in Clinton support are not fully translated to downticket Senate races.

As of today, Senate candidates are underperforming. In the closest races, Democratic candidates are running 4-6 percentage points weaker than Clinton in North Carolina, New Hampshire, and Pennsylvania. It would seem like a good idea for those three candidates to tie themselves to Clinton – and tie their opponents to Trump. Conversely, the optimal strategy for Republican candidates is to take up Obama and Clinton’s invitation to separate themselves from Trump.

Tags: 2016 Election · House · Senate

60 Comments so far ↓

  • liberal

    “It would seem like a good idea for those three candidates to tie themselves to Clinton – and tie their opponents to Trump.”

    I’ve seen the claim that Democrats fail to “nationalize” Congressional races to an optimal extent.

  • Michael

    Are you factoring into “senate control” a 50-50 split between the parties, but a majority based on the vice president being, say, on the winning Democratic ticket?

    • Matt McIrvin

      His charts do seem to be taking that into account.

    • babar

      in that case it comes down to whoever replaces tim kaine

    • Prehistorian

      On babar’s point, Kaine’s replacement will be appointed by the Virginia Governor until a special election takes place in 2017.

  • bks

    (Slightly off-topic) How are we doing on the Internet polls versus live-telephone polls question? I think the thesis was that Clinton has an advantage in the latter category.

    • George

      I think you can go to and create your own graph doing exactly what you are asking.

    • Some Body

      Not quite as off-topic as it seems. Because a gap between the presidential and Senate candidates in each state, which Sam mentioned, can also be a measure of a potential social-desirability-based gap. (Also, note that Clinton rarely breaks the 50% barrier in national and swing state polls; Trump, of course, is further back).

      A couple of months back, 538 ran a story on live vs. non-live polls, and found Clinton performs better in live polls in blue states and better in non-live polls in red states. What I noticed was that this is exactly the pattern you’d expect if social desirability bias was in play (with both “shy” Trump voters and “shy” Clinton voters, but with the former outnumbering the latter overall).

      On you can essentially check how things stand now for yourself (by going to state poll charts and creating two customized charts with only live and only non-live polls). I hope to do that bit of research if I get the spare time at some point, and post it as a comment on this site (if Sam allows that).

    • Chillax

      I guess I just don’t understand this thesis. Trump supporters seem to be a loud, proud bunch eager to show their support for him, and yet they are afraid to tell a stranger conducting a phone poll that they support him?

    • bks

      Going to HuffPolls and selecting just Internet pollsters:
      47.7 to 41.6
      just LivePhone:
      47.6 to 38.6
      So apparently it’s not that Clinton has an advantage but rather that Trump has a disadvantage on live calls.

    • GB

      In the 3-way poll option on Pollster (which is, IMO, the correct one, given that Johnson is running and polling significant support), it plays out like this:


      Clinton 45.0; Trump 38.3; Johnson 6.8

      Phone Only:

      Clinton 47.3; Trump 38.4; Johnson 5.8

      Live Phone Only:

      Clinton 47.4; Trump 38.4; Johnson 5.8

      Internet Only:

      Clinton 43.6; Trump 39.2; Johnson 5.5

      What’s interesting to note is that Trump’s support does not shift in a statistically meaningful way, regardless of how polls are conducted, but Clinton’s does–yet only in the internet-only polls. I suspect the internet-only polls are capturing a degree of millennial ambivalence toward Clinton, which has been documented elsewhere, but that the internet-only polls are overemphasizing this effect (while the live phone polls may be underemphasizing it).

      And there does not appear to be any kind of “silent Trump majority” or whatever. He’s stuck just under 40% and it seems unlikely that he’ll gain more than half the remaining undecideds.

      My guess is, barring unforeseen events, it ends up something like:

      Clinton 48; Trump 42; Johnson 7; Stein 2; Other 1.

