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


March 6th, 2016, 5:28pm by Sam Wang

In Saturday’s Republican elections, Ted Cruz outperformed his pre-election polls in all three states where data was available: Kentucky, Louisiana, and Kansas. This is in contrast with Donald Trump, who performed within 1-2 percentage points in Kentucky and Louisiana…but underperformed by a remarkable 12 percentage points in Kansas. What’s going on?

Looking at the graph above, Cruz has always outperformed his polls – in 17 out of 18 states so far. The higher he is positioned in polls, the bigger the effect – and this weekend’s states were good ones for him to begin with. Overall, his eventual vote share is typically 1.3 times higher than his pre-election polls. That is a considerable difference.

It has been said that Cruz benefits from caucuses and closed elections in which independents cannot vote. However, when these features are present, he also runs stronger in polls. Overall we have three correlated variables – Cruz support, closed primaries, and caucuses – and so it is hard to say what is causative. One way to restate the phenomenon is to say that Cruz voters are almost 1.3 times more likely to vote than other voters (Trump, Rubio, and Kasich supporters) whom pollsters reach. Cruz voters are a dedicated lot.

The tendency for pollsters to undercount his supporters may affect estimates of what will happen in elections from now to March 15th, most of which are closed and/or caucuses.

That brings us to Marco Rubio. His home state of Florida’s primary is closed. The last three surveys (2/24-3/2) give a median of Trump 45%, Rubio 25%, and Cruz 15%. However, this weekend, Rubio underperformed by 6 percentage points in Kentucky, and by 4 percentage points in Louisiana. It seems that Rubio’s attempt to use Trump’s highly personal attack style has not worked well for him. Combined with Cruz’s tendency to overperform, there is some chance that Rubio will end up in third place in his own state.

Tags: 2016 Election · President

23 Comments so far ↓

  • JayBoy2k

    Nice post, Sam . Full of data and crisp analysis. It is what attracts me to PEC.
    It certainly feels like a two-way race, but the playing field is going to get better for Trump going forward with Northern/Western states with few evangelicals and a winner take all format. Surging will not make much of a difference unless you win outright.
    I note that 2 Californians, Ron Reagan and Arnold both endorsed Kasich today.
    Waiting for Cruz to make it a race .. or not.

  • A New Jersey Farmer

    Are we seeing some effect of late deciders moving to Cruz in the Saturday results? Let’s see if the pattern continues on Tuesday. If Rubio loses FL, I think he leaves the race.

  • bks

    There are about 1500 delegates yet to be apportioned and 20% (303) of them are on the final day of the primary season (7 June). A candidate does not have to be overly optimistic to assume that prior to that day the convention might be contested. Very few states are truly winner-take-all so each of the candidates can pick up dribs and drabs.

    • Sam Wang

      I guess. But to rephrase what I said earlier, the farther Trump gets above the 50%-of-delegates-so-far threshold, the less likely it will be that the remaining delegates can reverse the trend.

  • Amitabh Lath

    Is 1.3 the slope of the linear fit to all Cruz? I agree this looks fairly compelling, but a likely voter filter that fails just for one candidate’s supporters?
    What could make that happen?

    Also, a linear fit to Trump looks like it would probably have slope 1, but the intercept may have an offset from zero.

    I can conceive of an offset to the intercept being caused by new voters not being picked up by the LV filters (which I gather Trump is doing). I cannot figure out what could cause Cruz’s slope to be different from 1

    • Sam Wang

      1.3 is the median value of Cruz_actual/Cruz_poll, state by state. Effectively, this is a fit but with a forced zero intercept and logarithmic scaling of errors.

      Trump: probably picking up undecideds, who are present in similar numbers across states. If they divide equally, this would lead to a constant y-offset.

    • Matt McIrvin

      Isn’t the slope greater than 1 what would happen if his supporters are simply the most likely to show up and vote, as the original post suggested?

      You can think of everyone’s raw vote numbers as incorporating some multiplier less than 1 that gives the turnout probability. If Cruz’s multiplier is closer to 1 than the other guys’, that would give him a percentage advantage that increases with his level of support.

    • Amitabh Lath

      Now that you have this data you could test for last-minute migration of voters between candidates by looking at the Deltas (actual – poll) to see if they anti-correlate.

    • Matt McIrvin

      Incidentally, this kind of behavior is what I would have expected for Trump, on the basis of the early reports of Trump-mania. That it’s not happening for Trump is interesting and probably important for the general election. That it is happening for Cruz is also interesting.

