Princeton Election Consortium

A first draft of electoral history. Since 2004

Does Trump’s ceiling matter?

January 7th, 2016, 9:55am by Sam Wang


Today, Ross Douthat quotes my post yesterday about Donald Trump’s current strength. He also says Trump is doomed because he will hit a ceiling of support around 30%. But even if that ceiling holds, it might not matter – because of how delegates are chosen.

The Republican Party has adopted a set of rules in which delegates are said to be allocated proportionally from Iowa/New Hampshire until Super Tuesday on March 1. However, there is a catch. In most states, if a candidate gets below some threshold of the statewide* vote, he/she gets no delegates.

Between now and Super Tuesday, 15 states will caucus or hold primaries. Here are some thresholds:

  • 20% threshold: Alabama, Georgia, Tennessee, Texas, Vermont.
  • 10-15% threshold: New Hampshire, Alaska, Arkansas, Minnesota, Oklahoma.

Graphically, this is what the rule looks like:

The arrowheads indicate approximate current standings in national and Iowa/New Hampshire polls. Two observations:

  • Trump is the only candidate who is consistently above threshold.
  • To get above threshold, other candidates have to claw their way over one another.

Obviously, as some lower-tier candidates drop out, others will get above threshold. This is a situation that bears watching.

I have done some preliminary simulations. The results suggest that with the field divided as it currently is, a candidate at 35-40% popular support can easily get over 50% of the statewide delegates in these first 15 states. Soon, I will share the results of these simulations.

Generally, these rules seem to be designed to avoid splintering of the field into many minor candidates. They reward candidates who can get a reasonable level of support, even if that is not a majority. Starting a few weeks after Super Tuesday, GOP primaries will start to use winner-take-all rule, in which a plurality is enough to win all the statewide delegates.

So, if you want to say that Trump is doomed, it would be best not to hang the argument on a ceiling of support that is below 50%. That ceiling will matter more if he faces only one other opponent. Calling the “establishment”!

*Some delegates are assigned on a Congressional district or region-by-region basis, carving a state into smaller areas for purposes of assigning delegates. These rules are byzantine, and I will address them at a future date. Let’s just say that rendering them in code is a long process.

P.S. Oh, yes…there is also this:

Tags: 2016 Election · President

36 Comments so far ↓

  • JesseE

    Dear Sam-

    This is a bit off-topic, but I’m curious about your general take on the Nate Silver/538 perspective on the Trump phenomenon. (Since you’ve battled him in the past, you seem uniquely positioned to offer perspective/critique).

    As I’m sure you know, Silver et al have been consistently bearish on Trump’s chances; through the summer and fall, they put Trump’s chances in the low single-digits (i.e. borderline Pataki-terriorty). Recently, they’ve started to concede that Trump might have a reasonable chance; yet they continue to peg his odds in the 10-15% range- well below what the polling data (and even the betting markets) suggest.

    I understand the theoretical reasons for this position. In addition to adhering to “The Party Decides’ thesis, they are deeply skeptical re early polling, believing that most voters aren’t paying attention or making up their minds. On this last point, Nate had a useful post in November, arguing that because only 1/3 or Iowa voters had actually decided at that point in the cycle, a more accurate estimate of vote shares would look something like this:

    Trump: 5
    Cruz: 4
    Rubio: 3
    Bush/Christie/Kasich: 2
    Undecided: 80

    I don’t necessarily buy this conclusion, but given 538′s Theory of the Case, I understand how he arrives at it.

    What bothers/surprises me, however, is their analysis of the ‘Establishment Lane’ candidates. Since October, they have consistently argued that: 1) that Rubio is the clear front-runner, pegging his odds in the 30-40% range, and 2) that Bush/Christie/Kasich are all single-digit underdogs. (In a November ‘Slack Chat’, Nate had Rubio at 45%, and B/C/K at 5-9%).

    My question is this: Given their skepticism re early polling, why do 538 forecasters unanimously consider Rubio to be the overwhelming favorite, both in absolute terms, and vis-a-vis his intra-establishment rivals. The same analysis of Iowa that put’s Trump’s vote share at 5% (just 1 point ahead of his closest non-establishment rival), also puts Rubio’s vote share at 3% (just 1 point ahead of his closest establishment rival). Why is Trump’s advantage met with such skepticism, while Rubio’s advantage is met with such certainty.