  • Ed Wittens Cat

    i would just like to point out
    the brilliant timing of Obama’s Mosul offensive.
    Effectively spikes Trumps guns on Iraq and ISIS– its now a convo Trump dont wanna have.
    Right up to the election (3 weeks) Mosul will do fine, grab headlines, and drum up chestbeating war porn.
    Americans and the media both love remote war on brown ppl. Another sunni population center turned into a smoking ruin is just what Hillary needs to avoid Trumps claims that she and Obama started ISIS and are “weak” on “radical islamic terrorism”.

  • Stephen


    Is it possible to change the yellow in your outcome, median EV graphs to something with better contrast with the white background?

  • anonymous

    “It would seem like a good idea for those three candidates to tie themselves to Clinton – and tie their opponents to Trump.”

    The Senate candidates should probably support Clinton’s GOTV operation and coordinate with the Clinton campaign wherever possible, but do so behind the scenes. There is probably little upside to publicly associate themselves with Clinton. Their underperformance can probably be explained by the fact that their own opponent is not Trump. So tying their opponent to Trump is likely to be much more fruitful for these Senate candidates.

    • FlyingSquirrel

      Yeah, I’m not sure down-ballot Dems have that much to gain from “tying themselves to Clinton.” If voters are planning to split their tickets in response to an anticipated Clinton win, then a better strategy might be (a) tie the GOP candidate to Trump, (b) make the case that the result of a Clinton victory and a Republican Senate will not be reasonable compromise but pointless confrontation and obstruction, (c) tie themselves to the positive aspects of Obama’s record and make the point that the GOP are still mostly committed to tearing down much of what he accomplished.

  • Daniel Barkalow

    I think it would be interesting to have a non-causal meta-margin data set for this sort of analysis. I.e., to get the value for each point in time, smoothly combine the polls that were being taken before and after that point in time. You obviously can’t do that in real time, but it might be a good way of doing this sort of analysis of recent history.

    • Bob Felt

      Has anyone seen the latest polls from the Washington Post? They really seem way off. Especially for FL. Outliers?

    • Matt McIrvin

      WaPo/SurveyMonkey surveyed a lot of states and most of their numbers don’t look all that weird. You’d expect a few of them to be weird outliers just by chance. The one showing Texas close is getting a lot of attention, but their numbers for Florida and Ohio are a little redder than average. Probably just random variation.

  • Brian

    Are you having some website problems?
    Data on banner and “tabular data” seem to be stuck on last night’s 12:05 am installment.

    • Brian

      FYI: The tabular data file finally has an entry for today (as of the 8:05 PM update). The banner (apparently) is still not reflecting the most recent meta margin and other president #’s.

    • Sam Wang

      yeah, sorry about that. An attempted fix to deal with Maine and Nebraska CDs is a bit crash-y. Need to fix that.

      It should be updating now – let me know if anything seems amiss.

  • Adam

    Does the calculation use 435 house races as a base or does it use as a base the number of seats with competition. Since many seats have only one party on the ballot, including many districts that were less than 10% Romney in 2012, it seems as though a smaller number than 435 should be used as including the single party races is probably skewing the NPV level democrats need to achieve artificially downward.

  • DeanH

    It should be interesting to review all the data a week after the last debate. Reviewing it now seems premature. If there is minimal change, we may see the Dems start a Senate blitz against the Rs.

  • Anthony

    It is very interesting that although in theory every election should be better for Democrats from a demographics standpoint, Clinton is likely to get almost the same numbers as Obama in 2012. So much for the demographics is destiny argument.

    I think what we are going to see in future elections is that as the minority share of the vote continues to increase, the Republican nominee will continue to see a similar increase of the vote share of the white vote. You would think as older whites die off this would not happen, however the alt-right movement contains a lot of young white voters replacing older white voters who die off.

    • Olav Grinde

      Antony, after 30 years abroad, I am still stymied by the fact that the Democratic Party has managed to lose so much of the white blue-collar vote. I just don’t get it!

