Princeton Election Consortium

A first draft of electoral history. Since 2004

Current Polls Favor A Trump Delegate Majority

April 9th, 2016, 4:59pm by Sam Wang


This week in the Republican nomination race, Ted Cruz’s win in Wisconsin triggered buzz about how front-runner Donald Trump might be in trouble. Doubtless today’s win in Colorado will intensify the chatter, and will involve words like “momentum.” It is best to ignore all of that coverage – at least until some national polling data shows a sustained change. Why? Because states differ from one another, mostly in demographics but also in rules and various local factors. It is almost impossible to learn something new from a single race. To know where the race stands as a whole, it is necessary to consider all states at once.

In several ways, Wisconsin was typical. With a pre-election poll median of 36.0 ± 1.5% (median ± estimated SEM), Trump’s vote share of 35% was on the mark, continuing his close match between polls and outcomes. Cruz’s finish was also typical, but for a different reason: he was, and is, outperforming his polls. Cruz’s pre-election polls were 39.0 ± 1.2%, and he ended up with 48% of the vote. In previous states, Cruz has overperformed by a median factor of 1.2. Either Cruz’s supporters are exceptionally committed, or he is the beneficiary of anti-Trump votes liberated from their previous first choices, or undecided voters break hard for him, or some combination of the three. In Wisconsin he may also have benefited from the fact that trailing candidates like Kasich often underperform their polls when it is time to vote.

Where is the national race now? The current 6-national-poll median (March 29-April 6) is Trump 39.5 ± 1.2%, Cruz 31.0  ± 2.1%, Kasich 19.0 ± 1.1%. If we were to apply a 1.2-fold bonus to Cruz’s numbers to allow for his overperformance, the corrected numbers are Trump 39.5%, Cruz 37.2% – extremely close. Either way, Cruz has risen quite a bit in the last month, and national opinion is now closely divided.

I have updated the polls-only snapshot of the remaining Republican primaries through June 7th, when voting ends. As I pointed out months ago in The New Republic and The American Prospect, Republican rules are complex and tilt the playing field toward the front-runner, even if he/she doesn’t get a majority of the popular vote. Therefore it is essential to emulate the state-by-state delegate rules with close attention to quantitative accuracy.

Even after getting the rules right, this is a challenging calculation for three reasons: (a) many states lack polls; (b) Cruz overperforms his polls; and (c) delegates may not follow the rules. Today I describe one way of dealing with all of these issues.

For those who just want the bottom line: Since my last update, a poll-based snapshot has moved – in Trump’s favor. If current polls accurately measure voter behavior, then Donald Trump would get a median of 1,356 delegates – almost 120 more than the 1,237 he needs for a first-ballot victory at the national convention in Cleveland. For this probability to drop to 50%, his national lead would have to drop by 8.0% – this is Trump’s Meta-Margin, a measure I have previously developed for general-election Presidential races. However, if Cruz’s overperformance continues, Trump’s lead would narrow considerably, to a count of 1,280 delegates and a Meta-Margin of 2.0%. After allowing for Cruz’s potential overperformance, the probability of a Trump majority is 70% – probable but uncertain. Under such closely divided conditions, the outcome won’t be known until the last primaries, on June 7th.

And now I will explain at length.

Here is a snapshot of current polls, with no correction for Cruz. It gives a median of 1,356 delegates for Trump, 119 more than the 1,237 necessary to get a majority on the first ballot.

It was done under the following assumptions:

1) State polls. Between now and June 7th, the 16 remaining states have 769 delegates, 31% out of a total of 2,474. Only four of these states – New York, Maryland, Pennsylvania, and California – have polls conducted in the last two weeks. A central problem is therefore how to construct the natural variation in state-to-state support. This can be done using national surveys, based on the fact that the national average contains respondents drawn from across the U.S.

To estimate support across the 12 states that lack polls, I used the fact that Trump’s vote share will fluctuate around his national support by some standard deviation (SD). From 2000 to 2012, the Republican front-runner’s SD has been 10-12%. This year, Trump’s SD has been 10%. I assumed SD=12%.