    In a way, I’m much more sympathetic to 538′s bearishness re Trump – given their perspective re early polling and Party Power, I understand how it follows. It’s their bullishness re Rubio that I find so baffling. It would be one thing if he was dominating in other Party Decides metrics: money + endorsements. But he isn’t. Instead, his only real advantage vis-a-vis other Establishment candidates is his polling. But 538 has told us again and again not to trust this polling (at least when it comes to Trump). So why is Rubio considered the run-away-favorite?

    • Sam Wang

      Jesse, a quick take.

      In summer/fall, it was probably appropriate to rely heavily on The-Party-Decides because polls lacked any predictive power. The question in my mind is why the FiveThirtyEight people have not updated that prior using polls. It is Silver’s style to react slowly to new data.

      Also, correct me if I am wrong, but I do not think they have done an analysis like what I posted last week. Even if they did, they (and I) now have a problem: The-Party-Decides and poll-based indicators are now pointing in very different directions. What now???

      In my view, this is because the national GOP has been moving toward crisis since 1994. Therefore I would say

      Probability that The-Party-Decides will fail = 30%.
      Probability that poll-based predictions will fail =15%.

      Based on that, I would guess that Trump is favored now over Rubio. (For now, I think Cruz is less likely because he scores so low on ranked-preference polls.)

      As for “Why Rubio?,” this is a consequence of The-Party-Decides. If one accepts that premise, then the only alternative is Jeb Bush based on endorsements, money, and officeholding experience. This prediction fails is if The-Party-Decides has waning influence. In national HuffPost averages, in January 2012 current and former officeholders were supported by about 80% of respondents. As of today that number is about 25%.

    • Kevin

      The case for Rubio has been an argument from elimination from the beginning. If you start from the premise “not Trump/Cruz/Carson,” then who? Well–not Bush (unpopular name plus everyone’s favorite punching bag thanks to Trump), Christie (Bridgegate), or Kasich (Medicaid expansion). Ergo, Rubio.

      The problem is that there is no evidence that actual voters make their decision this way. The parsimonious explanation is that voters choose the candidate they like best. Rubio may have more in common with Pawlenty/Walker than McCain/Romney.

      What is the positive case for Rubio? That he’s young and telegenic? That’s also how Dan Quayle and Sarah Palin were selected, but they didn’t turn out to be popular with voters en masse for long.

      I’ve heard two other positive cases for Rubio: 1) that his selection would build bridges to the Latino community, perhaps by magically moderating GOP voter preferences on immigration and economic issues; and 2) he’s really strong on foreign policy. On the latter, I don’t get it–I can’t be the only person who finds his pronouncements on foreign policy unhinged, and besides, 2016 voters don’t care much about foreign policy. On the former, this reasoning seems more compelling to center-left pundits (who also won’t be voting for Rubio) than actual GOP voters.

      It is ironic that Nate Silver built his brand on fact-checking punditry, yet keeps doubling down this cycle on pronouncements of opinion divorced from data. It’s one thing to (appropriately) point out that the data doesn’t tell us much at this stage, and another to leap from that position to telling everyone ad nauseam what’s bound to happen with Trump.

  • A New Jersey Farmer

    Is is still reasonable to assume that Carson will get votes if Trump somehow falters? I could see that in October, but he’s fallen quite a bit since then.

    Cruz is peaking at the right time but Christie seems to have stalled in NH after some brief press in December. All signs, as of today, point to Trump’s being ahead and at 35% is in very good position.

    I see this as great news for Democrats.

    • Amitabh Lath

      If Trump is gaining support within the GOP because of his message of independence from big money donors, it may resonate with Democrats in the general as well.

      Apparently 20% of Dems said they would defect and vote for Trump over Clinton.
      (Note that 14% of Republicans said they would go the other way, so not as big an advantage as it would seem).

      http://thehill.com/homenews/campaign/265330-some-dems-would-defect-for-trump-poll-shows

    • Petey

      Amitabh,

      It’s really difficult to draw any conclusions from either The Hill article, or the USNews article it aggregates from.

      1) No clear identification if it’s D-registered voters, or D-leaning voters. If the former, it’s utterly meaningless, as many Southern/Appalachian D-registered voters have long voted R at the federal level.

      2) The survey seems (?) to be done immediately after showing the group a Trump ad. That methodology obviously isn’t going to measure anything normal.

      Without better details about 1), and without a better methodology about 2), I don’t think any real conclusions can drawn whatsoever.

  • Amitabh Lath

    I’m looking at the 2nd choice data (October, The Upshot) as Sam suggests and basically everyone’s favorite 2nd choice is Carson (except Bobby Jindal supporters who like Cruz but as expected the statistical uncertainty is huge).

