The polls weren’t that far off

Here’s Nate Silver:

But ignore that for now — elections, after all, are contested in the Electoral College. (Hence the name of this website.) So here’s another question. What would have happened if just 1 out of every 100 voters shifted from Trump to Clinton? That would have produced a net shift of 2 percentage points in Clinton’s direction. And instead of the map you see above, we’d have wound up with this result in the Electoral College instead:

screen-shot-2016-11-09-at-6-02-41-pmThat shift would have led to a 3.2% victory for Clinton, exactly what the pollsters predicted on average, instead of the 1.2% margin she’ll have after California absentee ballots are counted.

Earlier I pointed out that the Wisconsin polls were pretty far off, but it turns out the national polls were pretty accurate.  As with Brexit, however, the experts had trouble processing the idea that Trump could win.

Apparently the mistake was underestimating GOP turnout, which explains why the GOP senators also did better than expected.  Interestingly, the one poll that predicted a Trump victory (LA Times) did fairly poorly, with a 3% victory margin predicted.  That’s actually further off than my absurd 5.13% predicted Hillary victory.


Tags:

 
 
 

32 Responses to “The polls weren’t that far off”

  1. Gravatar of Lorenzo from Oz Lorenzo from Oz
    9. November 2016 at 15:25

    “Apparently the mistake was underestimating GOP turnout, which explains why the GOP senators also did better than expected.”

    If pollsters keep underestimating turnouts for a particular ideological direction/voting group, they still have a problem.

  2. Gravatar of Lorenzo from Oz Lorenzo from Oz
    9. November 2016 at 15:28

    With Brexit, I said on Facebook before the vote that it was looking bad for Remain, because undecideds were breaking for Leave. But you could tell that because of the way the Wikipedia site was averaging undecideds as well as Leave and Remain.

    With this election, I went with the polls’ aggregation because they did not tell you undecideds. Fairly clearly, the pattern with the undecideds is crucial.

  3. Gravatar of Gordon Gordon
    9. November 2016 at 16:16

    I saw an article from Vox late last night which described how the political science models were largely correct in their predictions. Maybe Obama should have heeded Christina Romer’s recommendation to fill the Fed vacancies with monetary policy doves.

    http://www.vox.com/2016/11/9/13571872/why-donald-trump-won

  4. Gravatar of Patrick R. Sullivan Patrick R. Sullivan
    9. November 2016 at 16:24

    Wonder what Merrick Garland is up to today.

  5. Gravatar of Patrick R. Sullivan Patrick R. Sullivan
    9. November 2016 at 16:27

    Since The Donald was open to the idea of privatizing infrastructure, maybe he’d be open to other ideas. Like NGDP targeting…if someone could explain it to him.

  6. Gravatar of Brett Brett
    9. November 2016 at 17:03

    So here’s another question. What would have happened if just 1 out of every 100 voters shifted from Trump to Clinton? That would have produced a net shift of 2 percentage points in Clinton’s direction.

    That would fit with what Kevin Drum has been reporting on in terms of results. Essentially, Hillary had declines in a wide number of demographic groups of voters, and a very small (1%) increase in the percentage of white voters over Obama in 2012. For all of Trump’s nastiness, black and latino voters gave him a higher percentage of their votes than they gave to Romney.

  7. Gravatar of Matthew Waters Matthew Waters
    9. November 2016 at 17:19

    The exit polls have been cited many times, but it just strikes me as wrong that Clinton did better among whites than Obama. The county-level voting data doesn’t bear that out.

  8. Gravatar of Matthew Waters Matthew Waters
    9. November 2016 at 17:41

    Also, RealClearPolitics has looked at these “missing white voters.” The general story is these rural, blue-collar white voters voted largely for Obama in 2008. The Iraq War and then the recession hurt McCain among these voters, enough to get the blue-collar voters to vote for Obama.

    In 2012, the voters simply didn’t show up. They no longer like Obama, but Romney didn’t have the populist appeal.

