The national polls will probably end up missing by around 2-3 points hardly catastrophic. And as I mentioned earlier, polls predicted Hillary’s state shares quite well. Where they completely failed was predicting Trump’s state shares which of course ended up determining the outcome. So while polls definitely missed the mark, it was somewhat narrower than it first appeared.
As a compulsive 538-watcher, the most stunning development to me was the shattering of the Blue Wall.
Every day, I’d check 538 and see the swing states changing – from solid blue at Hillary’s peak (post-convention, and again after the first debate) to light pink at her worst (pneumonia). But the blue and red core never changed.
So my assumption was that those states were locked down, and all that mattered was the swing states: Trump needed to sweep the swing states; Hillary just needed a split.
So I was stunned to see the upper midwest in play. Michigan and Wisconsin (and to a lesser extent, Pennsylvania) were rock-solid blue all year long. WTF?
My theory is that when people are asked if they are bigots, anti-gay, anti-abortion or anti-women, there is a natural reservation to not claim membership in those groups because polite society considers that to be odious.
But when they are somewhere private where no one’s watching, like a voting booth, their true nature comes out. Who’s going to know I voted for Trump if I tell them I voted for Clinton?
Not everybody thinks that way, but when the margins are so narrow, it only takes a few votes to completely change the outcome.
ALL phone numbers are available to pollsters, or anyone else. Dialers just call all numbers in a block and don’t care if it’s a cellphone or not.
Please, correct me if I’m wrong, but the polls were off by about 2%.
That’s within the margin of error.
Hillary lost by 108,000 votes scattered over three states (Michigan, Pennsylvania, Wisconsin).
The issue is not the polls.
The issue is 1) as has been shown time and time again in many countries, National polling numbers get incorrectly interpreted (what matters is how much your vote is concentrated in each district, be it state or a smaller level) and 2) How do you divvy up the undecideds.
It seems that many pollsters take the undecideds and split them up exactly according to the existing proportions.
I have, however, talked to several polling analysts who think this is incorrect, and you have to divide undecideds based on whatever other social issues are at play — and if you cannot do that, then just state the number of undecideds.
Yeah, and with Caller ID on all cell phones, and most home phones, people don’t pick up anymore. I answered less than ten percent of the phone calls I received over the past three months, as they were almost all telemarketers or political robocalls of some sort. I received several that seemed to be pollsters, but many could have been campaign “surveys”. I didn’t answer most of them.
I spent Sunday making phone calls for Hillary. I placed 200 calls, to Virginia and Arizona. Maybe 15 of them were picked up by a human.
When people are asked if they support a position that hateful people erroneously call bigoted, there is a natural reluctance to admit it.
Do you live in Virginia or Arizona?
As I mentioned, the non-massaged benchmark of the final RCP avg in a 4 way race was 3.2 for Clinton. She appears to be around 0.2% ahead in the popular so that’s 3 not 2. If any major poll massaging aggregators had her up by less, that could be due to massaging the numbers: good for them if they adjusted in the right direction but doesn’t make the polls themselves more accurate.
That said to also repeat, in 2012 the final RCP avg had Obama up only 0.7% national popular but he won by 3.9% so it was further off. But in 2008 RCP avg was only a fraction of a % off, and more like 1 usually before that.
Margin of error in pure statistical sense doesn’t apply directly to an average of a bunch of polls. Anyway MOE only refers to pure statistical error, how close is your sample % of 1000, say, red and blue marbles drawn from a giant jar of 130 mil of them to the real % of red marbles in the jar, at a 95% (by convention) level of confidence. That’s only part of what is going on, ever, in polling errors in the real world. There are marbles which refuse to be counted or tell you they are red when they’re blue, you may unwittingly sample only in a portion of the jar that’s unrepresentative, and so on. None of that is included in statistical MOE for a given poll.
I don’t understand the reliance on 538-type polling. As I understand it, 538 averages the individual poll numbers to make it’s assumptions/predictions. Usually, it can be assumed that an average of poll numbers should negate any erroneous high/low poll numbers and produce accurate results. While that certainly sounds good, it fails to take into account the large number of faulty polls.