    And by far the largest 2nd choice of Carson supporters is Trump.

    So what happens to the Fiorina or Christie supporter who goes over to Carson, if Carson drops out?

    Do they behave like all other Carson supporters and move over to Trump (at the 24% level) or go over to their 3rd choice (if still available).

    In other words, are voters like electrons which “forget” their original polarization direction when you measure it along another axis, or classical particles which retain their identity (what Einstein called “hidden variables”)?

  • Amitabh Lath

    Douthat’s article is a mess. First he makes hay of the fact that Trump is behind in Iowa even though the polls there have him within the margin of error of 1st.

    But polls that show him ahead by large margins in other early states? Never mind, polls cannot be trusted.

    But the key to this Rossplanation is that as “establishment” candidates (as defined by Douthat and his ilk) drop out, all their support will go only to the remaining establishment ones and after a few iterations one candidate emerges to rule them all and slay Trump.

    This seems a bit much. The whole “establishment vs. outsider” frame may turn out to be a media creation. Candidates will drop out, and as to where their voters go, if your Bayesian prior is truly flat then they go everywhere including Trump (and some who stay home because if they can’t vote for Rand/Christie/Kasich then it’s no one).

    • Petey

      “Douthat’s article is a mess.”

      Most definitely. But it’s sorta emblematic of all the anti-Trump GOP pundits, who have been going through the Kübler-Ross stages of grief.

      Douthat is still stuck on Denial, though others have moved on to Bargaining or Depression. But both Denial and Bargaining require bad/spurious reasoning.

      (Standard disclaimer: I’m not asserting Trump will be the nominee. Things are still fluid, and I found the most recent Charlie Cook article not implausible. But there’s a whole lot of Trump Can’t Win punditry “mess” out there, on all sides.)

    • Sam Wang

      I agree, the establishment-voter idea is potentially a myth.

      One could test it by analyzing the Upshot’s old Condorcet data from October, using clustering analysis. I think the Dartmouth people who gathered it might have made the answers available – that’s potentially a gold mine for identifying voter types. If not – hey, Dartmouth people, read this thread!

    • Amitabh Lath

      Clustering analysis assumes there is a distance metric on a some rational issue-space, and voters move adiabatically on this space from one candidate to another. That may have been true in the past but perhaps no more.

      This time, the main issue underlying all others is billionaires funding candidates who then allegedly carry out their orders, be it increasing immigration or privatizing social security or trade deals with China etc.

      I suspect there is a one way valve in issue space. A Trump supporter thinks this is the highest priority, and is not about to turn into a Bush voter. However, a Bush voter may be persuaded this is important and go over to Trump (modulo personal distaste). This may be why Trump’s topline and favorables have been inching up.

    • Sam Wang

      You are being way too substantive. You’re missing my point – you are assuming the conclusion. Were you replaced by a pundit? Tsk tsk! ;-)

      I was imagining using only rank-order candidate preference, nothing else. This is the point of using the Dartmouth dataset. If there is such a thing as an “anti-establishment voter,” one would expect a Trump voter’s 2d/3rd choices to prefer Rubio as a second choice less often than, say, Christie voters do.

      That particular hypothetical is testable using a two-sample t-test. Now broaden it to all ranked preferences, and design a distance metric or some other clever math. Then look for clusters of candidates. If no clusters, then the pundits have it all wrong. (Again.)

    • Petey

      “Were you replaced by a pundit?”

      No spitting. No cursing. No fighting.

    • Amitabh Lath

      Yes, I’m theorizing above my pay grade. Wouldn’t be the first time.

      Looking at The Upshot excel file labelled “second_choice_by_first_choice_wtagend.csv”

      Trump people move mostly to Carson (36%), everyone else in single digits.
      Bush (Rubio) at 8% (7%).

      Carson people’s 2nd choice was Trump by a larger margin than anyone else (24%).

      Rubio people also move to Carson (22%). Christie and Trump are about the same at 8% and 6% respectively.

      Bush people go to Carson, Rubio, Trump at 18%,14%, 11%

      Carson stood to gain as the 2nd choice of most candidate’s supporters. This may have changed since October.

      Bush –> Trump is higher than Trump –> Bush, but I guess that’s not a surprise.

  • Matt McIrvin

    Douthat cites three early states in which Trump’s support level is a little below his national share, then goes on to cite that number as if it were the national share. The national aggregates are running closer to 40% than 30%, which says to me that Trump is getting a lot more in some states that aren’t getting a lot of polling attention.