    • Shawn Huckaby

      I’m a little puzzled by some of your assumptions. Arguably Clinton is a more flawed candidate than Obama, so would be expected to slightly underperform his numbers given the same demographics. Since the demographics in Clinton’s favor have increased slightly since 2012 it would explain her getting approximately the same share of the vote.

      I see no reason to think white voter distribution across the political spectrum would change solely because of it being a smaller portion of the electorate. That would imply all whites feel equally embattled by their shrinking influence. While that might be true on the margins (emphasize might), there are many urban and college educated whites who have no problem with seeing a more multicultural society emerge.

      The older racist/nativist contingent may indeed be replaced by today’s “alt-right” movement, but I don’t see it becoming an attractive stance for the public at large. Donald Trump may be emboldening these people to be more open and vocal in their racism, but I don’t think the reason they remain a fringe element is because people haven’t heard of them before. The Klan could run advertisements on every media outlet there is and not greatly increase their numbers.

    • Anthony

      Shawn Huckaby, im not trying to make any assumptions, I’m just trying to interpret the data that we have in polling so far. Clinton may be a flawed candidate, however from the polling data it seems that:

      (1) Clinton is doing similar to Obama in every demographic except whites, dramtically worse with non college educated white men.
      (2) Whites are less % of the electorate than in 2012, therefore if Clinton just held Obama’s demographics she should be doing better, but shes not due to her during worse numbers with whites
      (3) Even though she is unpopular, there does not seem to be a massive shift towards third party as we get towards election day. Jill Stein is now between 0-2%. Gary Johnson pulls about evenly from Trump and Clinton so hes a wash.
      (4) Clinton’s meta margin has been stable around +4%, Obama finished 2012 with 3.9%
      (5) I cannot think of any other conclusion as to why this is other than my original post. Her being more unpopular than Obama does not fully explain it, especially since Trump is even more unpopular and far more unpopular than Romney ever was.

    • alurin

      Clinton is running ahead of Obama ’12 in both the meta-margin and national polling. I don’t know how much of that is demographics and how much is the T—p effect, but certainly the demographics is destiny argument is still alive.

    • Rob


      In addition to what others are saying, don’t discount the power of labor unions of the past. In 1954, the US workforce reached peak union attainment with 35% of the workforce being in labor unions. These were mostly blue-collar workers. Now, union members are about 11% of the workforce and are mostly white-collar government workers. The blue-collar worker likely felt more protected then they do now and the unions were effective tools for mobilizing voters to vote for the Democratic party (the GOP was considered the party of the upper management class).

    • Roger Tang

      Heh. Don’t forget to add “to white voters” when you say Clinton is unpopular. We’re really not seeing THAT much difference between her and other Democrats among non-white voters.

    • GB

      Flawed reasoning.

      The demographics argument says that, under same or similar conditions, the Democrats will exhibit an incrementally growing advantage. That is to say, a generic Democrat running against a generic Republican in 2016 will have maybe 1-2% advantage from demographic changes relative to 2012.

      And that’s all, as the conditions and candidates are in no way, shape or form constant.

    • C Hart

      ” However the alt-right movement contains a lot of young white voters replacing older white voters who die off.”

      No, it really doesn’t. They are a twitter phenomenon. Their real numbers are objectively small.

    • Vatnos

      The parties change constituencies over time. They don’t represent fixed ideological viewpoints. Progressive Democrats and young voters voted for Obama in large numbers in 2008, but were disillusioned with him and had weaker turnout in 2012. Those people supported Sanders in the primary this year, but their support for Clinton is more tempered. They know they’ll be getting more of the same. Fool me once… etc.

      Donald Trump is an extremely different candidate from Romney, or McCain, or Bush. He’s playing on right wing economic populism to some extent, opposing trade deals and deporting immigrants. There’s some overlap in his constituency, but he also draws in some conspiracy theorists–the Alex Jones, David Eich crowd who aren’t necessarily right win at all, but think Hillary wants to start World War 3. Hard to say how the Republican party will change after this election. The establishment will obviously try to blame the loss on Trump and go back to the same-old same-old next time, while republican voters who like Trump more than the sponges the establishment offered will probably try to find a “more reasonable, politically shrewd Trump” for the next go-round….

