Princeton Election Consortium

A first draft of electoral history. Since 2004

Sound bites and bug bites

November 4th, 2016, 9:19am by Sam Wang


We’re entering a period when all the math and data gets converted to short quotes. The above quote is pre-Comey, but I will live with it.

I have some interviews this weekend. I doubt that I will be asked about median-based probability estimation. With luck I will get to the question where there’s real suspense: who will control the Senate?

Here are the weekend’s planned TV hits. I’ll update this as things develop.

All times are Eastern.

Friday: 6:00pm, CNN Money cancelled!. 10:00pm, MSNBC with Lawrence O’Donnell. 11:00pm , CNN w/Don Lemon to talk about this.

Saturday: 9:00am, CNN with Mike Smerconish (possible rebroadcast at 6:00pm).

Tags: 2016 Election · President · Senate

227 Comments so far ↓

  • MonteNOLA

    Sam, I am on the fourth election cycle with PEC. I can’t wait/am sad to see this end. From a keen observer, Thank You for your lucid comments, science and your stamina during this huuuuge election cycle!

  • Howard B

    Sam,
    if Trump is victorious many of us will relocate to an island where bugs are the main dish,
    but where democracy can begin all over again

  • Greg Gross

    Yep, the MM has been simply hovering for more than two days: 2.6, 2.7, 2.8, 2.6, 2.6…. if it’s moving toward the mean, it is doing so in a very meandering way, no?

  • Andrew C

    I am pleased by Dr. Wang’s dignified silence about Mr. Argent’s recent use of “trendline adjustment,” that is, abandonment of defensible method and crossing over into the dark side of punditry.

    Perhaps Dr. Wang has adopted as his motto Wittgenstein’s great sign-off “Wovon man nicht sprechen kann, darueber muss man schweigen.”

    Mr. Argent has claimed Dr. Wang is “dishonest.” Perhaps he should learn there’s a Japanese proverb that runs something like, “If you point at someone with one finger, you are pointing back at yourself with three.”

    • 538 Refugee

      Well, not everyone is silent about Mr. Argent’s methods this year.

      http://www.huffingtonpost.com/entry/nate-silver-election-forecast_us_581e1c33e4b0d9ce6fbc6f7f

      “By monkeying around with the numbers like this, Silver is making a mockery of the very forecasting industry that he popularized. ”

      The word “unskewing” is thrown in for good measure. I found it an enjoyable read.

    • Sam Wang

      Tell me he said that without naming me. It would be better.

    • Gelatinous_Cube

      I didn’t see “dishonest”–I saw “indefensible” used to characterize a 99% prediction.

      Not sure who started this, but Nate is coming off pretty badly in his replies. I wish all parties would cool it and consider whether there’s anything intellectually interesting about the disagreement (obviously, Nate does) and how we could resolve it. Obviously, when you’re estimating probabilities of a binary outcome, there’s no way you can be wrong as long as you don’t go all the way to 100%. State-by-state predictions could help, though, no?

  • Swami

    If it’s any consolation, some bugs are actually pretty tasty if prepared and seasoned well. Fried termites with chili powder, for example.

    Not that it will come to that.

    • James Orr

      Lobsters are just big bugs. If it does come to that, I’ll be happy to buy Dr. Wang a lobster.

  • Eric

    I’m interested in Alaska. They haven’t voted Democrat in a Presidential election since LBJ in 1964! Yet, in this election, Trump and Clinton are in a tie. There are some strong indications that it might well go into the Clinton column. What does everyone think of this? What does the tie in Alaska suggest about other states (if anything)?

  • Robert Johnson

    To prevent cruelty to bugs, I suggest that if Trump does win you should pick a bug which has already died of natural causes.

    • Neon Vincent

      I’ve suggested it before, so I’ll do it again. Eat a mescal worm. They’re already dead, they’re pickled so the are safe and they taste like booze, and they’re an insect, specifically a caterpillar. I’ve eaten them and they’re OK. If you need any help, take a shot of the mescal first.

  • 538 Refugee

    Nevada as a tossup doesn’t square with this assessment. It could just mean more people are voting early and the final results might still be in line with current polling data or it could mean a big miss. The Democratic party has a much higher favorable rating than the Republican party. I keep looking for evidence of that in the polls so I’m probably reading way to much into this. ;)

    http://www.nytimes.com/2016/11/05/us/politics/presidential-election.html

    “According to Jon Ralston, a political analyst for KTNV, high turnout from Latinos and Democrats in the state’s crucial Clark County is a bad omen for Republicans.