    Now, since we don’t have a national primary and they’re later primaries, that might not matter if Trump can get knocked out early. But it doesn’t look to me like he gets knocked out early unless he responds in some unusually bad way to losing Iowa.

    • Sam Wang

      In 2012, from Iowa to Super Tuesday the leading four candidates (Romney, Santorum, Gingrich, Paul) each had vote shares with a state-by-state standard deviation of 12-13 percent.

      Any deviations from national polls would have to be >10% and in the same direction for me to take notice. Trying to read much from a few differences of a few points each is attempting to extract meaning from noise.

  • Phillygirl

    Proportional allocation actually extends through March 14, giving eight more states after Super Tuesday a chance to torment the Republican field further. Probably not what Reince Priebus had in mind.

  • Michael K

    How do you account for all the “soft”, “unbound”, “unpledged”, etc. delegates?

    According to this (it’s from 2012, has it changed much?)– http://fivethirtyeight.com/features/the-g-o-p-s-fuzzy-delegate-math/ — only about three-quarters of delegates are bound to a candidate on at least the first ballot of the convention.

    Is it not fair to assume that many of those unbound delegates (lots of them party officials or chosen by party committees / conventions) will be eager to support anyone but Trump? And that Trump would therefore need to win a large majority of the bound delegates to compensate for that?

  • Petey

    Bravo, Sam. Just the article I was waiting for.

    My months-long hobbyhorse of GOP delegate allocation heavily benefiting winners finally gets a model!

    My only very minor quibble:

    “After Super Tuesday, GOP primaries will start to use winner-take-all rule, in which a plurality is enough to win all the statewide delegates.”

    Well, actually, only 16% of delegates come from WTA states. But 25% more of delegates come from ‘hybrid’ states. And my lazy ‘n’ sloppy analysis is that many of those ‘hybrid’ states are pseudo-WTA with any kind of halfway decent margin. For example, SC, which is ‘hybrid’, even though it’s inside the ‘proportional’ window, seems likely to award 90%+ of delegates to a candidate that wins 35-25.

    Yet another reason, (among many), to see the SC Cruz/Trump showdown as pivotal, as bks points out. Win that, and ride it to semi-sweep down South 10 days later, (and probably throughout March), and you’ll be damn hard to stop.

    Tangentially, interesting Amy Walter piece on open vs closed primaries. The upshot is that Trump non-R-reg voters have it easy early, and hard late. So, if we go with the Trump builds massive delegate lead through March scenario, one more way for ‘the party’ to try a last ditch effort to fight back…

  • bks

    Put me down for thinking that Trump going Birther on Cruz will be very successful. Is the GOP anti-intellectual commentariat going to suddenly embrace empiricism, science and the Harvard Law Review? No, they try to create their own reality and just reminding them that Cruz was born in Canada will be sufficient. The story is already creating a frenzy in the RWEC. Most commented story at Breitbart and most commented story at Hot Air.

  • TC

    A couple decades ago, the Republican Party abandoned all constituents except the One Percent. The RP needed proxies though to get the OP elected. They chose a) White Supremacists and b) Evangelical Christians. Now is fruition. Trump is the WS proxy candidate, while Cruz is the EC proxy candidate. The other candidates are considered Phonies by the Republican voters. It appears doubtful that a Phony can win. Also doubtful that the EC proxy candidate can win, as the US is more racist than religious. Seems the largest question is will a victorious Trump choose Cruz as VP and then be able to beat Clinton. Looks like a coin toss. A horrible coin toss. It would be statistically interesting and I think enlightening to analyze the degree of White Supremacy and Evangelical Christianity in the early and mid primary states.

  • Nick Warino

    Playing with the RCP numbers:

    On May 3, Rubio’s measured support peaked at 14.3%. He got up to 14.8% a few weeks ago.

    Right now, he’s at 11%. While a 3.3% decline isn’t that bad, what makes this even worse for Rubio is to consider how the field was different in May: Bush’s campaign hadn’t collapsed, Walker was still a top contender, and Perry, Graham, Jindal, and Pataki were there too.

    May 3rd RCP Polling Average:

    Jeb! Bush: 15.4%
    Scott Walker: 12.3%
    Rick Perry: 2.5%
    Lindsay Graham: 1.7%
    Bobby Jindal: 1.3%
    George Pataki: 0.5%

    COMBINED: 33.8%

    Since then, only Bush remains, currently pulling in a mere 3.3%. That’s a disappearance of 30.5%, entirely from candidates in the crowded “establishment lane.” You would think that Rubio would be best suited to grab that support.

