Princeton Election Consortium

A first draft of electoral history. Since 2004

Florida Man Starts To Flag

October 25th, 2016, 10:33pm by Sam Wang


Last week I suggested that Senate polls might move toward Democrats, as part of their general pattern of following Presidential-race ups and downs. The Senate Meta-Margin has indeed ticked in that direction, driven by small movements toward Democratic candidates in Nevada, New Hampshire – and Florida??

That’s a surprise. Marco Rubio had been assumed to be the best candidate to hold the seat for Republicans, which is why he got back into the race after his Presidential bid failed. But his median poll lead has shrunk to 2%. That explains the recent push by President Obama and other leading Democrats in Florida. It’s a twofer: Florida is also the site of several competitive House races, thanks to a Florida Supreme Court ruling that struck down several gerrymandered districts.

Florida activists on both sides how have a lot to keep them busy. With several races on a knife edge, their get-out-the-vote efforts will be especially powerful. Therefore I have added the Florida Senate race to the ActBlue page on the left. Republicans can give via the NRSC.

In other news, the Presidential Meta-Margin has ticked toward Trump by a fraction of a percentage point. This has no bearing on the win probability – he’s too far behind. But it does indicate that the race is, once again, moving toward its average. Truly this is a dull year, statistically speaking.

The House generic is also ticking toward Republicans. This increases the likelihood that Republicans will retain their majority. However, the GOP could still be in for a rough ride after the election. With a smaller majority and a more extreme caucus, they might not be able to hold it together to elect a speaker. Jeff Greenfield at The Atlantic speculates.

The title of this post is a play on the Florida Man meme.

Tags: 2016 Election · House · President · Senate

72 Comments so far ↓

  • Jeremy

    Think I’m about done with the “other” site. The batch of Quinnipiac polls released today all showed Clinton with moderate-to-significant leads in three swing states, while tied or nearly with Trump in two states he has been projected to carry, including in past Quinnipiac polls, yet her win likelihood percentage somehow managed to trend downwards.

  • David F.

    Shawn H. said to watch for “the expected domination of Clinton’s turn-out operation over the GOP’s almost non-existent one. I expect Clinton to out-perform her projected margin by 2-3%.” I agree this could be one of the more interesting post-hoc investigations: what difference can GOTV superiority make? In the past, the presidential candidates have enjoyed rough parity in GOTV, so the effects have washed each other out. But this time around we should get a good read on the underlying potential of GOTV by comparing polls-vs-actuals in states with and without a competitive Senate race. (I imagine HRC’s effort is best in class, Trump’s is non-existent, and RNC’s is good but not great.)

    Looking forward to this.

  • Ed Wittens Cat

    via the indispensible Drew Linzer–
    Trumps data science ops
    !!Project Alamo!!
    hahahahaha
    https://www.bloomberg.com/news/articles/2016-10-27/inside-the-trump-bunker-with-12-days-to-go

  • M. Leo Cooper

    Sam,

    I can’t help wondering what effect the early voting could have on the forecasts and whether it’s factored into the models. Seems to me that based on the number of ballots cast or requested by registered voters of each party that a pretty good guess could be made about how many early votes each of the candidates have gotten. And, these are voters now off the table, not undecideds. Do the polls take this into account — do they just poll voters who haven’t yet voted? And what does this mean for your model, Sam?

  • A

    Sam has said over and over that this race is very stable, boring even.

    It continues to be so, even with a bit of a MM drop the last couple of days. There’s been some noise in the recent polls, but the reality is that not much has changed.

    When you’re nervous as most of us are, every little bad poll feels awful–but it’s not real.

    Let’s keep our composure and listen to Sam’s wisdom. He isn’t worried, and he knows a lot lot more about this stuff than we do!

  • Richard

    High anxiety.

    Sam Wang: Clinton 96-98%
    Huff Pollster: Clinton 97.2% (gotta love the 0.2)
    Drew Linser (Daily Kos) 95%
    NY Times Upshot: Clinton 92%
    The other guy at 538: 83.7-87.2%

    What’s it all mean? Clinton will very likely win but there is a better than 1% chance that Trump will become President. That scares the shit out of me, as it should everyone. Can’t wait for this horrible election to end. Oh, and the Senate is too close to call.

    • Jay Sheckley

      “a better than 1% chance that Trump will become President. That scares the shit out of me, as it should everyone. ”

      Cmon, Trump’s the GOP nominee. That ought to be good for a greater chance. To me the margin [and metamargin] are scarier, because people don’t understand that’s what to worry about before you “protest vote” or don’t vote. (Sorry people: Not showing up ain’t protest, it’s abdication.)

