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

This weekend’s primaries: More of the same?

March 5th, 2016, 2:15pm by Sam Wang

Update: a surprise in Kansas, whose primary is closed to independent voters. Cruz is winning with 51%, outperforming expectations by over 20 percentage points. That’s a spot where the paint isn’t drying the way I expected. Overall, though, results are consistent with previous primaries and caucuses: spots of strength for Cruz, and Rubio lagging in third place.

Greetings from Hong Kong. Pardon the delay in this post.

Thirteen time zones away from home, I am amazed at how closely people here are following the Republican primaries. Yesterday two autism therapists (my trip here is a scientific one) told me they thought Trump might be OK for domestic policy because he’s a businessman, but when it came to foreign policy he would be incompetent and catastrophic. Not a large sample…but I was amazed that the conversation went in this direction.

This weekend, Democrats* and Republicans in four states vote: Kansas, Louisiana, Nebraska, Maine. Also, Puerto Rico votes (Republican only). I see these contests, as well as other contests March 7-12, as being like watching paint dry. Sure, the paint will upend one major political party and is degrading our national public conversation. There’s that. But from a polling/delegates standpoint, it is not clear how much the candidates’ standings will change.

FHQ estimates that Trump has 336 out of 722 of delegates awarded so far. In my view, the fraction of delegates to date, 47%, is the key number to watch. I think the question of whether he gets to 1237, a majority of all delegates, is less useful to watch. Horserace journalists use it as a memorable benchmark, but you may find it easier to follow along if you calculate the fraction. Anyway, because races so far have been relatively weak states for Trump, this fraction will tend to increase over time. When it gets over 50%, that will be a milestone that suggests the paint is dry.

On the Republican side, reports** of this week’s debate resemble the last few debates, only worse. In July I did not anticipate Trump’s persistent lead. Nor did I think he and Rubio would now be having a public go-to over personal endowment. Or that Kasich would express a desire to invade a few countries. But we are adapted to all of this freakishness now. Maybe I’m wrong, but it does not seem that much was revealed to sway GOP voters.

Likewise, polling data suggests that this weekend’s states might resemble Super Tuesday states on average. To extract an aggregated measure, I calculated a delegate-weighted average of poll medians (step 1: take poll medians for each state. step 2: calculate an average weighted by each state’s number of delegates). For three states, Kansas, Louisiana, and Nebraska, this calculation gives a result of Trump 33%, Cruz 20%, Rubio 17%, and Kasich 7%. Note that there is no polling for Maine. Those percentages are similar to the Super Tuesday states:

Even if Maine (23 delegates) goes for Kasich (or Cruz – closed primary! and early returns do favor him), it seems that today we will get a few more data points that are representative of the primary process to date.


To continue my postmortem from Wednesday, compare the bar graph above with the actual results:

As I said a few days ago, Trump, Cruz, and Rubio all outperformed polls. But Trump won less delegates than my calculation predicted. There is an error that I should have avoided: I did that calculation by hand, and did not allocate undecided voters (my MATLAB script does this, but I didn’t use it). Cruz and Rubio are often near 15-20%, a threshold below which many states don’t give any statewide delegates. By outperforming polls on Election Day, they did better than I predicted.

This raises the problem of how to allocate undecideds. Here is a state-by-state comparison of opinion polls and actual outcomes for all five candidates.

The data points for Trump, Cruz, and Rubio are mostly above the black diagonal, which reflects the fact that they outperformed polls. However, points for Kasich and Carson are on both sides of the diagonal. Kasich’s outcomes were a median of 0.2 percentage points above polls, and Carson underperformed by about 1 percentage point. These results are consistent with the idea that undecided voters mostly end up committing to the most viable candidates. This suggests a rule for simulations: when candidate support adds up to less than 100 percent, the remainder should be divided equally among candidates who are at 10% or above in polls.

Update: in Kansas returns so far, Cruz leads with 51%, followed by Trump 24%, Rubio 15%, and Kasich 9%. That’s a surprisingly good showing for Cruz compared with the only fresh poll (Trump 35%, Cruz 29%, Rubio 17%, Kasich 13%). Cruz is developing a track record of places where he overperforms – caucuses and closed primaries, it seems.