This SD can also be used to estimate how much Trump’s average vote share in these 12 states will deviate from the national average: the standard error (SE) of Trump’s 12-state average is 3.5%, the minimum amount by which Trump’s average will deviate from national numbers. Combining this SE with this year’s polling inaccuracy (about 4%), I estimate that Trump’s average is uncertain by +/-5%. Therefore I varied Trump’s 12-state average by +/-5% around national polls, and assumed that individual states varied around this average by +/-12% (all values are one-sigma).

2) Statewide delegate rules. These rules, as well as district-level rules, were implemented using the detailed descriptions available at The Green Papers. Colorado has no election and was omitted from the calculation.

Ten remaining elections assign at least some statewide delegates on a winner-take-all basis. The exact probability distribution of all possible outcomes is easy to simulate using the same method I have used for the Electoral College. NY and CT statewide delegates are winner-take-all if the top finisher gets above 50%, proportional otherwise. The remaining four states  (RI, OR, WA, NM) are proportional.

3) Congressional district-level rules. In nine states, 3 delegates per district are assigned locally. In the past, a candidate’s district vote share typically has varied around the statewide average with an SD of 3-5%. I simulated this with a t-distribution to allow for outliers. Under winner-take-all rules, which apply in most states, the rule is well approximated by an S-shaped curve. The curve is very steep – think of South Carolina, where Donald Trump won all nine Congressional districts. In the code, I also dealt with additional subtleties in the rules for New York, Rhode Island, and Washington that go beyond winner-take-all.

(For an extended discussion of how 2) and 3) above play out in assigning one state’s delegates, see my discussion of California here.)

The biggest uncertainty in the calculation comes in Pennsylvania, where many delegates declare a preference, but strictly speaking are are unbound. I assumed that Pennsylvania delegates will vote according to their district’s voters. This gives Trump 54 delegates on average. In real life, the true allegiance of these delegates is somewhat uncertain.

>>>

The calculation above has two important features. First, in the histogram above, 92% of the probability is at 1,237 delegates or greater. Second, the probability is reduced to 50% if all margins are reduced by 8.0% across the board. This is very similar to the Meta-Margin that I have defined for general election races. For example, if the second-place finisher (Cruz) is underestimated by 8%, that would even up the race. Alternatively, if 4% of GOP voters switch from Trump to Cruz, that would reduce margins by 8%. Either way, a Meta-Margin of 8.0% means that effectively, Trump is 8% ahead in polls.

I do not think Trump’s probability of getting a majority is actually 92%. The biggest reason is that Cruz overperforms his polls. If we reduce Trump’s margins by 6 percentage points based on Cruz’s national numbers, the median outcome is 1,280 Trump delegates – only 43 delegates to spare. Under this assumption, Trump’s probability of a delegate majority is 70%, about 2-1 in Trump’s favor.

Here is what the histogram looks like with a correction for Ted Cruz’s overperformance:

Finally, a note on non-pledged delegates. I have left out the approximately 120 delegates who are either uncommitted or not bindable (see cells B11 and B13 of Taniel’s spreadsheet), and therefore potentially recruitable for the first ballot. With these, Trump’s possible median could be anywhere between 1,226 (no Pennsylvania district-level delegates, no non-pledged delegates) and 1,400 delegates (all of both groups of delegates).

Finally…the scripts, somewhat ugly for now, can be found here, here, and here. I’ll document them better in a little bit. Guardedly, I welcome corrections and comments.

Correction: For NY and CT, the script calculates whether Trump gets over 50%. The Cruz bonus was erroneously applied, but is now removed. This adds 8 delegates to Trump’s total.

Tags: 2016 Election · President

72 Comments so far ↓

  • 538 Refugee

    Found this buried nugget. There may be one rule change after all and it would guarantee either Trump or Cruz would be the nominee.