      Though I admit wondering if the 99% banner means Gary Johnson is as likely to win, or that Sam isn’t using decimal points.

  • David Auerbach

    There is some awful vote suppression going on in (at least) Mississippi and Indiana. In Indiana, voters trying to vote early are told that their registration is still “pending” and they need a letter from the DMV (I think) that seems to never arrive. This despite having gone through the registration process and having all the requisite ID. Similar foot-dragging techniques in Mississippi.

    • Anthony

      You would think democrats would have civil rights lawyers ready to go on call. We all know these kind of things will happen months (years) in advance. Republicans have been public broadcasting they intend to suppress the vote.

    • Matt McIrvin

      In Indiana the state police also raided an organization doing voter registration on grounds of a phantom nationwide voter-fraud conspiracy.

  • Mark F.

    I understand UK pollsters underestimated the Conservative Party vote by an average of over 6% in their last election. Is that possible here? That amount of error could mean either a narrow Trump victory or a Clinton mega-landslide. I suppose that’s why Sam has the odds of Clinton winning at “only” 96-98%.

  • Anthony

    I’m starting to worry about the MM, dropped from 5.1% to 3.9% in a matter to days, still two weeks to go until the election.

    • Ravilyn Sanders

      There is no reason to panic. Such changes happened before, sometimes glitch in data feed, some times it is an old good result expiring due to age …

      Sam will tell us if (and when) to panic. So far he does not seem to be asking for good recipes for bugs. So stay calm. And vote.

      If you really have to worry and despair, there is always the gerrymandered House to fill us with hopelessness and despondency.

    • Rob in CT

      I’m not worried, but I am frustrated. I have been waiting for Clinton & the Dems to pull away and it’s never happened. Sam’s “it’s 2012 redux” take looks rock-solid, and that’s almost infuriating, given who the GOP nominated.

    • Kanwaljit Singh

      ROB in CT, it’s a symptom of extreme polarization maybe? R’s can nominate a marmot and the final results will likely be similar to the ones we are seeing now.

    • Matt McIrvin

      The main difference from 2012 is the shape of the endgame: the 2012 Presidential race became scary close after the first debate, which was on October 3, then Obama recovered. And it’s still slightly mysterious why he seemed to recover in Sam’s state-poll aggregate before it was visible in national polls (or in some other state-poll aggregates, even).

      Also, the Democrats were starting from a slightly better position in the Senate.

      The situation now looks more like the endgame in 2008, when Obama progressively dominated more and more with each of the debates, except that we’re looking at more 2012-like margins.

    • truedson

      The probability went from 99 to 98 is another way to look at it…..Meanwhile CNBC got all excited today when one, ONE, poll showed Trump ahead by 2 in Florida.

    • Paul Ruston

      I am also not worried and also frustrated. I thought the PEC MM and senate race margin would from this point forward keep looking better for Clinton and the Dems. For comparison sake this is what PEC 2012 presidential election eve margins compared to where we are today.
      2012 10/26/16
      CO +2 +4.5
      FL T +3
      IA +2 +1
      NV +2.5 +4
      NH +3 +9
      NC -1 +2
      OH +3 T
      VA +2 +7
      In 2012 the MM from 10/26/12 through election day 11/6/12 ranged between +1.78 to +3.02. I think the final MM was +2.46 99.2%

    • Matt McIrvin

      The slight retrenchment is actually another similarity to 2008: there was a little drop in Obama’s margin and Meta-Margin right toward the end. (Before that, he was getting some crazy polls where he was tied up in places like Montana and North Dakota.)

      So this endgame seems kind of like 2008, miniaturized to 2012 size. This year, the key event seems to have been the first debate, which finally made it so obvious that Trump was an ignoramus that a lot of marginal voters actually paid attention. Back then, it was the collapse of the entire global economy, McCain looking feckless about it, AND the awfulness of Sarah Palin landing around the same time. Looking at it that way, I can understand the relative scale factor.

    • Shawn Huckaby

      Two things to watch for: One is that Obama’s job approval rating is back on the upswing, which may bode well for more movement upward again, or at least prevent further erosion.

      Two, although not visible in advance, is the expected domination of Clinton’s turn-out operation over the GOP’s almost non-existent one. I expect Clinton to out-perform her projected margin by 2-3%.

    • jculb69

      It could be mean reversion. The current +3.9% is right on the linear trendline from the tabular data.

    • Slartibartfast

      Shawn,

      There is also the possibility of a Bradly effect for Clinton adding another couple of percent to the margin. In a 2010 study it was shown that females in Senatorial and Gubernatorial races over-performed their polling by better than 2%.