    So what’s changed since then?

    Trump: +35.3
    Cruz: +11.2%
    Carson: +4.7%
    Fiorina: +1.0%
    Kasich: -0.3%
    Pataki: -0.5% (ELIMINATED)
    Christie: -0.8%
    Santorum: -1.3%
    Jindal: -1.3% (ELIMINATED)
    Graham: -1.7% (ELIMINATED)
    Perry: -2.5% (ELIMINATED)
    Rubio: -3.3%
    Huckabee: -6.2%
    Paul: -7.0%
    Jeb!: -12.2%
    Walker: -12.0% (ELIMINATED)

    All the shift has been away from establishment candidates and towards non-establishment, mostly Trump, then Cruz, and a bit for Carson.

    This is why the assumption that the as the field consolidates, Rubio will pick up most of that support is suspect. For the past 7 months, the field has been consolidating, but mostly around Trump and to a lesser extent Cruz.

    So maybe it’s Rubio who has a low ceiling?

    • Kalil

      This is something I’ve been wondering for some time.

      There’s been a lot of discussion about Trump’s ‘ceiling’, but almost none of Rubio’s. Trump, Carson, and Cruz have gone up and down, but it’s been a very long time since Trump + Carson + Cruz has been less than 60%, and a very long time since Bush+Rubio+Christie has been greater than 30%. Even adding in semi-outsider and briefly serious contender Fiorina doesn't really push them over that. This suggests to me that if anyone has an inability to win, it's the Rubio establishment wing.

  • Amitabh Lath

    You made a Monte Carlo! Awesome. Code up on github?

    Anyway, how does one calculate this “ceiling” with any confidence? Is the assumption that as lower tier candidates drop out, absolutely none of their support will go to Trump? Zero? Is that plausible?

    Maybe it is people saying “I would never vote for candidate X?” During early 2008 a whole bunch of Hillary Clinton’s supporters were saying that about Obama. They got over it.

    • Sam Wang

      The code is currently hideous to behold. My least pretty effort in some time.

      Ceiling estimates:

      (1) First-choice plus second-choice as Public Policy Polling does.

      (2) Condorcet ranked-preference polls, as appeared in the Upshot in fall – though this gives a rather higher ceiling.

      (3) Favorable-unfavorable ratings, though as Harry Enten has pointed out, these turn out to be malleable.

      Basically I think the ceiling could be there, but somehow we don’t hear it when other candidates are discussed. Could be motivated reasoning.

    • Matt McIrvin

      Salon seems to have a long-running line now in articles by Sanders supporters urging liberals to throw the general election to the Republicans if Clinton is nominated. I’ve been trying to figure out if this is just the normal nonsense like the 2008 PUMAs or if it’s a real danger. I currently suspect it’s not much to worry about, but some Democrats are starting to openly worry.

  • Paul

    I am looking at the delegate allocation rules in IA (no threshold) and NH (10% threshold). In each state, three delegates are chosen by the party, so there will be 27 delegates chosen according to the IA caucuses and 20 delegates chosen according to the New Hampshire primary. If the current state polling holds (I use an average of all state polls conducted within the last month), the results follow (total of IA+NH):
    Trump 17
    Cruz 11
    Rubio 7
    Christie 5
    Carson 3
    Bush 1
    Fiorina 1
    Paul 1
    Huckabee 1
    Kasich 0
    Santorum 0

    Specifically. Cruz and Trump would tie with 8 delegates each in IA and no one else has more than 3 (Rubio and Carson). In NH, due to the 10% threshold (only Trump, Rubio, Christie and Cruz have poll averages over 10%), 7 delegates end up being reallocated from those candidates under 10%. The net effect of the 10% threshold is that the allocation in NH ends up being Trump 46%, Rubio 21%, Christie 17% and Cruz 17% with all the other candidates at 0% because they do not make the 10% cut. By the way, the two big losers with this 10% threshold in NH would be Bush and Kasich who are currently polling at 8%. So, how many candidates make the 10% cut has a significant effect on the net benefit to the top tier candidate(s) in this threshold state.

    • 538 Refugee

      I was wondering what happens to the allocated delegates of people that drop out. Maybe that is why no one ‘drops out’ they “suspend” and hold onto their bargaining chips (delegates)?

  • bks

    I’m looking at the South Carolina primary, Saturday 20 February as the GOP watershed. More delegates than Iowa + NH, and a clear window onto the SEC primary ten days later.

Leave a Comment