Thursday, November 10, 2016

Pollsters missed the call by failing to spot Trump's support among white voters


Pollsters are diving deep into their data to try to figure out why things went so wrong for them on Tuesday.

Answers may not begin to emerge for some time. But the results of the U.S. election in the closest swing states provide clues to what the polls might have missed — and how those misses led pollsters astray.

The error the polls made at the national level was both relatively small and enormously consequential. Hillary Clinton's margin over Trump averaged a little more than three points in the polls. As votes continue to be counted, Clinton could end up ahead in the popular vote by about one point.

You can read the rest of this article here.

9 comments:

  1. The L.A. Times Day break poll had support for Trump among "Whites" at 58% and had Trump well above 50% among this group for months. They had Trump at 40%+ among Latinos and Trump received about one third of the Latino vote. The L.A. Times Day Break poll had a large margin of error 4.5% but, Trump had statistically significant leads among many of the subsets: race; age; income; sex. The failure of the so called "polling experts"; including on this site to recognise and accept the L.A. Times Day Break poll results shows unfortunately, the ingrained and systemic partisan bias that inhabits the polling industry. Polling in many ways is more an art than a science however, this miss was caused by human error not an inability of the numbers to project the outcome.

    ReplyDelete
  2. Kyle H.,

    Please find my responses to your questions in bold type below.

    WGS,

    Secondly, where are you getting your numbers from? REAL CLEAR POLITICS. Michigan is still a fairly solid Democratic state. NO, IT WAS NOT (By Sunday) THAT IS WHY RCP MOVED IT FROM LEANING DEMOCRATIC TO TOSS-UP. THE LAST STATE POLL (Nov. 6) GAVE TRUMP A 2 POINT LEAD. The closest Trump got to winning it recently were ties in three polls in the last month, and one of those I believe is Ipsos' fifty state poll which is fairly notorious for its small sample sizes and odd results (Clinton winning Kansas, Trump winning Vermont). The other one is Google Consumer surveys, which also has crazy results. CRAZY RESULTS THAT TURNED OUT TO BE MORE ACCURATE THAN ESTABLISHED POLLS. IT WAS NOT THE POLLS THAT WERE CRAZY BUT THOSE WHO ARBITRARILY DISMISSED THEM (including Kyle Hutton). All other polls show Clinton leaders (sic) ranging from 1% lead to a 19% lead, with the recent average about 4%. Now, that's a close race to be sure, but it isn't "tossup" territory, especially when the lead is still reliably for Clinton. ACTUALLY KYLE, ANYTHING LOWER THAN 5%+ IS GENERALLY CONSIDERED A TOSS-UP IN CANADA, AMERICA, AUSTRALIA AND THE UK. THEY ARE OFTEN REFERRED TO AS MARGINAL CONSTITUENCIES, SEATS etc...

    The last time Trump led or was even close to Clinton in New Mexico was the end of August. TRUE. BUT HAD YOU LOOKED AT POLLS AFTER AUGUST ONE WOULD HAVE SEEN A NARROWING RACE THAT DIMINISHED CLINTON"S LEAD TO ONLY 2 POINTS BY NOV. 6th WITHIN THE MoE. THAT IS WHY RCP CHANGED NM FROM LEANING DEMOCRATIC TO TOSS-UP.

    Florida is a genuine toss-up, but Trump's campaign also thinks they can win or come close in Miami-Dade, which currently has something like a 150K lead in early votes for Democrats. THIS IS UNTRUE! ALL BALLOTS ARE COUNTED ON ELECTION DAY.

    New Hampshire actually oddly enough not trending for Trump was much as Iowa or Ohio are. Why would you use NH as your example when those other, much better states exist? SIMPLE: BY NOV 6TH IOWA AND OHIO WERE SOLIDLY IN THE TRUMP COLUMN. OHIO WAS AVERAGING A 3.5% TRUMP LEAD WHILE IOWA WAS CLOSER TO 6%. I CHOSE NEW HAMPSHIRE AND MICHIGAN TO DEMONSTRATE HOW POORLY THE CLINTON CAMPAIGN WAS DOING-TRUMP WAS CAPABLE OF WINNING "SECOND TIER" SWING STATE NOT JUST THE USUAL TOSS-UPS; OHIO, FLORIDA AND PENNSYLVANIA. Maybe you should actually do some work, I can be a better Trump shill than you. NO YOU CAN'T! YOU DON'T UNDERSTAND PEOPLE, POLITICS or POLLS.

    80% of Catholics voted for centre-left, pro-choice parties in Canada in the last, I don't know, six elections? NOT TRUE. ANYBODY REMEMBER HARPER'S 2011 MAJORITY? Not every voter decides based on their religion. The fact that 72% of Hispanics voted for the pro-choice Obama in 2012 should already show that to you, but oh my gosh 80%? What a leap. TRUMP WON ROUGHLY 1/3 OF THE HISPANIC VOTE!

    ReplyDelete
  3. I have re-produce most of a comment directed at me from Kyle H. as I believe it shows why and how so many people failed to call the U.S. election correctly.

    Yes, it is partly a case of mass delusion, however, it is more a failure of accepting change among polling numbers and an inability even hostility toward accepting numbers unfavourable to one's preferred candidate. Specifically, it was a failure to recognise a broader trend these numbers were demonstrating. When Trump closed within 2 points in New Mexico and closed the gap in New Hampshire and Michigan it says something; Clinton has not connected with these voters-she has not closed the deal even among states that should be solidly Democratic. So, a lead <5% is unexpected and significant. When this is multiplied across several states (as it was) then people should take note.