    Democrats outpaced Republicans in Clark County by 11,000 votes cast on Friday, Mr. Ralston reported, giving Mrs. Clinton a larger firewall than President Obama had in the state when he won by 7 points in 2012.

    Thousands of mostly Latino voters lined up late into the night to vote, meaning that Republicans could also face losses down ballot on Tuesday.”

    • Matt McIrvin

      I’m seeing people faulting Nate Silver for not incorporating this into his model somehow, but for all the justified stick Silver gets, this is an unfair complaint to me–a poll-based model is a poll-based model, and there’s just no legitimate way to tweak it to incorporate this kind of analysis without making a bunch of arbitrary assumptions.

    • DonC

      Polling is very hard in Nevada. Of course you could increase accuracy, but to do that you’d need more money and there isn’t an entity willing to step up to do that. This is no doubt known, but you can’t expect this site to change its methodology. If it did then it would be entering the world of punditry.

      You can always use the map!

  • Rudy

    Sam Wang: Saw all three shows and congrats on how on-message you were on a fairly complex subject and how you justified your conclusion to the extreme polarization that exists in our country! How simple and yet counter-intuitive!

  • Howard B

    On your podcast, Professor Wang, you made New Hampshire a canary in the coal mine. As New Hampshire is stuck in a tie, do these figure a lost night for the democrats?
    It’s almost as though Trump actually heard your podcast.

    • Matt McIrvin

      I think the idea with New Hampshire is that it provides calibration for any national swing across the entire model. If the model itself says NH is tied, and results reveal that it is in fact really close, that suggests that the rest of the model won’t be way off.

      I wonder, though. It appears to me that NH going to a tie suggests that we may need to pay attention to what’s abnormal about NH: It’s a true battleground state that is extremely white. That may actually mean it has a swing not precisely correlated with the national one.

    • Josh

      I had a similar thought recently. In past elections I’d definitely have looked to New Hampshire as a bellwether, but because its demographics are now so different from those of the country as a whole, I wonder if maybe North Carolina or Florida–whose polls close at the same time as those in New Hampshire–wouldn’t be worth watching more closely instead?

    • Matt McIrvin

      What makes NH convenient is that it’s a small state without any huge cities in it, so results actually get reported early after the polls close. With larger swing states, there’s usually a delay in reports from the most urban and Democratic-voting parts of the state, so the state usually looks like it’s leaning very Republican until late in the evening.

      As long as most polling error is strongly correlated across the whole country, it’d still work–not as a “bellwether” in the sense of how NH goes goes the nation, but as a calibration for the accuracy of a national election model.

    • Matt McIrvin

      …Meanwhile, a lot of election stories out there seem to be reporting NH as the crack of doom for Clinton, which I don’t think is what Sam means at all.

    • Sam Wang

      It’s probably not really a tie. Anyway, we’ll see. North Carolina is also worth watching.

  • DonC

    It does look like the cake is baked. In Nevada the Democrats had a huge turnout yesterday, fueled by early voters in Latino areas. This gives the Dems a 72K+ lead in Clark County, which is where all the voters are. In 2012 the lead was a bit less than this and Obama won Nevada by 6% or 7%.

    This development inspired Jon Ralston, the well known political journalist, to tweet:

    “They’re bringing drugs. They’re bringing crime. They’re rapists. And some, I assume, will stop me from being president.”

    I believe we’re seeing something similar in Florida along the I4 corridor. No doubt also in Arizona (the Clinton campaign is so data driven that if she’s spending resources in Florida she thinks there is a payoff).

    • Matt McIrvin

      One difference in Arizona is that the post-Shelby County vote-suppression regime is probably going to have a major effect there. They closed a lot of polling places in majority-Latino districts.

    • anonymous

      Dr. Wang. NC is worth it because of early voting trends or overall polling.

    • Debbie Lefkowitz

      Similar reporting on turnout in Florida to NV. Huge jumps in Hispanic voters and many are NPA and new or not likely to vote folks which is really good for Clinton. The spate of concerts, Beyonce etc.. is also helping Clinton. Classic GOTV effort. concerts are free but attendees sign up and then are contacted and encouraged to vote. They even have ticket distribution near polling places. She’s running master class in GOTV. Also, word I hear from NH is there are so many Clinton volunteers, they have run out of areas for folks to canvass.

  • Sandy S.

    I can only hope Dr. Sam doesn’t have to eat a bug. Still, I wonder:
    The meta-margin is 2.7%. What are the chances that the polls are, collectively, off by 3% in one direction?

    Poll errors are not independent of each other. They all have a collection of similar assumptions, strengths and weaknesses. Polling errors of 3% in one direction have occurred before, from the ensemble of state polls summed nationally. (Didn’t this happen for the Senate in 2014?)