      I would definitely take the “over” on the meta-margin (the actual results being more pro-Clinton) as there seem to be more plausible reasons for Clinton to do better on election day than in the polls than for Trump to do likewise.

    • Mark R

      Seems like there’s a recent divergence between national polls (especially those released today) and the meta-margin. Haven’t been paying attention but does anyone know offhand which states she’s been slipping in? Conversely, I assume she could be gaining in states she has no hope of winning as well.

    • Matt McIrvin

      It was looking like Clinton was leading in Arizona; now it’s leaning pink again. There was slippage earlier in Ohio, and there was one Bloomberg poll with Trump up in Florida that probably shifted the median slightly less solid blue. Also all that giddy pinkness in Texas is gone. On the other hand, there seems to have been a gain in Iowa.

    • alurin

      I suspect that the difference between the meta-margin and the national polls is that Clinton is doing a lot better in states that are normally deep red. Texas is the paradigmatic example, If Clinton trails by less than 10 points in states like Texas, that’s going to boost her national total, relative to Obama’s 2012 performance, while not having much influence on the meta-margin, which is driven by the states that are closer to the tipping point.

  • Jeff Alworth

    This perhaps oversimplifies things too much, but: it seems that whenever Trump spends time in front of a national audience, his numbers tank and then only start ticking up once he’s been out of the public eye for a while. The very best thing he could do, then, is hole up in Trump Tower for the next two weeks.

  • Slartibartfast

    What is the cause of the drop in the meta-margin? Is it all from Remington Research dump or is there tightening in other polls as well? I had expected the meta-margin to drift downward, but not quite this sharply.

    • MNP

      Republicans are coming home again. There hasn’t been fresh atrocities to shake them loose of Trump so they drift back showing stability. Whether this is partisanship, something HRC especially gooses or Super PAC spending who knows.

      Senate Dems are mostly running behind HRC.

  • Arthur Neelley

    I have to say I’m pretty shocked to see the Meta-Margin shrink so rapidly in the past two days. I recall about 5% the other day, then 4.4% yesterday and then today… Pretty scary and I pray this does not continue. Any thoughts as to which states’ polling are causing this ?

    • Matt McIrvin

      There is not any trace of a drop in the national numbers, so I’m thinking this is mostly noise. In any event, there’s so little time left that even if this were the beginning of a sustained slide similar to what we saw the first half of September, Clinton would be set to win handily.

  • James

    I enjoyed watching Newt argue with Megyn Kelly about the polls (amongst other things). It is so reminiscent of Karl Rove’s meltdown four years ago and the unskewed polls website. Now Trump’s people have the added charge that all the mainstream news sources are biased against trump, even Fox News. Now I see in the news this morning that Trump and his surrogates are arguing that the polls don’t take into account his supporters’ enthusiasm and new voters. Even Fox News is running “Early voting suggests tight race in key states despite Clinton camp boast.”

    The main difference this year is no war on Nate Silver (aka in this comment section as “the other guy”). Surely everyone will learn their lessons on who to trust once the polls are in.

    • Sean

      The far right is getting more extreme. It’s to the point where fox news is no longer right-wing enough for them. I think all the allegations are in part to build support for a new TV network launched by Trump & Ailes.

    • Deb

      Well I suppose attacking fox now allows them to steal that audience over to the new media empire.

    • Matt McIrvin

      Instead of blaming Silver, Trump and Roger Stone are saying outright that the polls are all rigged by some fantastic conspiracy run by John Podesta.

  • anonymous

    Could the DNC not have kept their lower cost ad time in Florida, and made the possible withdrawal argument to attract donations for the other races, before actually withdrawing? Nevertheless, I am not convinced that the future presidential threat of Rubio is a good enough reason to lose Senate control this time around (i.e. I tend to agree with Chuck Schumer’s assessment of the state of affairs). This is primarily because I now have increased confidence in Hillary’s ability to school the Republican contenders, and the likelihood of extremists still controlling the party in 2019. The Supreme Court is going to be moulded for a long time in the next four years, so the time for Democratic Senate control is now.

  • Michael

    PEC Contest: Which will turn out to be the biggest canard of the election season: The LA Times/USC Poll, the Investors Bus. Daily Poll, “Brexit,” Trump’s Bradley Effect, Clinton’s Bradley Effect, Trump’s “Hidden Vote,” Trump’s rallies, One of several October Surprises, Rigged polls, Rigged elections? If there were a way to count votes from readers, it would be fun to know which gets the most.