Looking ahead: recall that Trump was not as strong in Super Tuesday states as he is nationally (current median support, 44%, 4 independent surveys with samples spanning February 24-March 2). Even if Mitt Romney gets his wish and Kasich and Rubio stay off one another’s turf, it’s a three-way race. That, combined with the increasing prevalence of winner-take-all contests, means that Trump’s fraction of delegates awarded so far (again note: not fraction of all delegates!) will increase in the weeks ahead. As I said above, Trump could get over 50% as early as March 15th if wins Florida and Ohio. Ohio could conceivably go for Kasich, and Florida is getting somewhat closer. But keeping Trump’s eventual delegate count below 50% would require not just these two upsets by Kasich and Rubio, but for Trump’s national support to drop a fair bit. A combination of all of these events is what it would take to put a dent in Trump’s inevitability.

* I’m not going to write about Democrats. If we use the majority of pledged delegates as a criterion, the paint became dry as of South Carolina/Nevada. If one includes superdelegates in the count, Hillary Clinton started above the 50% criterion even before Iowa/New Hampshire and has not dipped below.

** “reports”: I didn’t watch it. Honestly, I might be done watching debates until the general election season.

Tags: 2016 Election · President

48 Comments so far ↓

  • mediaglyphic

    closed vs. open primaries

    have the polls been as unbiased in open vs. closed primaries? there seems to be a feeling that Cruz might outperform Drumpf in the closed primaries. any thoughts?

    • Sam Wang

      One could drill into the polls to see if they make a point of matching party affiliation to the state’s open/closed rules. It does seem that Cruz has outperformed more than Trump/Rubio.

    • Amitabh Lath

      Cruz is outperforming Kansas just called for Cruz. He is significantly ahead in Maine as well.

    • Olav Grinde

      How surprising are today’s GOP results in Kansas and Maine? I understand there is a paucity of polls in most of the states being decided this weekend.

    • Sam Wang

      Cruz overperformed in Kansas by a lot. Closed primary.

    • bks

      I think we’re seeing a flow of votes from Rubio to Cruz. Today the Cruz camp said they were going to fight for votes in Florida. For low-information Rubio voters, Cruz may be an easy leap if they (rightfully) think that Rubio is finished and they’re at all anti-Trump. Plus Cruz may be getting Carson residuals.

    • Froggy

      Two interesting questions from tonight’s Cruz rise and Rubio stumble (which is just a continuation from Tuesday): 1) How much does Kasich benefit (leading today in an ARG poll in Michigan)? 2) If Trump is actually brought down, how fast will the pivot be from #NeverTrump to #NeverCruz?

    • bks

      Cruz overperformed, but still KY & LA for Trump. On to Michigan!

    • Josh

      In closed caucuses, yes, Cruz should (based on past performance) overperform. The problem for Cruz is that A) there are only a handful of closed caucuses left and B) in primaries–even closed ones–Trump tends to do significantly better.

  • 538 Refugee

    “Yesterday two autism therapists told me (snip) but I was amazed that the conversation went in this direction.”

    When speaking of Trump and autism therapists I thought that conversation was headed to a different direction too. ;) Feel free to delete this, some open doors I just have to walk through. :D

  • Olav Grinde

    If Trump is actually brought down, how fast will the pivot be from #NeverTrump to #NeverCruz?

    Froggy, that’s a good question. But I am even more interested in the pivot to #NoneOfTheAbove.

  • Art

    What is Cruz’ objective? I don’t see him getting to 50% nor winning a brokered convention. Maybe the VP, if Trump doesn’t get to 50 on his own?

    • Art

      PS hope you’re enjoying the dim sim!

    • Josh

      Maybe Cruz believes that, because there’s not a lot of daylight ideologically between himself and Rubio, if he can lean on Rubio to drop out, a disproportionate share of Rubio voters will go to Cruz.

      The problem with this line of thinking is that, if Sam’s logic holds, and roughly 1/3 of Rubio’s following is split between Kasich, Trump and Cruz, you have, nationally, something like Trump 43%, Cruz 27%, Kasich 17%, with ~12% undecided. Even if all 12% of those undecideds broke for Cruz he’d still be down 4% to Trump. And the majority of the remaining states should be more favorable to Trump.

      Cruz definitely had a better night than average–that’ll happen. Was it a three- or four-sigma above average kind of night? I’m not sure, but it doesn’t seem that way. It still seems like Trump is in the driver’s seat.

  • Ian

    There is a huge movement towards Cruz. He overperformed in all 4 states tonight — not just Kansas. And there’s significant evidence that except for early voting, Trump would have lost Louisiana — where he was ahead by 20+ points in the polls. Sam, I think you are really missing a key dynamic here.