    ” Ash intended for the committee to consider an amendment to the party rules to prevent the reopening of nominations on the convention floor—the pathway to the nomination of a “white knight” in the event of a deadlocked convention.

    That amendment would have shifted the convention procedural rules from the rules of the House of Representatives—which has been in place for generations—to the stricter Roberts Rules of Order, thereby weakening the convention chairman in a move viewed as weakening the party establishment’s control over the convention. Hours later, RNC Counsel John Ryder told the committee it would take up the measure after all.”

    http://fortune.com/2016/04/18/republican-chairman-seeks-to-halt-rules-changes-as-trump-complains/

  • Don

    And by the way, I am an evangelical Christian way to the right of Cruz, who I am for, but I’m voting for Trump–a fact I think is overlooked in Trump supporters. There are millions of voters like me, a positive/negative in the works on which the outcome is predicated, and should be looked into by those who use prevalent factors on which to project outcomes. I should say, Sam, that your facts and figures are extremely interesting and I will visit you more. Thanks!

  • BJ

    Has any consideration been given to the idea that Cruz’s “overperformance” in the polls may be a regional phenomenon? I know there’s a lot of Northeastern states that haven’t voted yet.

    Also could this “overperformance” be related to negative campaigning against other candidates? I know they went negative against Kasich in Wisconsin.

    Another thing, where the polls in New York adjusted to account for the fact that polls list undecided voters. Getting over 50% in each district on election day where there’s no “undecided voters” in the count will be easier than doing so in the polls, especially if there’s a high undecided voters count.

  • Bloodstar

    Dr Wang,

    I believe your a analysis is mistaken for several hypotheses:

    1) national polls serve as a poor proxy in general for state voting.

    2) national polls are increasingly composed of voters in states that have already cast their primary ballot.

    3) The Colorado Debacle for Trump gives potential voters questions on the competence of Trump. If he can’t manage a campaign (or surround himself with people who can) can manage the country.

    4) New York is not conducive for Trump to sweep the delegates. At this point trump trump needs to gather all the delegates possible in the north east and mid Atlantic before the contests in WA, OR, NM, and CA and it looks like Trump will get 170 to 180 delegates before Indiana. (Not counting the goofy PA district allocation)

    Assuming IN and WV breaks trump’s way that will be another ~80 delegates. SD, NE, and MT are going Cruz’s way. So Trump is left with NJ (51), and then OR, WA, NM, and CA.

    He will likely have 1071 delegates going into those remaining states. Give him 25 for OR and NM (proportional), so Trump will need ~140 delegates from WA, CA, and uncommitted. Which is not an impossible task. It’s just there are major assumptions WV and IN being the big ones, that may be wrong.

    5) Kasich staying in the race is hurting Trump in the NE and will likely be a net negative in OR and NM for Trump.

    Just my thoughts and unprofessional analysis. :)

    • Sam Wang

      I really think you’re missing the point about national polls. The point is that national polls *by necessity* are the average of all the states. This is ineluctably true. Similarly, it is true that there is variation from state to state. This is measured by quantities such as the standard deviation (SD). By using the fact of that variation, one can estimate how many states are likely to fall one way or another. Finally, the average of the remaining states will deviate from national opinion by some amount. That is expressed by the standard error of the mean (SEM). In short, your concerns are contained in my assumptions.

      Basically, I am trying to show you how to think about this problem without all the details that fill most commentary on this topic. Your other comments offer detailed commentary on individual states. I am sure they have merit, but they constitute one of many possible opinions. The point of the calculation I offer is to probabilistically consider all possibilities, weighted by likelihood.

      Finally, I note that the race appears to have been static for the last three weeks. So sure, Trump could decline. But he hasn’t since June 2015.