    • Matt McIrvin

      …The turnout effect would saturate at high levels of support, and the curve would bend over to a lower slope near 100%. But Cruz isn’t polling with a large majority anywhere, so we don’t see that on the chart.

    • Matt McIrvin

      …Though for my hypothesis to have an effect this big, Cruz’s turnout advantage has to be gigantic! If his people turn out 98% and the other guys’ collectively at 70%, that would just about do it.

    • Bill Herschel

      I suspect that Cruz’ turnout operation is extremely efficient, targeted and financed. His “ground game”. It is also said that Trump does not have a ground game.

    • MAT


      Amendment One (prohibiting gay marriage) was on the primary ballot in NC in 2012. At this time I was on our county board of elections and noticed an abnormal amount of ‘deactivated’ voters showed up (I forget the actual legal term, but it basically means someone who is registered but hasn’t voted in 2 straight presidential elections and hasn’t responded to voter registration cards being sent out. They can vote, but there are a few more gyrations that take place, hence how I knew this was occuring).

      There also was a huge amount of undervotes (no vote registered on a ballot) for every race except Amendment One. Obviously, there was a significant boost in turnout for this single issue vote. I did an analysis later after the general election to see if these voters showed up in November – they overwhelmingly did not for the subset that I checked.

      Now, of course, I don’t know how these folks voted, but based on all the empirical evidence on hand, it wasn’t hard to come to the conclusion that the message on Sunday from the pulpit was to turn out on Tuesday and vote for Amendment One. Having done so, these voters disappeared again.

      I suspect something similar is happening with Cruz.

    • InmanRoshi

      It wouldn’t surprise me to see Cruz’ ground game operation and data analytics on a different level of sophistication than Trump’s. His biggest financial backer is “quant” hedge fund manager Robert Mercer. Mercer is also a partner in “big data” analytics firm Cambridge Analytica, who is c onsulting Cruz’ campaign. Mercer is part of a new breed of benefactor that isn’t just writing blank checks to a candidate’s CPAC, but taking an active role in managing the CPACs. You can read up on Mercer’s background on his wiki page, it’s quite impressive.

  • MartinM

    If Cruz is over-performing, somebody else must be under-performing, even if that somebody else is ‘Undecided’. Is there a discernible pattern to who’s losing out to him?

  • Anthony

    I will have to say I certainly underestimated Cruz. Im surprised he is racking up so many delegates. I still think he has no chance to actually win, I just though that Rubio would be the one in a solid 2nd place, not Cruz.

  • bks

    In the 50 years it’s been since I read Asimov I have to admit I’ve forgotten if Hari Seldon gave a figure for Cruz in Michigan tomorrow?

    • Sam Wang

      He said the candidates to get above the 15% threshold for delegates will be (n=4 polls) Trump 38.5%, Kasich 22%, Cruz 21%, but that considering Cruz’s history of overperformance that Cruz could come in second.

    • JayBoy2k

      Always searching for polls to calibrate results the night before elections. Thanks to Sam, we can compare results in Michigan. There seem to be expectations for both Kasich and Cruz.
      Few/No polls in Hawaii and Idaho, but the last Mississippi poll on Feb 29th had Trump by 24!! Hard to believe that he will perform that well.

  • 538 Refugee

    Best I can come up with is a whisper campaign amongst Cruz supporters to dupe pollsters so they can always look to be over performing. Keeping that quite and still reaching the random sample of those polled has got to be one long odds shot.

    The only other thing I can think of is to look where that difference is coming from. Undecided will always under perform.

    • 538 Refugee

      This from Fox.

      “There may be another factor at play, though, in the contests thus far – and that is Cruz is performing better in the smaller, caucus states where organization matters most.”

      They have a map and it shows all but two states west of the Mississippi are closed primaries.

  • Eric Walker

    Somewhat off-topic, but: are there any reasonably reliable numbers that show (or even imply) whether Trump is drawing any formerly Democratic voters in blue or battlefield states? (I see no point in looking at states already behind the Red Picket Fence.) That is a roundabout way of asking whether a Trump candidacy is any realistic threat to the famous Blue Wall. I have not seen any such numbers anywhere.

  • MAT

    and now for something completely different! Picked up an interesting tidbit today reading 538 and verified on the Green Papers. There are 108 unbound delegates on the GOP side, from North Dakota, Wyoming and Pennsylvania. That’s enough of a block to make a difference if things are close come convention time.

    2nd, Florida binds their delegates for 3 ballots. They were really greasing the wheels for Jeb! and boy did it backfire…..

Leave a Comment