Page 2 of 2

Re: Drunken Politics

Posted: Thu Nov 10, 2016 6:51 pm
by Urbain
The polls were close but being off by 2 points nationally can make a real difference in the outcome. Polling is a difficult business: people aren't always truthful when asked about who they are voting for; a pollster will only get answers from those who pick up the phone and agree to answer their questions (truthful or not); often people who state their intentions don't get out and vote. How do you, as a pollster, account for this kind of uncertainty at a national level? It's tough for sure.

I closely followed the FiveThirtyEight this year. Nate Silver (538) had the Clinton/Trump ratio at about 70/30 for the last few weeks and were warning people not to discount a Trump victory. At the same time I saw a lot of op-ed pieces calling the FiveThirtyEight's methodology into question as Nate gave The Donald such a strong chance because he used a fat-tail method of analysis. It turns out he was right to do so. While Clinton won the popular vote, Trump won the right states and, as a result, the electoral college. As Nate asked a few times on his blog - would you get on a plane with a 30% chance of crashing? Not bloody likely.

Re: Drunken Politics

Posted: Sat Nov 12, 2016 1:50 am
by Voom
I followed the 538 website too and I agree with all the points you made. Even so, the polls were consistently favorable to Democrats. Polls about Florida were quite accurate, but Wisconsin showed Trump down 4 to 5 points. The result ended up placing Trump 4 points ahead. That's an 8 to 9 point discrepancy. And that's just one example. Pennsylvania behaved similarly and more. Personally, I thought Trump was a bad candidate, but it turns out he was right about the polls. It really was shocking.

Re: Drunken Politics

Posted: Sat Nov 12, 2016 1:30 pm
by rijackson741
Polls are just samples, and usually small ones at that. Pollsters try to make the samples as representative of those that will actually vote as they can, but that requires certain assumptions. Those assumptions are mostly based on behavior in past elections, and if the election is unusual (as this one certainly was!), and the assumptions are wrong, then so is the poll. Hillary failed to inspire young people, who lean heavily Democratic, and their turnout was unusually low. That cost her, especially in Wisconsin. The "left behind" in the rust belt were instrumental in putting her husband into power, because he recognized their needs (whether he actually did anything for them is another matter). They continued to vote mainly democrat after that. Until now. Hillary largely ignored them, at the same time that Trump promised them the world (he cannot deliver it, but that doesn't matter). So they not only flipped to Republican, but they turned out to vote in unexpectedly large numbers. Unexpectedly low turnout for Democrats, unexpectedly high turnout for Republicans, and the polls are wrong. There's not much the pollsters can do about this, because there are too many variables in an election and they have to make assumptions. The one thing they can do is allow for the unexpected by using a fat tailed distribution. So yes, IMO Nate Silver was right to do so.

Re: Drunken Politics

Posted: Sun Nov 13, 2016 4:35 am
by Voom
That's great analysis. It helped me understand what happened. Given that online polls are inaccurate, pollsters should stop calling only landlines too.
As insane as this election was for the U.S. we do have to recognize that Trump will be our next president for better or worse. And whether you agree or disagree with his seemingly imaginary plan you probably want to check out what he is going to try to do in his first 100 days here.