    In 2016, Trump really had the populist appeal. There are two reasonable explanations:

    1. Trump voters have a nonresponse bias. Both Trump voters and Brexit may be more suspicious of cooperating with a mainstream media poll. Even if the poll tried to weight responses for rural whites, the rural whites who DO respond may be more likely to not vote for Trump.

    2. The polls weighted based on what replicated the 2012 result. The polls in fact missed the other way in 2012, underestimating the Obama vote. The 2012 weights assumed whites would turn out similar to 2008 and the whites didn’t. The 2016 weights assumed whites WOULDN’T turn out like 2012 and whites did turn out.

    The second point describes the weights for statewide polls in the midwest states which missed horribly. The nationwide polls were closer to the nationwide results only because of an offsetting miss on Latino turnout. Unfortunately, Latinos weren’t enough for AZ and Latinos didn’t matter for CA and TX.

  9. Gravatar of bill bill
    9. November 2016 at 17:41

    I can’t find the old post, but IIRC 538 did a study a few months ago where 538 surveyed like 834 people with lots of questions, then gave the results to five of the best polling companies and asked them to review it. The polling projections from the 5 companies differed by up to 4%, I think. One example I remember is this: some companies assumed you would vote if you answered “yes” to these two questions (Did you vote in 2012? Do you intend to vote this year?), but other companies assumed you would vote based just on the second question. So even with the exact same data, the pollsters make did projections. It’s not easy work.

  10. Gravatar of ssumner ssumner
    9. November 2016 at 18:09

    Patrick, You said:

    “if someone could explain it to him.”

    That’s a big if . . .

    Seriously, that’s not the sort of decision that presidents typically make. No president decided on a 2% inflation target.

    Everyone, Try digging up those old Sean Trente articles on the missing white voters. He basically wrote the script for what happened yesterday. As Tyler would say, his status should be rising.

  11. Gravatar of Benjamin Cole Benjamin Cole
    9. November 2016 at 18:37

    Pat Sullivan:

    The GOP will be very interested in a more-expansionist monetary policy now, and may well have use for NGDPLT. Hopefully Sumner and the George Mason-Mercatus Center team can get a hearing from the Trumpsters. The Trumpsters will be all ears to any plans for taking off the monetary noose.

    “The market’s being driven (after Trump’s victory) by a pickup in inflation expectations –- from quite low levels -– on speculation that fiscal policy next year will be more expansionary.’’– said Brian Nick, chief investment strategist at TIAA Investment Management.

    So print more money and run bigger deficits! Yahoo!

    Maybe Trump can do a Reagan. Protectionism married to non-entanglements in foreign affairs and big deficits.

    And cheapen the US dollar, ala the Plaza Accords.

    This could be a great eight years!

  12. Gravatar of Matthew Waters Matthew Waters
    9. November 2016 at 18:57

    I also saw this from Nate Silver a day BEFORE the election:

    “Throughout the Rust Belt and the Upper Midwest, Clinton is running 3 or 4 percentage points worse than President Obama did in our final 2012 forecast, which is a big reason why Trump has a better chance of winning than Mitt Romney did. And the scenario by which Clinton loses a state like Michigan or Pennsylvania isn’t hard to fathom: First, enough white voters without college degrees turn out for Trump, and second, African-American turnout is depressed as compared with four years ago (which isn’t crazy to think could happen).”

    http://fivethirtyeight.com/features/election-update-the-state-of-the-states/

    Sean Trente also did look at the 2012 results and found the “missing white voters.” The polls in MI and PA found them to an extent, but either the missing white voters had nonresponse bias or the polls weighted out their existence.

    Also as Silver says, 1% of the electorate changed their vote and Clinton wins FL, PA, WI and MI. Equivalent to 1% changing their vote is 2% coming from third parties or those who left President blank.

    http://fivethirtyeight.com/features/what-a-difference-2-percentage-points-makes/

    Like with Brexit, there was a large bias against even saying the vote was close. I had that bias as well.

    The real question is why the vote is close to begin with. Monetary policy is part of it. Also our country’s health care policy is abysmal. Liberals have made fun of “economic anxiety” among Trump’s hardcore supporters, but very poor economic policy switched more than at least 1% of the population towards Trump.