    • Sam Wang

      If we’re really going to get into individual states, we can all take our favorite guess without resorting to a probabilistic calculation. Based on polls and the “Cruz bonus,” here is mine. Using the votes-to-delegates conversions contained within my code, I estimate the current modal (i.e. single most common combination) Trump outcome as follows. Start with 758 pledged delegates. Then:

      NY: >50% statewide, >50% in about 16 of 27 Congressional districts, for a total of 14+70=84 delegates.
      CA: 13 delegates statewide, plus 32 of 53 CDs, for a total of 109 delegates.
      NJ, MD, WV, PA, CT: wins, including all CDs (ignoring PA defections), for a total of 222 delegates.
      NM, OR: proportional for 7+11=18 delegates.
      RI, WA: elaborate rules with narrow outcome range, 7+19=26 delegates.
      DE, NE, MT, SD: winner-take-all, but no polls. If we guess that Trump only wins DE, 16 delegates.
      IN: highly uncertain. I would assign between 0 and 57 delegates.
      Total: 1233 to 1290 delegates.

      It is a pity that Indiana does not allow automated phone polling. This is probably a big reason why there is no poll there. An Internet company like YouGov should get on the stick there…

    • Bloodstar

      Thank you for the response, I find it fascinating how we can both look at similar data and have totally different takes!

      My main issue with a probabilistic methodology for determine outcomes is that I am hypothesizing that here is an over estimation of the correlations.. By that I mean I think the data from polls, particularly national polls, will not well reflect races in a state level. I suppose if I get time I can try to do some regression and correlation analysis and see if I can get some data that can support (or reject) my hypothesis, but that’s definitely out of the question until May at the earliest.

      Of course, the thing I like and hate about a probabilistic approach is that no matter what the outcome, it can’t be shown the model is wrong (Unless you can get a large enough set of results).

      thanks again for the response!

    • Sam Wang

      To argue your side, I agree that a main problem with my way is that I can’t guarantee where the actual 9-state (or whatever we lack polls for) average sits within the +/-SEM window. Until now I have avoided estimating unknown states. I guess one could try a demographically-inspired approach, which is what your state-by-state guesses (or mine) amount to. The biggest problem is how to estimate Indiana. I saw an attempt in the NYT and was not convinced.

    • Commentor27

      You raise good arguments, but it only ultimately amounts to a quality pundit-style horse race analysis, which can almost always find ways to rationalize not believing the polls.

  • A New Jersey Farmer

    The reality of denying Trump the nomination, even if he is under 1237 will be too daunting a challenge for the party. But as Sam says, that’s likely a moot point given that he will go to Cleveland with enough delegates.

    • Sam Wang

      It is by no means definite that Trump will get >1237 pledged delegates.

    • 538 Refugee

      I heard some Republican official, forget who, say that 1100 would do it because if Trump reaches that figure the party won’t deny him the nomination. My guess is the prospect of turning the primaries into a sham would just be too damaging to the party.

    • bks

      In two weeks Cruz will be eliminated from a first-ballot victory. That should make a compelling argument for Trump going into CA.

  • timothy

    this professor today reveals that the Trump won’t win the majority of the delegates and GOP will pick Kasich and that will defeat Hillary.
    http://www.huffingtonpost.com/seth-abramson/john-kasich-will-be-the-r_b_9638598.html

    • Sam Wang

      clip ‘n’ save

    • Amitabh Lath

      Professors say the darndest things.

      Remember those Colorado professors who predicted a Romney win? Not just a win, but a 330 EV landslide.

    • kahner

      he “reveals” it? i gotta find out where he got the time machine. the rest of us are schlubs are relegated to just making predictions.

    • Woochifer

      Yes, a professor of English.

      Abramson’s the same guy who predicted the collapse of Clinton’s African-American and Latino support, based on crosstabulations from some cherrypicked outlier polls. If the exit polls from NY suggest anything, it’s that Clinton’s AA and Latino support has hardly wavered at all.

  • Ondrej Nezdara

    First, thank you for your work. I’m not the U.S. citizen but I’ve been following the U.S. elections and coming to your site occasionally since 2012. Excellent stuff.