    Then there is the possibility of shifting voter sentiment, which follows national trends across all states.

    I hope Dr. Sam is right. But a 3% error looks a lot more likely than the 1% chance in the model. I’d guess more like 15-20%.

    • A Tai

      It certainly is not out of the question that the error will turn out to be 2 to 3 pts. But if I had to make a bet, I think it is more likely the error will be in favor of the Democrats due to the fact that the polling results do not factor in the their superior ground game. If I remember correctly, Trump generally under-performed his polling during the primaries and while that is no guarrantee, I would not be surprised that the same thing will happen when it is all over.

  • Andrew C

    As much as I love the back-and-forth between Mr. Argent and Dr. King, particularly royal counter-punches such as the one on Twitter yesterday, what I wish that folks would learn from both of them is that data matters, facts matter, method is important, and that reason is still the best tool we have for making decisions about this complex mutual endeavor we call the USA.

  • J. WALKER

    Fantastic interview with Smerconish! It is so refreshing to see some intelligent conversation about the state of the election in all this noise. Bravo to both of you on an excellent segment!

  • Glenn G

    CNN’s breathless, melodramatic polls coverage was driving me to distraction when Mssrs. Wang and Zelizer made their Don Lemon appearance Friday night. I’m not a scientist, but I know enough to be able to weigh trends, medians, and all those good things that really tell you something. Thank you, thank you, thank you. (And you upstaged Beyonce.)

  • Roger

    How much would it change the Clinton win probability to shave off 1 electoral vote of all her outcomes? Would it drop any/much?

    It was reported months ago, and the person is confirming that he still will not: one of the Washington Democratic electors, who is a Bernie Sanders supporter, states that he will not vote for her if she wins Washington. Under Washington law, he cannot be forced to do so. So, how ever many electors she wins, you have to reduce it by 1.

    https://politicalwire.com/2016/11/05/elector-says-wont-vote-clinton/

    • songsmith7

      I read that story, and it’s my understanding that there is no plausible way that HRC hits exactly 270 EV’s, which would make that single vote irrelevant.

  • Hem13

    Other than O’Donnell pronouncing the Professor’s name and the distraction of flashing lights from Beyonce’s, it was an excellent appearance!

    • Sam Wang

      It was pronounced correctly.

    • Jake Ingram

      “Wong” has always been the correct pronunciation for the Chinese surname Wang and, no, that’s not just an affectation to avoid, umm, penis jokes. I think people are rarely corrected on this because ethnic Chinese folks with this surname find it easier just to accept the incorrect pronunciation than to be correcting people all of the time, although I don’t presume to speak for Dr. Wang.

    • Stuart Levine

      I understand that “Wong” is the correct pronunciation. Of course, it could always have been worse. O’Donnell could have pronounced it “Wiener.”

  • Bob Spitak

    Great job Dr. Wang on both Don Lemon and Lawrence O’Donnell shows! I am a fan of both shows and also respect the work that you and Nate (@ 538) do.

  • Debbie lefkowitz

    Sam, super fun to see you on MSNBC and sharing a screen with Beyoncé! Your summary was clear and to the point. So nice to hear a nice rational explanation of the polls versus 538 clickbait.

    Debbie

  • Ed Wittens Cat

    watched CNN while waiting for Dr Wang and Beyoncè and i SAW David Axelrod SAY that their internal polls showed Obama with a +4!
    …because the OFA relational DB showed who had actually voted in the past instead of self-selection of likely voters.
    Natalie Jackson (Huffpo) is WRONG.
    Internal polls DO exist– and i know this because i worked on OFA.

  • Prairie Pundit

    Sharing a split-screen with Beyoncé! Professor Wang: Superstar!

    • anonymous

      I saw him too! With a pen in his hand for that extra scholarly look. WTG.

    • John Stevenson

      Yeah, that not a bad gig sharing the split screen with Beyonce. Nice job Prof Wang!

  • Andrew

    OMG Sam Wong on my TV set seeing him with Lawrence O’Donnell!! You are the man Sam!! With love from CT!

    • Ed Wittens Cat

      split screen with Beyoncee….
      doesnt get any better — the Beauty and the Brains
      ;)

  • emmanuel

    I stopped watching CNN’s electoral analysis. Sam, if Trump wins, get an extra bug for me. The meal is delicious when we are two at the table. Besides, my dietician recommended my diet be fortified with some bugs to improve nutritional in-take.

  • Rickey Bartlett

    I am one up on you Sam, I have promised to eat an Opossum if Trump wins…..