  13. Gravatar of Capt. J Parker Capt. J Parker
    9. November 2016 at 19:29

    Silver is so full of it. He aggregated a bunch of poll results and applied his little fudge factors and was oh so pleased with himself to find that his favored candidate was the favored to win. Now in abject misery over the loss of the progressive utopia he’s playing self delusional games about what might have been. He ought to be admitting he blew it and figuring out why. Trump flipped a lot of hard hat dems to his side which is different than underestimating GOP base turnout. Not every down ticket GOP candidate was treating Trump like kryptonite so the fact that the senate is still in GOP control could simply mean some down ticket GOP candidates figured out, like Trump, how to steal some of the populist baloney away from the dems.

  14. Gravatar of Steve F Steve F
    9. November 2016 at 21:13

    I remember arguing for months that the polls were either being calculated with great laziness by the pollsters or by their own agenda. This was back when the averages had Clinton beating Trump by some ridiculous digits.

    If it was true that the polls weren’t that far off, it would also be true that Silver wouldn’t have given such silly probability win chances to Clinton and it would also be true that the many, many millions of people who were shocked, astounded, baffled, surprised today that Trump won because they were told he had it in the bag wouldn’t be, well, shocked, astounded, baffled, surprised.

    On election day, the guy sitting behind me in class told me that he doesn’t even know why I was reading election news since it’s already obvious that Clinton was going to win since Nate Silver said so.

    I spent a good deal of effort on other forums telling people Clinton wasn’t nearly as far ahead as they thought. Saying the polls weren’t that far off is revisionist history.

  15. Gravatar of Steve Steve
    9. November 2016 at 21:58

    Here’s a fascinating data point:

    Minnesota 47% Hillary
    Georgia 46% Hillary

    In fact, Minnesota would’ve flipped Trump if McMullin hadn’t been on the ballot and his voters had come home.

  16. Gravatar of E. Harding E. Harding
    9. November 2016 at 22:03

    Heck, even Dearborn, MI was less pro-HRC than it was pro-Obama (thus cutting into Her margins). Detroit was far less pro-HRC than it was for Obama, leading Trump to win MI, and was slightly more pro-Trump than it was pro-Mitt.

    “The exit polls have been cited many times, but it just strikes me as wrong that Clinton did better among whites than Obama. The county-level voting data doesn’t bear that out.”

    -Maybe the shift was especially strong in places like San Mateo, CA and where our man Sumner resides?

    Yes, Trump did seem to do much better among Hispanics than Willard S**t, check the returns from Starr and Zapata counties in Texas.

    BTW, wasn’t it Steve Sailer who was the biggest figure in 2012 to point out the GOP autopsy was BS and that the GOP needed to capture a few Great Lakes states to secure a Lincolnesque so-good-it-should-be-illegal EC advantage?

    BTW, how ironic is it that Trump did better in Missouri than Montana, and in Iowa than Georgia?

    As I pointed out on my blog, the 70%+ White town where I live in trended slightly Dem. The heavily minority precinct I live in trended GOP (percentage-wise, not margins-wise, due to rising turnout on both sides, but there being very few Republicans there in 2012).

  17. Gravatar of E. Harding E. Harding
    9. November 2016 at 22:04

    In any case, any person who denied my claim Trump and Kasich were the only electable candidates the GOP had on offer wanna comment?

  18. Gravatar of Lorenzo from Oz Lorenzo from Oz
    9. November 2016 at 23:12

    The Donald got less votes than Mitt Romney.

    Hillary got a lot less votes than Obama.

    The Donald did better than Romney amongst black and Latino voters because so many of them stayed home.

    The Donald didn’t win: Hillary lost, The Donald was the one left standing.