    Obviously, there has been an over-performance of Cruz. However, I wouldn’t be surprised if it is not necessary Cruz but the biggest Trump challenger in the state. That would explain Ohio and also it is (or may be) pretty rational behavior from non-Trump voters. My guess is that if we see Kasich doing better than Cruz in coming opinion polls this week, Kasich may get even 10% more votes than Cruz in New York primaries. Same for Connecticut.

  • Will Hutchinson

    Coming from Illinois myself, let me amplify a little on Amitabh Lath’s posts. Scott Walker has an 80% approval rating among Republicans. He has a 43% rating overall, but that’s beside the point. He endorsed Cruz, which counted for a lot among the Republican faithful and might have handed Cruz the victory. Then Trump compounded his own problem with some very rude comments about Gov. Walker. That undoubtedly pushed even more Walker supporters into voting for Cruz. I happened to be in Wisconsin on primary day and saw no one who was surprised by the Republican result.

  • Causes and FX

    Interesting breakdown. Thank you.

  • AySz88

    Sam, which version of MATLAB are you on? Don’t want to start introducing too much stuff (ex. Live Scripts, classes) that you can’t use.

    • Sam Wang

      Currently R2013b…there are newer versions floating around this place. I am under some pressure from lab members to be more recent.

  • Daniel Barkalow

    It seems to me that people who plan to vote for whichever non-Trump candidate will be most likely to deny Trump delegates in their state, and wait until the last minute to decide who this is, will not be counted usefully by standard polls. If there is a substantial group of such people, it could account for Cruz overperforming polls. (Interestingly, Cruz didn’t overperform polls in Ohio, and Kasich did, while Kasich hasn’t elsewhere.)

    So maybe that’s an effect that we can expect to see through the rest of the election, and we shouldn’t be too surprised if we see an outcome like what we get by giving Cruz an extra 6%. On the other hand, I’m not convinced that 20% of Cruz’s other supporters is a good metric for the size of this bump; if there is a group that’s indistinguishable from “undecided” in polls, but goes as a bloc to some particular candidate in each race, the size of the group wouldn’t tend to vary with Cruz’s poll figures. I’d look for correlation between people who poll as “undecided” and local non-Trump front-runners outperforming their polls.

    • Sam Wang

      I was thinking this same thing. If you look at national surveys, Trump+Cruz+Kasich adds up to a median of 96%, which suggests that 4% of people aren’t able to articulate a preference. Note that Cruz’s overperformance was largest during March, when his national numbers were moving most rapidly. I think that is consistent with your suggestion. It would not be hard to get 6% from a combination of this and last-minute defections from Kasich supporters.

      My original reason for expressing the offset as a ratio was the fact that the data seemed to be well fitted by a multiplicative factor rather than a fixed offset. Now both ways work about equally well. I agree that a fixed percentage makes more sense, and this is in fact how I offset the Trump-Cruz margins in the calculation.

  • Mark F.

    Sam’s analysis is sound. However, my gut feeling is that Trump will come up a little short and they will nominate Cruz on a 2nd or 3rd ballot. As for Kasich–he is delusional.

  • C.S.Strowbridge

    Dr. Wang,

    Does Colorado change the numbers? Donald Trump’s campaign was about as organized as a rolling dumpster fire and that cost them dearly.

    How many simulations had Trump winning enough in Colorado to make that state essential?

    And does the possibility of another self-inflicted wound further reduce the numbers below 50%, in your opinion?

  • mediaglyphic

    Dr. Wang,
    in your model does Sum of all ExpectedValue per sate, = ExpectedValue national?

  • Austin

    Mr. Wang, your analysis has been wonderful and very informative, and I look forward to more articles.

  • P G Vaidya

    Excellent analysis!

    It seems that it is being read and discussed widely. I just noticed that in Predictit Trump has gone back up to 50 percent.

    • Amitabh Lath

      Trump’s share price moving up might not be an indicator that bettors believe he will get greater than 1237 delegates, it might be that they believe he will prevail in a contested convention.

      The share price for a contested convention has remained steady around 0.65 or so.