    • Ed Wittens Cat

      The biggest canard is that polarization is 2D.
      we are seeing polarization on multidimensional axes.
      one axis is education, another is gender–its complex– we are seeing traditional republican demos moving to democrat, like white females and white college educated men.
      thats why i like simulated annealing in multiple dimensions as a model. Is college education a new polarization axis? Is it persistant over time?

    • Ed Wittens Cat

      When this whole Trump episode started i ax Dr Wang if a movement could survive wholly bereft of intellectuals.
      I do not think the GOP can survive.
      but this isnt new.
      http://www.3ammagazine.com/3am/calvinist-roots-american-anti-intellectualism/

    • Scott J. Tepper

      Votes from readers? You mean — a poll?

    • Lorem

      How can we tell which one’s the “biggest”? It seems likely that all of them will have approximately no effect (or be an outlier in the case of the polls). Google search result counts?

      (And if one does have an effect, it’ll be really tricky to disentangle it from other factors.)

    • Ed Wittens Cat

      this helped me think abt the nuances of polarization–
      http://www.newyorker.com/magazine/2016/10/31/hillary-clinton-and-the-populist-revolt

      more and more i only care the Math– could Jefferson’s ploughmen learn to understand Math?
      would that immunize them to carny-barker exploitation?

    • Ed Wittens Cat

      i guess that makes me an educationist :(
      maybe the axis of polarization is like Packer claims– upwardly mobile and the opposite–
      down, down, down

  • ravilyn sanders

    At the beginning of the season, there were many incentives for the pollsters to put their thumbs on the scale. Outlier results garner media coverage. Consistently bluing or redding (I verb all words) the results builds a fan base. Attracts funding and support from political factions with axes to grind. etc etc.

    But as the election draws near, they need to get the answer right, to earn some credibility and to crow about their accuracy etc. So all the pollsters who were intentionally biased their results will slowly wring it out of their system and try to get the results match the election. They will blame the sudden swing on, “late breaking developments” and “voters deciding in the last week”.

    But is it possible to work on the moving averages and other statistical metrics and calculate which pollsters were intentionally biasing their results, which way and by how much?

    I can only wonder. May be there are others who can answer?

    • Bill G.

      > But is it possible to work on the moving averages and other statistical metrics and calculate which pollsters were intentionally biasing their results, which way and by how much?

      One facet of Sam’s model is that it does not feature any sort of “balancing” of polls. If I remember correctly, the explanation for this is that existing house biases will tend to cancel each other out over the long run (an assumption that I, personally, agree with). By taking the median of a large number of polls you can get a good idea of where the race is at in a given state.

      If, as you say, there is some sort of wringing out of house biases as the election draws to a close, then those firms’ results should move towards to median, which PEC has been using to make its predictions from the beginning.

      This is all based on my understanding of the model, I’m sure someone can add more.

    • Sam Wang

      The house effects have to cancel out, since the only benchmark for calibration is other houses. The median takes care of this, not always as well since not every house covers every state. But it is more stable than other approaches, which is empirical evidence that it works.

    • Charles

      There is no reason to believe pollsters are intentionally biasing results. Pollsters are not successful because they get media coverage, they are successful because they get it right.

    • Olav Grinde

      Charles, I respectfully disagree! I do suspect some “pollsters” are in the business of influencing public opinion – rather than faithfully and accurately measuring it.

      And as long as they ease up on “putting their thumb on the scale” just before Election Day, they can have a credible claim to “getting it right.”

      However: as Bill points out, PEC’s median-based model is not unduly influenced by such outliers and biased polls. Sam doesn’t make “corrections” because with his model there is little need to do so!

    • alurin

      Some pollsters do seem to have consistent house biases (e.g., Rasmussen), and some pollsters are more accurate than others (e.g., PPP). But I don’t see any evidence that pollsters with a house lean are changing that lean to “get it right” closer to election day.

    • Erik N

      It is certainly possible that pollsters are underestimating Latinx voter turnout in Florida.

      In Florida, they made up 15% of the electorate in 2008, and 17% in 2012. As per the NY Times, 5 pollsters are giving a turnout estimate of 13%-15% for 2016.

      http://www.nytimes.com/interactive/2016/09/20/upshot/the-error-the-polling-world-rarely-talks-about.html

    • Matt McIrvin

      Rasmussen has a reputation for making their Republican house bias go away at the 11th hour, but that didn’t happen at all in 2012. I suspect they’re not doing it intentionally.

    • Josh

      I guess in theory what you’re saying makes sense, but…do you have any actual evidence of it? It seems to me like by far the most important incentive for a pollster is…accuracy. You’re basically alleging that pollsters would intentionally undermine their own accuracy for the sake of building a cult following?