    • Sam Wang

      I think I need to calculate whether this is a 1-2 sigma event, which is within normal limits. Also, if it’s real, it should also show up in national surveys as an uptick for Cruz.

  • Mark F.

    Not a bad night for Trump, and a good night for Cruz. Michigan and March 15 should be interesting. Not sold on Cruz having much chance right now despite tonight’s results.

    • Josh

      I concur. Nate Silver however characterized the night as “2/10” for Trump and “9/10” for Cruz. It sucks to see someone you admire shilling for page views like that but c’est la vie.

    • David D.

      Slightly off-topic, but what happened to Nate Silver and 538? I don’t remember him being so sensationalistic in 2008 as an independent blogger or in 2012 with the NYT. Is ESPN turning the screws? Is he getting bored with presidential politics?

    • Mark F.

      I’d say it was 5/10 for Trump, 8/10 for Cruz and 1/10 for Kasich and Rubio

    • Josh

      @David D: Maybe both? Working for yourself and nobody/nothing else definitely discourages or at least de-incentivizes sensationalism.

  • Mark F.

    So, what happens if this becomes a real 2 man race after March 15?

  • CalStateDisneyland

    If Rubio decides to withdraw from the race, can he release his delegates on the first ballot and let them vote for anybody?

    • Olav Grinde

      I have been wondering about that, too. Likewise I have wondered about the possibility of a Trump–Cruz deal, which would surely clinch the nomination for Trump regardless of any shenanigans by the GOP brass.

      It was not entirely meant as a bad joke when I suggested Trump could promise Ted Cruz a nomination to the Supreme Court.

    • 538 Refugee

      I believe ‘bound is bound’ and these votes will be cast as allocated whether the candidate is still in the race or not.

  • John


    I wonder if that will be the case going forward. Trump appears to have been given a real run for his money in Louisiana.

    I know the past polls don’t seem to indicate that, but isn’t it a mistake to assume that these aren’t dynamic races?

    • Sam Wang

      (John, use a real address please.)

      There could be a new dynamic. However, the race has been static for so long that we should look for conventional (i.e. not game-changing) explanations first. In this case, Cruz sometimes overperforms (see Iowa), and he might get Carson’s support. Recall that Cruz has a serious analytics effort, and he’s been working on getting Carson out of the race. Maybe he knows something.

      If Cruz picks up strength in surveys post-Carson-suspension, that would support the idea I am describing.

    • Josh

      Sam said it much better than I could have.

      Also, re: Louisiana, people are making hay about the fact that pre-election day voting was more favorable to Trump and using this to argue that the race has fundamentally changed. An alternative argument might be that 1) Carson was in the race up until Friday, and 2) Trump supporters in general appear to be fairly passionate about their candidate, though I don’t know what the numbers say about whether or not there’s a link between this and voting early?

  • Richard

    I want you to compare the map of economic distress in the US:

    with this map that has the GOP winner by county:,_2016

    Other than the weirdness in New England and Cruz’s strength in TX and regions adjacent to TX, Trump is winning in the most economically distressed rural areas.

    One problem is that those areas are most concentrated in the Sun Belt, specifically the South, and most Southern states are already done voting. He may still pick up some states in the Rust Belt and AZ and that may power him to victory, but losing in ME has got to be a giant concern not so much because of the delegates but what that signals. There’s no reason for ME Republicans to turn away from Trump more than NH and VT Republicans, but they did so, it seems because of what happened recently. Maybe it was the KKK thing. Maybe the exploitation/con man attacks are working, but it signals a turn away from Trump. With Cruz consolidating non-Trump support, the Rust Belt may not be in the bag for Trump, which means that it will be a dogfight to 50% of delegates. And it may come down to the RNC delegates who make the difference.

    • Sam Wang

      Since we don’t have Maine polling data, how can we know if voters there turned away from anyone?

      I believe pre-election polls showed a 3-state average of 33% for Trump, which is what he got in those states. Therefore I think maybe he didn’t lose that much support.

      But by the same reasoning, Cruz did pick up a lot of support from someplace – more than the number of Carson supporters. I am a bit perplexed. Strategic voting? Last-minute deciders? A great turnout machine? I note that Cruz has outperformed before.

    • mediaglyphic

      could it be that in closed states, Trump’s ability to draw new participants is thwarted and cruz because of his organization wins. we don;t have a lot of data points, just a thought.