    • Josh

      The two are not mutually exclusive. If Trump finishes, say, 20 delegates short, there’s still a reasonable chance he’ll shore up enough unpledged delegates to clinch the nomination.

  • bemused senior

    The just reported Field Poll in California gives info regarding Trump-Cruz-Kasich strength by region. Since the delegate allocation rules are wta by district, it looked to me like the delegate estimation could be refined with this. Do you think it is worth looking at?

    • Sam Wang

      I looked at the Field Poll. Their data do suggest that the SD statewide might be somewhat larger than my 5% assumption. The subsamples are ~140 people each, which creates some problem because of small sample.

      I used past districtwise voting for other candidates (see thegreenpapers.com for returns). SD is fairly similar across many states, which is interesting and useful. I could examine this more, but it seems to work well in its current form.

    • bks

      More fun to look at New York than California as we only have to wait nine days to see how the model (mis)behaves.

    • Sam Wang

      P.S. regarding the Field Poll: the standard deviation among the four regions is 7%. It is not hard to get this down to 5%: it could be done if the support in “Other Southern California” were not 45%, but 40%. For the moment I am inclined to leave the SD at 5%.

  • BF

    Could you break this out state-by-state? I am trying to reconcile this with other analyses, but hard to see why you model is so much more optimistic for Trump than the others. Presumably, CA is the big difference, where I suspect your model suggests something like a Trump sweep?

    • Sam Wang

      I agree that California is critical.

      In CA, Trump leads Cruz by 7.5% (n=2). With a 6% correction that is a 1.5% lead, with an uncertainty of 5%. The 17 statewide delegates go to Trump with win probability = tcdf(1.5/5,2)=0.65, more or less (not at computer now, sorry). This is then used as a probability to construct the histogram.

      The fraction of the 159 CD delegates going to Trump is again 0.65 because I assume SD=5% based on historical patterns. (Generally because of the similarity of the two calculations, the share of district-level winner-take-all races should be equal or greater than a front-runner’s statewide win probability.) This leads to a fixed assignment of 103 delegates.

      Doing the above, over and over, for all other states goes into making one histogram of outcomes.

      Then calculate a whole set of histograms, stepping through a range of differences between 12-state average and national polling median. This offsets the candidate margin in states without polling data by varying amounts*. Weight the histograms according to a t-distribution.

      *In this logic, CA should not be varied. However, I currently have it varying along with the unpolled states. I think the effect of changing this would be small.

  • Ed Wittens Cat

    i wonder if Cruz overperformance could be related to perception of relative electabilty or perception of down ballot damage? Conservatives I know believe Cruz is more electable than Trump. Granted, thats anecdotal– i dont see much difference between them in the polls. But the media will lather this up as Cruz “momentum” to increase the horse race aspect.
    My personal opinion is that Trump is more telegenic than Cruz. i wish there was polling to measure telegenicity.

  • Matt McIrvin

    This is a case where I’m itching to see the critique of other media analyses, which seem to have settled on the idea that Trump is not going to make it on the first ballot (and frequently go on to insist that he can’t possibly get it on subsequent ballots either, the fix being in).

    • Matt McIrvin

      …Of course, the way the political media excitedly speculate about the return of the “brokered convention” in every single election cycle, even though it never happens, is famous by now. Since this time around we have probably the highest actual probability of one since 1952, it’s only natural they’d go nuts.

  • D Lu

    Does this model incorporate demographic or regional differences? If I understood this correctly, it sounded like you assumed the numbers from all unpolled states were iid. So the model would give an equal probability to Trump winning Connecticut and Trump winning Nebraska. That seems like a bad assumption.

    • Sam Wang

      One could do that, though I don’t think it would make much difference. Anyway, you do not pay enough for this content to get that!

      Isn’t there some other guy who became well-known in 2008 for this kind of thing? Oh wait, he’s no longer doing math.

    • Amitabh Lath

      If there is no state polling, using national polling and eating a large systematic as Sam does is preferable to creating some sort of “regional voter” model.