    • Ravilyn Sanders

      @Josh // Oct 26, 2016 at 11:40 am
      “You’re basically alleging that pollsters would intentionally undermine their own accuracy for the sake of building a cult following?”

      Their accuracy is measured only when the election happens. So there is incentive for them to be click baits early in the cycle, and switch to “true mode” to gain accuracy ratings.

      I don’t have any evidence, I don’t have the math fu to tease it out of rolling averages and correlation coefficients. That is why I asked people who would know more/better.

    • Froggy

      There was an article in the Washington Post yesterday on the IBD/TIPP pollsters. They change their likely voter screen about two weeks before the election (tightening it), and then go on to “tweak this model further, customizing thresholds for older and younger voters.” That provides the capability to adjust the poll results before election day to make them agree with other pollsters. It doesn’t have to be done deliberately with that goal in mind, or even consciously.

    • Josh

      Right. I totally buy–and there’s clearly evidence of this–that pollsters will change (or “tweak”) their screens or modeling or whatever as the election draws near. But what I see no evidence for whatsoever is that the only reason pollster do this is because they’ve spent the previous several months publishing results in order to appeal to groups of like-minded followers.

    • Charles

      “So there is incentive for them to be click baits early in the cycle”

      No, there’s not. Pollsters aren’t ad-supported. They don’t make money per page view. And people look at the accuracy of their entire polling history, not just the last few before the election. Not only do you have zero evidence for your crackpot conspiracy theory, but it’s a theory that doesn’t make logical sense from any perspective. This kind of nonsense doesn’t belong here.

  • Ed Wittens Cat

    will there be Vote Protectors in Florida?
    nice scoop by HuffPo
    http://www.huffingtonpost.com/entry/vote-protectors-voter-intimidation_us_580e4e63e4b0a03911ee03bc?ncid=engmodushpmg00000004
    In Colorado its technically illegal to even post a selfie with ur ballot because of anti-voter influencing laws.
    Q: Will the news media discuss Stone and the Vote Protectors or breathlessly dwell on the ACA “October Surprise” and leaked emails?

  • Jerry

    You mention GOTV (get out the vote) in Florida. But given that Hillary Clinton has a good GOTV effort nation wide and Donald Trump does not, I wonder if the polls might be a bit misleading. Has there been any studies about how much GOTV matters? And I wonder how strong the Republican GOTV effort is given the situation?

    • Ken Arnold

      Charles,
      Re: Pollster motivation
      Many partisans are most comfortable in a friendly echo chamber, which means that pollsters and media outlets will garner audience in return for biasing results. If “getting it right” means making the proper call in early November just before election day, then a pollster can optimize for both audience and track record by “intentionally biasing results and then adjusting in the final days to get the results match the election. Exactly what Ravilyn suggested.

    • bks

      Is there some evidence for these last-minute adjustments?

    • llywrch

      I’ve seen reports that activists believe a good GOTV effort can move the results as much as 7%, but in the past both sides in a contested election will make a similar effort to get their supporters to the polls, so no one really knows. I’m guessing that in this election, it’ll be 2-3%, but then you have to account for Republican disillusion with their candidate, so it could actually be more or less than that number.

      We’ll have one possible datum November 9, but I expect the discussion will continue long after that.

  • Frank

    “The race is moving toward its average.” Using your tabular data, and starting where they start back in May, the Presidential meta-margin would achieve its average on Election Day if it (linearly) drops only another 0.4, to 3.7. From its second peak (two days after the final debate) of 5.1, this would be a 27% drop. In any case, how much movement is typical in the final two weeks? It’s hard to tell from the great graphs in your “How Should Volatility Be Defined?” posting.

    • Matt McIrvin

      Electoral-vote has graphs of how their aggregator evolved over every presidential election from 2004.

      The only one where there was really critical movement right near the end was 2012, when Obama made that recovery after the first debate, and I think Sam’s aggregator registered that a little earlier than theirs did (this one makes it look like Hurricane Sandy saved him, which I don’t think it did).

      This year looks more like 2008, though the scale is a little smaller–the current situation is more comparable to the 2012 final count.

    • Matt McIrvin

      …Though the polled lead was swapping around right in the home stretch in 2004, because it was so close. A lot of that was probably noise.

  • JPI

    Florida Man thinks the Speakership should be decided by arm wrestling.

  • Matt McIrvin

    Whoops, Senate Meta-Margin now back down to D1.4%. I guess there was a worse poll in NV.

  • GM

    This is great news! For John McCain!

    (Sorry, sorry, but we only have a couple of weeks until it’s retired for another four years.)