  • Alan Cobo-Lewis

    How bug-averse are you? Your current prediction puts P(clinton over 298 EV) at about .16. Thats almost 1 in 6.

    • Froggy

      We should be getting a special meta-bug-margin, and a no-bug probability.

    • anonymous

      I think the EV margin will stick. Even if she loses NH, that is only -4. So still at 308.

    • Arun

      Don’t you mean P(Clinton < 298)?

    • Alan Cobo-Lewis

      @Arun

      Yes, thanks, P(Clinton 240) = P(Sam eats bug) = .16 ~= 1 in 6 which is a real risk (esp to the bug)

  • Bob S

    Re: the original tweet, you now say you’ll live with it. Is that the same as saying you’d post the same tweet today or would you up his number? My other question is probably silly and if it is, feel free to say so. How confident are you that a Clinton win is >99%. I realize that saying you’re 100% confident of a >99% probability may be both unscientific and redundant, but if you were to assign a confdence %, what would it be? I have pretty much given up watching the MSM, but for you, Dr. Wang, tonight I make an exception. Thanks for all your great work.

    • Sam Wang

      To put it technically, the exact probability is a matter of parameter estimation. What is the effective uncertainty of the Meta-Margin? Based on 2000-2012 it appears to be sigma<1%, therefore the high win probability. As a comparison, I think the equivalent parameter at FiveThirtyEight has a value of about sigma=5%.

    • Roger

      Sam, In that context is Sigma the same as Standard Deviation?

  • Anne

    Dear Sam, Today according to 538 there were decent national polls, and not so great state polls for Hillary, and as a result on that site Hillary’s chances of winning went down. As I understand it your model uses only state polls, so why do your results not change so much? Thank you.

    • David

      Not to answer for anyone else (namely Sam)-but aside from the major differences in model, 538 uses/does 3 things that are really different:

      1.They adjust their polls via a methodology that is unclear to me-it is not just the “house effect” of the pollster. So they do not take the poll number at face value.

      2.They don’t use precisely the same polls. Sam I think uses the Huffington polls. Things like Google Consumer Surveys are not in the Huffington database but are in 538′s. Their polls are interesting, but hard to know their meaning-for example they consistently have Clinton with problems in states like MI/FL/SC but have her consistently winning in states like IA/KS/AR that she will not win for sure (KS in particular).

      3.Most importantly-they apply a potentially dramatic trend-line adjustment which partially comes from the national polls and partially from the trend in the state poll. So right now-all of HRCs running poll averages are getting massively adjusted down by 1.5 or more points.

  • Tony Asdourian

    My son notes that on today’s EV histogram, the “snapshot”, the red line is inside the blue, so the chances TODAY of Trump winning is >2.5%. Weird that the chances today of Trump winning are less than in 4 days, even with the “random drift” model, never mind the Bayesian. Will the snapshot really be at >99% by Tuesday? Seems hard to grasp.

    • Donna Gresh

      That same question was asked yesterday (not answered). I am curious as well.

    • Dmitriy

      According to Bayesian analysis, values are more likely to regress to the mean than to move away from the mean. Currently, Clinton’s poll numbers are lower than their average value throughout the season. Therefore, it’s more likely that in the next few days, her numbers will increase (i.e., getting closer to the mean) than decrease (i.e., getting even further away from the mean).

    • 538 Refugee

      Possible ≠ probable?

    • Tony Asdourian

      Dmitriy, thanks for the answer about the Bayesian number. I did actually grasp that, although it seems odd at this point to assume that the snapshot is highly likely to regress to the historical mean in the next 4 days, rather than follow the current downward trend. I know that Sam has pointed out Hillary has stayed between 2% and 6% all year, but other than the model saying so, I’m not sure why he (or his model?) is so confident that the current trend downward will abate.
      Furthermore, on the EV histogram the red line is within the blue, so the actual chance of Trump winning today looks like it is around 3% or more. How does that improve to the 2% listed in the “random drift” prediction? I would think it would stay the same over 4 days, on average. Is the difference just about rounding?

    • Froggy

      There’s an over 99% chance that the snapshot will be at >99% on Tuesday. And no, I’m not going to show my math for that.

    • Scott H

      I believe the >99% represents the chance that the MM > 0, which isn’t the same as “Clinton Wins”.

      Maybe the win percentage should be based on the convolution of the MM pdf (represented in EV space) with the EV pdf at the predicted MM.

      Far from election, MM uncertainty dominates. Close to election, EV poling uncertainty dominates.

  • Veronica

    Why is Hillary at 191 EVs?

    • Rob

      He’s only counting the states that are considered “safe” … the darkest blue states.

Leave a Comment