      Look at the difference between IL and WI. What model could have forseen the large difference between Republican primary results in these otherwise nearly identical states?

    • D Lu

      @Amitabh Lath I’m pretty sure the reason Dr. Wang’s analysis is so divergent from that of other poli-quants is because almost everyone else is writing off Nebraska, South Dakota, and Montana (all WTA states with almost 100 delegates between them) as unwinnable for Trump based on his regional weakness. I think Dr. Wang’s model, on the other hand, treats them as identical to unpolled states in the Northeast.
      I agree you’re right in that it’s difficult to make a non-overfitting model more sophisticated than “Trump is weak in the Mountain states, so let’s just say he loses.” That doesn’t mean, however, that it’s necessarily wrong.
      So I think this is the critical difference. Dr. Wang’s model will be much more accurate than the Silver’s and the Cohn’s if making this fuzzy, human-knowledge based correction turns out to be an overfit. But I don’t know a good way to evaluate the probability of that.

    • AySz88

      Interestingly, Wasserman did give them a demographics-based model (the “delegate targets”). But the numbers are stale – I don’t know why it’s not being refit as states vote. Wasserman himself seems to be concentrating on the Democratic side first for some reason (safer call?).

      And yeah, their head honcho’s steadily getting less and less moored to the basics. It seems more like a bettor’s perspective now than a quantitative one.

    • Amitabh Lath

      D Lu, if one were forming a Bayesian prior about Nebraska, South Dakota, and Montana, one might assume Nebraska would follow Illinois (another midwestern state with large farms) where Trump won big. Or Kansas to the south where Trump is ahead by about 5%.

      One might also be tempted to lump South and North Dakota together but their economies are very different. SD is mainly agriculture and tourism while ND is energy extraction from the Bakken formation. So maybe SD votes like other agricultural states? Who knows.

      One could also spin a similar tale for Montanna but the point is that there are enough variables in the “let’s use fundamentals” game that one would surely come up with a prediction favorable to their biases. And for news agencies the bias is almost certainly towards a contested convention.

      Using the national average might have several other problems, but sneaking in personal bias isn’t one of them.

  • Doctor Science

    Does the stuff that’s been recently reported about Cruz lining up support among Trump’s delegates make any difference? Or will that only come into play if there’s no first-ballot winner?

  • Bill Herschel

    The thing that jumped off the page in the exit polls reported in the NY Times was that Trump had done better than Cruz with voters who had decided at the last minute, something that is unprecedented.

    The opinion pages in the Times have become an exercise in contortion. Krugman is doing everything in his power to appear not to be a Trump supporter, but he keeps writing columns indicating that Trump’s positions are far less nuts than the other candidate’s. Then there’s Douthat writing his contemplative column advocating disenfranchisement of Republican voters, although of course he doesn’t call it that. His contorted facies trying hard not to say he supports Cruz.

    • C.S.Strowbridge

      “The opinion pages in the Times have become an exercise in contortion. Krugman is doing everything in his power to appear not to be a Trump supporter, but he keeps writing columns indicating that Trump’s positions are far less nuts than the other candidate’s.”

      Both of those are true.

      https://www.isidewith.com/political-quiz

      I’m Canadian and a supporter of the NDP, so it comes as no surprise that I fit well with Jill Stein and Bernie Sanders (both in the high 90s) while Hillary Clinton is close behind at 95%.

      Cruz is at 13%, but Trump is at 25%.

      So yeah, according to my political views, Trump is half as nutty as Cruz, but I certainly don’t support him.

    • Matt McIrvin

      Krugman is most likely trying to forestall the political media going all-in for any non-Trump candidate who manages to get nominated, out of sheer relief that the guy is not Trump.

    • Mitch

      Now this isn’t quite fair to Krugman. His point is generally that people think that the others’ programs aren’t nuts because they’re put forth without profanity, but Krugman points out that they’re no less insane than Trump’s from a textbook economic point of view.