  19. Gravatar of Brian Donohue Brian Donohue
    10. November 2016 at 00:22

    Scott,

    Saw this map on counties who voted Obama/Trump and immediately thought of you:

    http://mapsontheweb.zoom-maps.com/post/152956392778/counties-formerly-for-obama-that-went-for-trump#_=_

  20. Gravatar of Ray Lopez Ray Lopez
    10. November 2016 at 05:17

    Nice post by Sumner, not controversial since it sticks to facts not unverifiable theories like NGDPLT, but the dunce’s cap belongs to Matthew Waters, who absurdly links Trump’s victory to US monetary policy! I suppose an equally ideological Marxist would link Trump to the exploitation of the working class or other such nonsense.

  21. Gravatar of XVO XVO
    10. November 2016 at 06:53

    The final polls weren’t that far off, but the polls 2-3 weeks ago with Clinton up 12 were unreal and pretty clearly inaccurate.

  22. Gravatar of Bob Murphy Bob Murphy
    10. November 2016 at 07:46

    Scott, it’s actually your fault Trump won. If only you had agreed with me that monetary policy was way too loose, we could’ve avoided this. Now it’s E. Harding’s world.

  23. Gravatar of ssumner ssumner
    10. November 2016 at 08:10

    SteveF, You said:

    “If it was true that the polls weren’t that far off, it would also be true that Silver wouldn’t have given such silly probability win chances to Clinton”

    Just the opposite, he gave Trump more chance than almost any other pundit—about 30% chance. His reputation is soaring.

    Capt, Parker, You and Steve F are really clueless. Don’t you know how to read?

    Harding, You said:

    “In any case, any person who denied my claim Trump and Kasich were the only electable candidates the GOP had on offer wanna comment?”

    Check out Rubio and Trump in Florida, and tell me that Rubio couldn’t have won that key state. He won the Senate seat by almost 8%. Trump by 1.5%

    The urban Dems did not turn out, that’s why Hillary lost. If they didn’t turn out to vote against Trump, they aren’t going to turn out to vote against Rubio. Meanwhile, Rubio would have done better with professional white voters.

    Where you were right is the Hispanics. I thought they be much more strongly anti-Trump, it turns out they didn’t really care about his racism. Not sure why. You were right about that.

    Lorenzo, Again, millions of votes still have not been counted.

    Thanks Brian.

    Bob, The markets think Harding’s about to find out that Trump will dump his alt right supporters, and join the GOP supply-siders. We will see.

    As for money being easy, I’m going to assume you are joking, as you are not that dumb.

  24. Gravatar of E. Harding E. Harding
    10. November 2016 at 08:45

    “Check out Rubio and Trump in Florida, and tell me that Rubio couldn’t have won that key state.”

    -Psst. Look at the electoral map. Trump would still have won had he lost Florida. Willard Sh*t and Rubot would still have lost had they won that state.

    Luzerne and Sandusky counties do not have major reserves of Cubans.

    “The urban Dems did not turn out, that’s why Hillary lost.”

    -Really? In Iowa? By more than in Georgia? Don’t say stupidities.

    “Meanwhile, Rubio would have done better with professional white voters.”

    -I honestly can’t believe you’re lying so brazenly. Willard Sh*t WAS the choice of “professional White voters” in Iowa, Georgia, Ohio, and everywhere else in the GOP primary in 2012! The chances Rubot would have done better than him in that demo (or any other except Cubans) are between 0% and 0%.

    “it turns out they didn’t really care about his racism.”

    -Because it didn’t exist AND because Clinton sent several K too many emails for Her own good.

  25. Gravatar of Bob Murphy Bob Murphy
    10. November 2016 at 09:06

    Scott wrote,

    “As for money being easy, I’m going to assume you are joking, as you are not that dumb.”

    Of course I was joking, meaning that I know you think money has been tight. But I think money has been loose.

    So, it was well-informed trolling, which is an art that I try to refine every day.

  26. Gravatar of H_WASSHOI (Maekawa Miku-nyan lover) H_WASSHOI (Maekawa Miku-nyan lover)
    10. November 2016 at 15:51

    > As with Brexit, however, the experts had trouble processing the idea that Trump could win.

    I can’t believe still people doing blah blah analytics.