  • Amitabh Lath

    Although Maine was unpolled, I confess I was quite surprised at the results being so different from NH and MA. I suppose like many who go to school in the Boston area and occasionally drive up to LL Bean on a whim, I expected there would be some regional homogeneity.

    Or (and this would be more interesting) Maine’s results could indicate some time dependent phenomena? Perhaps Trump has a decay constant?
    Insulting popes is okay with the Republican primary electorate, but penis references are not? If only the pollsters would ask the really important questions.

    • Olav Grinde

      Amit, while I don’t know New Hampshire, I can assure you that attitudes and political sentiments in Maine are radically different from those of Massachusetts. My wife and I spend most of the week in Boston, but every weekend we drive up to our homestead in central Maine.

      With the notable exception of our next-door Sanders-supporting neighbors (ten minutes walk away), we quickly discovered that the friendly Mainers we met at dinners and social gatherings hardly share our political views. My excellent joke about Governor LePage was once met with a stony silence.

      The F**k Obama bumper stickers are ubiquitous; and a few miles down the highway, there is a tree with hand-painted political signs: Impeach the Dictator and Terroist Lover (sic).

      We have since learned that rather than politics, it’s far better to talk about the weather, venison recipes and deer hunting, ice fishing, chainsaw models and the best methods for curing firewood. And of course the incredible beauty of the land, local traditions, and apple cider (our first batch is almost done fermenting).

    • amit

      Trump the wise offered up the real reason why Cruz won Maine:

      He congratulated Cruz for winning caucuses in Kansas and Maine — the latter victory being predictable, Trump said, because it’s “very close to Canada, let’s face it.”

    • Olav Grinde

      And I suppose Trump’s explanation applies to Puerto Rico as well. Given its proximity to Cuba, he was relegated to third place – behind Cuban-American Rubio and Cuban-Canadian Cruz.

      However, shouldn’t Donald Trump win the entire Eastern seaboard of the United States (including Maine) … given its “proximity to Germany”??

  • bks

    Polls in Puerto Rico close at 1pm EST. Open primary, but Cruz has put the most resources into the fringe contests.

  • JayBoy2k

    In the last PEC thread, I was trolling for predictions and did not get many responses because based on scarce data, there were not a lot of predictors out there beyond Sam and 538. It would seem that Trump was the one who overperformed 538’s predictions:

    Quote”..There is an Analysis at 538 that seems to say Trump will lose Kansas and break even in the other 3, Huffpost shows the polls in Kentucky and Louisiana with nothing in the way of predictions.
    Clearly Trump “lost” the debate, but it would be hard to link these states beyond Louisiana – it is a Primary and Trump had a significant poll lead.”

    There should be a lot of data/polls on Michigan to measure Trump’s rise or fall based on GOP establishment attacks or poor debate performance. Just 2 days until Michigan and Mississippi.

  • EmmaAnne

    I seem to recall reading here that general election polls aren’t very predictive during primaries, and maybe not even before both conventions are over. Can someone point me to this discussions? I tried the search function but didn’t get anywhere.

  • Kevin

    Real Clear Politics had Trump +15.6 in its pre-Louisiana polling average, and the actual result was Trump +3.6. According to 538’s live blog, there was a real split between the pre-vote in LA (Trump +24?), which led the networks to call LA for Trump immediately, and the late voters, who seem to have flocked to Cruz and abandoned Rubio. This could reflect demographic and dispositional differences between early and election day voters, or a late-breaking realignment from Rubio to Cruz, or both. Or it could be a fluke? Although Cruz seemed to beat expectations, and Rubio seemed to fail to meet them, in the other states as well.

  • Bill Herschel

    Mercer, the principal backer of Cruz, is a mathematical modeler. The company he heads, Renaissance Technologies, is built around mathematical models of various markets. They all have PhD’s in Applied Mathematics.

    In some ways they are the antithesis of Trump’s approach, although we don’t really know what Trump’s approach is. In any case, Cruz’s people are not to be trifled with.

    The blacker the box, the more likely Cruz will win.

    • mediaglyphic

      i understand that mercer is backing cruz. is there any evidence that mercer or his folks are doing any data analysis?

  • bks

    It looks like Rubio is on pace to take all 23 delegates from Puerto Rico:

  • Richard

    Might Maine be explained by the particular nature of its caucus set-up? I read somewhere that it had something like 20 sites for the whole state. That would certainly highlight the need for both commitment and organization.

Leave a Comment