    Secondly, False information obscured and gave false hope to Democrats. Kyle H. noted that Democrats had a 150K vote lead in Miami-Dade. I don't know where this untruth originated. I am sure Kyle H. did not make it up because I heard similar "news" from media sources. Since, votes are generally not counted before election day such "news" should have been refuted immediately by professional news organisations and others but, the systemic favourable bias among the fourth estate toward Democrats meant they felt no need to refute obvious untruths masquerading as facts.

    Thirdly, when using polls one must treat all polls equally within reason. It appears in hindsight that many people news organisations etc...either ignored the last round of state polls or missed them entirely. It is a reminder that you have to cross the finish line, you can't run only 90% of a race.

    In closing, I would like to say I find too many people on this site overly partisan. Having a preference is fine. When you willfully block out what the other side is saying and disregard their opinions then you negate the purpose of a democracy: free expression and amicable debate.

    ReplyDelete
    Replies
    1. Religious voting patterns in Canada don't reflect those in the US. Canada is a majority Catholic country, but there is a very weak correlation, if any, to a left-right voting preferences, so much so that it's not a typical polling question here.

      Delete
    2. @WGS - Re the idea that the Democrats had a 150K vote lead in Miami-Dade: You are correct that the votes are not counted before election day. The confusion arose because the voter registration affiliations of voters were released daily in Florida (and some other states, such as Nevada), and THOSE are the numbers that showed the Democrats with a 150K lead. While most Democrats and Republicans can be expected to vote with their party, it's not always 100% true. There were also a significant percentage of voters not affiliated with either party, many of whom were Hispanic, and thus assumed to be Democratic (but not necessarily so). It would be more accurate to characterize Kyle's statement as a misinterpretation than an untruth.

      Delete
    3. Jay,

      In the context of Kyle H.'s comment directed at me it was closer to an untruth than a misinterpretation. Either way he should have checked his facts.

      Delete
  4. Eric

    I think it might be helpful for you to read an article by Sean Trende senior elections analyst for RealClearPolitics..

    He says the polls were accurate enough but it was a failure of the punditry.

    It was almost as though he was aiming this article directly at you.

    http://www.realclearpolitics.com/articles/2016/11/12/it_wasnt_the_polls_that_missed_it_was_the_pundits_132333.html

    ReplyDelete
    Replies
    1. I'm not sure why you would think that.

      Apart from the delusion that he would even be aware of the work I was doing, I always couched my analyses with the potential for Trump to win as the margin between him and Clinton was too narrow to be overly confident.

      I was certainly not of the camp that thought Clinton was 98%+ favourite to win.

      In the Poll Tracker, Trump won all of the "Lean" Democratic states except Nevada. The only other state he won was Wisconsin, which was projected as a "Likely" Democratic win. That is the major "miss" of the Poll Tracker.

      That he won the Lean states is not something that was considered impossible. Those states were projected with less than 67% confidence. But I didn't want to mark any states just as "toss-ups", I wanted to show who was favoured.

      Had I marked states just as toss-ups and did not portion them out to Clinton or Trump, I would have had Trump at 186 votes and Clinton at 237, with the remaining 115 votes up for grabs.

      In the end, that would have been the more "accurate" approach and would have prepared readers better for Trump's win. I don't think people would have appreciated that level of hedging, however.

      Delete
    2. I am afraid I did not look at the CBC's poll tracker once and am unfamiliar with this site's or CBC's forecasts.

      I do however, agree with BCVoR's statement that it was human error not poll data that is to blame for this prediction or rather projection "failure".

      As an example: "Meet The Press" on NOv. 6th had Clinton with 274 electoral college votes versus 180 for Trump with only 84 undecided. Hindsight is a wonderful thing of course but, a critical eye can not help but observe that NBC-MTP had only 6 battleground states and had moved two battleground states: Pennsylvania and Nevada into the Clinton column while not doing the same for Trump. If Pennsylvania and Nevada were given to Hilary why were Ohio and Arizona not given to Trump? More importantly they failed to include at least five states in their battleground column: Minnesota, Michigan, Wisconsin, Virginia, Maine, New Mexico.

      There seemed to be a general trend to give the benefit of the doubt to Hilary regarding swing states while keeping similar states as toss-ups for Trump.

      Secondly, pundits either willfully ignored or missed the last round of polls that had Trump surging in Pennsylvania, Michigan and Wisconsin.

      Thirdly, too much emphasis was placed on Obama and his relatively high approval numbers while negating the difficulty of the incumbent party to win a third term and the general anti-establishment mood that often emerges after a two term presidency. Since, the Second World War the incumbent party has only won the presidency once for a third term; George H.W. Bush succeeded Ronald Regan in 1989. Lyndon B. Johnson had the opportunity to do so in 1968 but, his unpopularity and that of the Vietnam War caused him to withdraw from the Democratic nomination in March of 1968 endorsing the eventual Democratic nominee and his Vice-President Hubert Humphrey.

      Had I not apportioned out toss-up states I had the following result: Clinton; 203: Trump; 186: Undecided; 149.

      Real Clear Politics came close; With a battleground contingent of 14 states they had Clinton on 203: Trump; 164: Undecideds; 171.

      Delete

COMMENT MODERATION POLICY - Please be respectful when commenting. If choosing to remain anonymous, please sign your comment with some sort of pseudonym to avoid confusion. Please do not use any derogatory terms for fellow commenters, parties, or politicians. Inflammatory and overly partisan comments will not be posted. PLEASE KEEP DISCUSSION ON TOPIC.