    That’s simple, Hillary’s smile

    http://www.dailymail.co.uk/news/article-3854016/Could-Hillary-s-smile-cost-election-Twitter-mocks-Clinton-s-creepy-grandma-grin-smirks-way-presidential-debate.html

  27. Gravatar of Patrick R. Sullivan Patrick R. Sullivan
    10. November 2016 at 16:34

    ‘Seriously, that’s not the sort of decision that presidents typically make. No president decided on a 2% inflation target.’

    But this isn’t just any ordinary president. This is a blank slate. He’s a businessman, explain to him what’s in it for him; a successful presidency and re-election.

    He got through Wharton, he can understand algebra. Maybe an open letter to President Elect Trump explaining Obama’s mistakes–and the Fed’s of course–and how to avoid repeating them. I’d even help you write it.

  28. Gravatar of Patrick R. Sullivan Patrick R. Sullivan
    10. November 2016 at 16:40

    Besides, if you can convince him that monetary policy is where he should focus, he might forget about building that stupid wall.

  29. Gravatar of MFFA MFFA
    10. November 2016 at 23:47

    With Trump planning to blow up the budget deficit and engage in other failed Keynesian stimulus, will we see Paul Krugman coming back to his 90s self? Will Trump make Krugman great again?

  30. Gravatar of Jerry Melsky Jerry Melsky
    11. November 2016 at 02:07

    Steve F said “If it was true that the polls weren’t that far off, it would also be true that Silver wouldn’t have given such silly probability win chances to Clinton”

    Dr. Sumner said: “Just the opposite, he gave Trump more chance than almost any other pundit—about 30% chance. His reputation is soaring.”

    I say: What the heck does 30% probability mean for a one off election? How are we supposed to interpret it? How do people in fact interpret it? Let’s be generous and say Silver is saying that a 30% Trump – 70% Clinton forecast means that for the margin of victory from his aggregated polls and the measurement error of those polls the calculated probability of a Clinton win is 70%. The thing is: for a result like Silver’s (48.5% Clinton and 44.9% Trump – 3.6% Margin) The 95% confidence interval on Clinton’s margin of victory that is consistent with a 70% probability of a Clinton win is +/- 7.2%! – twice the size of Clinton’s predicted margin of victory! Any honest interpretation of that result would have been that the election was way too close to call statistically. But Silver’s 70% chance of a Clinton win presentation was not taken that way by most of his audience in my opinion (myself included). So, it’s damn lies and statistics once again. And how close the actual election results were to the poll results doesn’t make Silvers misrepresentations look any better. And for this Silver’s reputation should soar? I think not.
    One thing you can say about Silver, he has a great sense of humor. You can tell because in his probability forecast numbers he uses decimal places.

  31. Gravatar of ssumner ssumner
    11. November 2016 at 06:19

    Harding, You said:

    “Because it didn’t exist”

    That makes no sense. If Trump’s not a racist, then why do you like him so much? Stop lying.

    Bob, You said:

    “But I think money has been loose.”

    That’s just sad, inflation has been averaging 1% and NGDP growth is 3%. There’s good trolling and bad trolling. Saying dumb things is bad trolling. I already get too much of that over here, I don’t need more of it.

    What if I came over to your blog and said “Money was tight during the German hyperinflation, because they had high interest rates”? Would that be a smart comment?

    Patrick, I wouldn’t want the President making that decision, so I would not have any interest in convincing him, even if I could.

    MFFA, That’s actually a good comment. Thank you.

    Jerry, It’s not Nate’s fault that some people are too dumb to interpret simple probability. His site is wonderful for those of us who understand basic statistics.

  32. Gravatar of E. Harding E. Harding
    11. November 2016 at 12:41

    “If Trump’s not a racist, then why do you like him so much?”

    -I told you, his immigration stance, Supreme Court list, and stated foreign policy. His refusal to denounce most racists supporting his campaign is certainly a plus (it makes it impossible for the media to smash him with guilt), but hardly the largest factor in my support for him.

    Remember, I also supported Kennedy over Duke for Louisiana Senate, since Duke seemed incapable of handling his emotions. Also, like many fellow Michiganders, I supported Barack Obama over Mitt Romney in 2012 (though I didn’t vote).

Leave a Reply