    • Phil

      I don’t think we can weigh in on this until the voting actually happens. And even then there are so many confounding factors in this election due to Trump and resurgent white nationalism that it would be pretty hard to tease out.

      Since we live in the Electoral College world, I like to watch how the states are moving. The Midwest is definitely moving R as a whole, some parts faster than others. The Southwest is moving D. The coastal South is moving D. The Northeast as a whole seems to be more static, but individual states are performing differently.

      The bigger picture is that the Southwest and the coastal South are parts of this country that are growing faster than the Midwest. What’s happening is not an even trade. So if in eight years Iowa and Ohio are solid red, it won’t make up for Texas, Arizona, and Georgia being swing states and NC and FL leaning blue.

      But again, we have to wait to see the actual votes.

    • alurin

      While T—p is a very different candidate from Romney or McCain or Bush in many senses, it’s important to remember that the Republicans have been playing off white nationalism for fifty years now. Republican ideology has long been authoritarianism with a thin veneer of libertarianism. T—p has basically stripped off the veneer. That is alienating some people, the “limited government” true believers, but they were never the majority of the Republican electorate. Is there really anyone who somehow missed fifty years of racist dog whistles but is now switching over to T—p because he comes out and says what Reagan just hinted at? I don’t think T—p has really changed the composition of the Republican electorate that much, aside from driving away white women and George Will.

  • Sean

    I look forward to where we stand around the end of the month. The last debate will have been over and more time will have passed since the wikileaks and North Carolina incidents.

    I don’t expect much to change honestly. I think at this point people have chosen a side and no amount of sensational headlines is going to change that.

  • George H

    Allan Lichtman is not backing down. Last week in Chicago:

    Professor With Perfect Prediction Record Says Trump Will Win Presidency

    “The Keys often go against polling”

    Then there is the “silent army”:

    wsj: Donald Trump Says He Will Win on Surge of Silent Supporters

    Brexit was the hope.

  • AAF

    Why use the presidential meta margin here?

    The meta margin is a way to eliminate the effect of big leads vs. bigger leads on non-competitive states, which can affect a national poll but is irrelevant if you’re trying to predict EVs. But when correlating House races to the presidential race, there is no reason to eliminate that effect – in fact, keeping in all the information about presidentially noncompetitive states seems pretty important.

    If one candidate is leading by 60% in New York and California and New Jersey, and the other by only 15% in Texas, Missouri, Louisiana and Florida, that difference in lead sizes would be reflected in the national polls as increasing the first candidate’s overall lead, while being nearly invisible in the Meta Margin because all those states would be out of play.

    But, for correlating to House races, it would matter a lot that a candidate was absolutely crushing in their States will the other was just winning handily in theirs.

  • Sassy

    Big lie: Trump is focusing entirely on core supporters, who tolerate or endorse white nationalism and racially-driven sentiment. You seem to forget he reached out to the Hispanic and Black communities.

    All Trump supporters worry about the economy as their no. 1 preoccupation. Trump fans include Hispanics, Latinos, Hindus, Blacks… Very great shame on you.

    That said, the polls are not reliable. Results are weighted against the 2012 model/outcome.

    Not too long ago, a 10 million voter registration increase was noted for the Republicans while some 10 million voter decrease in registration for the Democrats. Thus, the 2012 model no longer applies.

    Also, I have often noticed (when methodology is available) a much higher percentage of Democrats polled.

    This is one election where the polls are to be ignored.

    • Sam Wang

      This comment by Sassy is loaded with falsehoods and misleading statements…but it is an interesting view of Trumpland.