  • Amitabh Lath

    What’s the mechanism behind this Cruz overperformance factor? As you say the polls have been fairly predictive except for this 1.2x for Cruz.

    My first instinct would be to blame polls that listed Carson, Rubio et al. As they dropped out the polls were not swift enough to catch the move to Cruz.

    Another hypothesis would be that Cruz pulls a new demographic voting for the first time in the primaries.

    Maybe NY will be instructive. Recent polls have just the three of them.

    • Matt McIrvin

      Cruz’s base is the evangelical religious right, and they’re highly motivated to turn out, which is particularly important in primaries and caucuses. I don’t think they’re a new demographic, I think they’re just people who vote very reliably.

    • Amitabh Lath

      So weighting for percent evangelical (ie, come up with a poll correction factor) might go some way to explaining this Cruz 1.2x? Does it work the other way (does he underperform in states with no evangelicals?)

    • 538 Refugee

      New York would be the “evangelical test”. Hint, look towards the bottom first.

      http://www.beliefnet.com/Faiths/2004/11/State-By-State-Percentage-Of-White-Evangelicals-Catholics-And-Black-Protestants.aspx

    • Some Body

      Don’t forget there are undecided respondents in all the polls. Cruz seems to be getting most of them most of the time.

    • LondonBob

      Another element I think is that open primaries are now an issue for Trump. I can only explain Wisconsin by supposing that a significant proportion of Cruz’s support were Democrats voting for a variety of mischievous reasons. Turnout, better ground game and rallying around nearest closest competitor all make sense too.

    • Amitabh Lath

      One explanation for Cruz in WI I have read is the “last gasp of the Establishment” hypothesis. Ryan and Walker are of the national establishment but still well regarded among WI Republican voters. Also, Walker faced a recall election that made the state party more cohesive . These are not properties that are transferable to other states. Of course the usual caveats for post-hoc explanations apply.

  • mediaglyphic

    Thank you for this. Is there an easy way to see expected value by state for trump (forecast)?

    http://www.centerforpolitics.org/crystalball/articles/assessing-trumps-path-to-1237/

    A couple of weeks back i saw the above from Larry Sabato’s website, others seem to have their own versions. It would be interesting to see if where the pundits differ most widely from the polls.

    • Ron

      Read the Sabato piece and wow, right off the bat he missed WI, which was enough of a swing for Trump NOT to have a majority (by Sabatos own numbers)

  • Joel W.

    Fascinating! Many people think Trump was in big trouble because of his terrible week, but that appears not to be the case.

    Thanks for this analysis. Oh, and one question: Do you think or know if any of the remaining candidates follow this site? They might learn a few useful things from it.

    • bks

      It’s safe to say that all of their numerate advisers are aware of Sam’s work, even if the candidates, themselves, are not.

    • Some Body

      @bks: Does that assume all of them have numerate advisers?

    • bks

      I admit that it does seem that Kasich and his entourage cannot count to 1237.

    • Matt McIrvin

      At this point Kasich has to be entirely motivated by the dream that Trump will end up under 1237 and then Kasich will end up as the white knight appointed by the party on a later ballot.

      He does have the advantage that unlike Ryan or Romney, he actually is running for President, and at this point he polls well in the general-election question, much better than anyone else running and even better than the Democratic frontrunner. On paper, that sounds like a good pitch. I have my doubts that he can actually get delegates that way; I think at this point that if it’s not Trump it’ll be Cruz, and everything else is political reporters spinning exciting drama.

    • Joel W.

      Kasich has stated that he’s not interested in the vice presidency. Nevertheless, an alliance between him and Trump might prove mutually beneficial. From Kasich’s point of view it would raise his profile for 2020, win or lose; and from Trump’s, assuming Kasich were to direct his delegates to support Trump at the convention, it would guarantee the nomination and increase his chances of winning Ohio. Trump and Kasich supposedly “ganged up” against Cruz in Michigan, which might signal future cooperation. Actually, Kasich staying in the race might be evidence of cooperation all by itself.