If you start with faulty data/input, your end results will be incorrect. The individual polls that predicted a Hillary win were proven wrong. When 538 chose to average incorrect poll numbers, they never had a chance of correctly predicting the winner.
No, you don’t really understand. 538 pretty specifically ranked and sometimes excluded poor pollsters. He adjusted the more reliable pollsters based on their percieved bias and on national trends. He also built a big uncertainty into his estimates to account for polling errors and unknown ways the undecideds/third party voters would break.
You gotta read as well as watch ( and that sounds snarky, but I absolutely don’t intend it that way ). There were reasons to skeptical of such a construct and we just saw a good example of why. Both Nate Silver and Harry Enten at 538 have challenged the Blue Wall hypothesis - Silver as I recall has hedged his bets, Enten flat out doesn’t believe such a thing exists/existed. Enten from last night:
- I had long thought that Democrats didn’t have a blue wall in the Electoral College. I had long thought that demographics weren’t destiny. Yet I didn’t think Trump would be the person to disprove these theories.*
Every election cycle we get crowing/doom-saying about the next “permanent majority”, whether it be Republicans or Democrats. It comes from both sides and IMHO it is usually mostly bullshit. The politics of a country of 300 million+ can be about as accurately plotted as the weather. You might get closeish, but it is too chaotic of a system to ever really nail down and predict the future.
538 either correctly predicted a Trump win/Hillary loss, or they didn’t.
Polling organizations that can’t predict the future aren’t of much use to anyone.
Nobody can reliably predict the future, once you cut through any hype or selective track record they offer you, about election outcomes, financial markets, sports outcomes or any other uncertain future event. What we should be looking for in regard to uncertain future events is a reliable estimate of % likelihood. Looking for somebody to reliably tell you every time what’s going to happen is a fool’s errand.
Now I’d stop short of completely contradicting you because there is no way to closely verify the % likelihoods given by Silver/538. It’s pretty much obvious Wang, for example, had it more wrong in saying Clinton was 99% likely to win (but last night was the one in a 100! would be a thin reed to stand on 
). Silver was fairly obviously closer in saying Clinton’s chance was 70-ish% (too bad for him his model output went up from 65% not long before). But it’s impossible to say how much closer. And we’re just not going to have enough presidential elections under similar enough conditions to test whether candidates given a 70% chance then win close to 70% of the time. Other types of elections and more than a few years’ past presidential ones might not be similar enough.
That’s silly. He was giving odds, not stating an absolute prediction. If you took it as such, that’s on you not him. Frankly, 538 is an odd one for you to pick on as he was the most cautious of the bunch with regard to the confidence in his “prediction”.
Depending on where exactly the national vote ends up (counting won’t be finished for a little bit), they may end up better than in 2012.
Indeed. He got into a Twitter war with the HuffPo’s version of Silver because the HuffPo was giving Trump less than a 2% chance to win, and he claimed Silver was *way *too bullish on Trump. And almost a month before the election, 538 published this rather prescient map.
Things can change if everyone starts trusting pollers again and tells them the truth. News outlets that sponsor polls and then ignore some of the results that they don’t like are a big factor.
Also “likely voters” has to be a very difficult judgement. It doesn’t help to let your biases affect those calcs. I’m sure that the obvious turnout and excitement at Trump rallys was rationalized away by some who were otherwise attempting to put out accurate poll results.
I guess you’ve never even glanced at the 538 website, nor ever read a single post about it. This is not what they do. You might find what they do worthless, but some people find probability-based analyses useful, so let them be.
This is as good a thread as any to mention this: As some of you know, 538 has also, for months, been calculating the probability of an EV/PV split, and of what nature (Dem/Rep, or the reverse).
I would say they got this one right, too, in that they warned us this was more likely to occur this election than many assumed – it was up to 11% at the end, which is pretty darned high for such a historically rare event – AND they noted that a Trump EV/Clinton PV split was MUCH more likely than the reverse.
(True, the latter prediction is hardly rocket science, when one considers all those nearly-guaranteed “wasted” Clinton votes, of both types: sure-to-be-red-this-time-but-getting-purpler states like Georgia, and sure-to-be-blue-even-if-millions-fewer-votes-went-to-Clinton states like California).