    • Michael Hahn

      Sassy: I can hardly wait to see what you will say on Nov. 9 when the outcome turns out as the aggregate polls suggest!! As has been the case for the last several presidential elections. Pollsters have become very good at what they do in Presidential election years.

    • Sam Wang

      …okay, I’ll take this one.

      “Big lie: Trump is focusing entirely on core supporters”: My statement of his focus is obviously true. We can break it down. His campaign staff is headed by inwardly-turned operatives from Breitbart and Fox. When he makes an “appeal” to minorities, it takes the form of telling African-Americans that they have nothing to lose, and assuming that they live in inner cities. These appeals have totally failed. Remaining silent in the face of endorsements from white nationalists is also conspicuous to any nonwhite voter.

      Polls have been reliable since 2004. Weighting to prior elections…that is the standard way to do it, and it has succeeded in the past. Occasionally pollsters will adopt a “high turnout” or a “low turnout” model. If you do not like 2012-base weighting, an alternative is Census-based weighting; for example, see this interesting analysis of the L.A. Times/Dornsife/USC poll.

      In this case, you are essentially arguing against the wisdom of the entire pollster crowd. You don’t have any evidence for this. It coincides with your candidate falling behind, which is suggestive of motivated reasoning, i.e. looking for reasons to disbelieve good evidence. For example, I do not recall hearing such objections when Trump led in primary-season polls.

      No matter who you support, drilling into internals of a poll (i.e. “higher percentage of Democrats polled”) is a reliable tell for motivated reasoning. Another interpretation here is that respondents have a slight tendency to report a partisan ID that matches their preferred candidate.

    • Sassy

      I am not saying polls are unreliable, only that this election in particular and they should be disregarded. Remember the Democratic nomination whereby Clinton was leading Sanders by huge margins. When all was said and done, results were much closer than anticipated. For example, how do you explain this?

      Something else to consider in this election: the Monster Vote estimated at some 10 million individuals that have never voted or bothered to vote before. Passions are high and I maintain that polls are to be taken with a huge grain of salt this time around.

    • Sam Wang

      Passions are high, but variability is low. Those go together – think about it.

      Comparison to primaries has to be done carefully, as I have discussed in my review of how well polling of Democratic primaries did. Caucus states, in which voting is more difficult than primary states, are to be viewed with special caution. The case of Michigan, where there was no usable voter model to begin with, is discussed here. Keep in mind the exuberance effect: in states where one candidate is considerably ahead, he/she often outperforms polls.

      “Monster Vote”: I challenge you to justify an estimate of 10 million missing voters. One could give voter-registration statistics to counter this claim, for example this report that the GOP registration advantage in New Hampshire has decreased. Generally, recent new registrations tilt Democratic. I am not aware of any reports of a differential surge in GOP registration. Finally, for a larger look, note that voting has started already.

      If you want, click on the “Trump +2%” link at the right to see how a realistically optimistic speculation would turn out.

    • Sassy

      Sam Wang: With all respect, we should continue this discussion on November 9. We live in interesting times, n’est-ce pas? Regards.

    • Shawn Huckaby

      Please go there, Sassy. It’s getting much too expensive to bet against your candidate.

  • Phoenix Woman

    Sassy, two things:

    1) Dunning-Krueger Effect. You’re manifesting it.

    2) You might find this link helpful:

  • Clay

    Hi Sam,

    Is it possible to predict turnout at this point in the process, perhaps from early voting data? I can’t help but feel that given the negative views of the candidates it might be suppressed but that Republican turnout may be lower given the rift in the party and that might affect the election results.

    Do you know if the likely voter screening has a way to catch this?

  • A

    Yay, Sam finally made the “eat a bug” comment on Twitter! That makes it official…

    Sam Wang via Twitter: It is totally over. If Trump wins more than 240 electoral votes, I will eat a bug.

  • Ru

    Is there any way of quantifying GOTV effects? Is that perhaps the reason Obama out-performed his polls?

Leave a Comment