I don’t think the results are outside the margin of error are they? I think part of the problem is that we see 48% in a poll and we think it means 48% today. It actually means 44-52% last week.
That’s within the margin of error of the polls, isn’t it?
Do any of the polls try to predict an Electoral College win as opposed to just a greater number of voters?
Some of them did, mostly using state-level polls and putting together a meta-analysis like fivethirtyeight.com did.
Trump wasn’t getting great odds through that analysis either. I think the consensus was still something like a 10-20% chance of Trump winning.
In a way, that goes back to how average people understand statistics. We hear 10% and mentally revise it to 0% because it’s so unlikely, but if every election had one candidate at a 10% chance of winning, you’d expect the underdog to win about once every forty years. I’ll let someone crank out those numbers, but I don’t think that’s far off from reality.
And Gary Johnson got 0? Jill Stein? The Sweet Meteor O’Death?
They may not be reporting those numbers in the headlines, but by the numbers I saw overnight Johnson was pulling anywhere from 3 to 6% in each state, pretty much nationwide. Easy to imagine then that neither Clinton nor Trump would reach 50% +1 in the popular vote in a closely contested race.
Go do a “XX vote by County” (where XX is a state abbreviation (Quirk of Google: “IN” gets you IL; that one you need to spell out).
It is amazing: the big cities are blue; the sticks are red.
I wonder what the folks there think of their cousins in the other camp: “Them Big City creeps are Commies”. “The hillbillys are all Rednecks”.
Meanwhile, they resent being called “Flyover” by the Commies in the crowded, over-priced coastal States.
“Sometimes, the problem lies not in others; it lies in you”
Each poll will have a margin of error in that neighbourhood, but aggregating them all together reduces the MOE; it’s in effect something like a single very large poll.
Wikipedia says Johnson got just over four million votes, or 3.2% while Jill Stein got just over 1.2 million votes, or 0.9%. I don’t know how the numbers work out by state, but if those third-party candidates had not been there, could Clinton have won?
Actually, Wikipedia does show the state by state totals for each of the four major candidates. I think, for example, that in Pennsylvania if some of the votes for Gary Johnson and Jill Stein had voted for Clinton, she would have won those twenty electoral votes.
A lot of us liberals were listening to Nate Silver, who has been basically saying all long “Clinton will probably win by a comfortable margin, but there’s a chance that there’s a systematic error in the polls, in which case she might win in a landslide, or Trump might get a narrow win”. And it turns out that, sure enough, the dice came up snake eyes this time, there was in fact a systematic error in the polls, and Trump got his narrow win. In other words, the source we were listening to was right. It’s just that, when you’re talking probabilities, “right” doesn’t always mean “what’s most probable”.
Now, what was the cause of that systematic error in the polls? That’s very difficult to say. It’s always very difficult to say with systematic errors, because if you knew about them in advance, you’d correct for them and they wouldn’t be systematic errors any more. It might have come from shy Trump voters, who didn’t want to admit even to an anonymous pollster that they were voting for him, but did so anyway in the privacy of the booth. It might have come from mistaken assumptions in the likely voter models the pollsters were using. It might have come from sampling biases of some sort. We don’t know yet.
If you read the fine print on those polls, they typically poll like 1,500 people give or take.
Yeah, science and math and statistics and shit, but I have a hard time believing you can a sample that small and be SURE you got all your little special sub groups adequately sampled.
Humans are not a random collection of rocks or failing light bulbs or the statistical like.
And sure enough, the 20/20 hindsight error analysis now appears that these polls DID mis/undersample some groups that were more pro Trump than not.
And from almost day one, I had my suspicions about these polls because all the models are based on previous elections. And this election was FAR from a normal one. Models used to extrapolate from the data use to build them often aren’t so hot.
I don’t mind being wrong here, but I don’t think all of the blame lays at the feet of the pollsters. I still get the feeling the media I was consuming was overlwelmingly supportive of Hillary and using the polls as just one piece of data. They didn’t emphasize the limits of the polls or the margin of error that much. It really did sound to me that it was going to be pretty much Hillary and the question was by how much? Trump was openingly derided when he said his comments about how he was going to win.
I think it really is the mindset not of white vs black, or men vs women…it
Really seems rural vs. cities…and the media is populated overwhelmingly by big city culture. They really should set up major media centers in Joplin missouri or Wausau wisconsin, to get more in touch with the rest of the country.
There is, more or less–an Android app called “Google Opinion Rewards”. Fill out a short 1-4 question survey and get about 30 cents of Google store credit. It doesn’t exactly exceed my normal time value of money, but it’s kinda fun and the credits do build up over time.
I’ve never gotten a political survey from it but that may just be luck. Of course, you’d have to be careful using it for politics–maybe iPhone users lean Democrat, for instance. But polls already make those sorts of corrections so maybe it wouldn’t be too bad.
Neither the Brexit polls nor the national polls on this election were all that wrong. Once all the votes are counted, the 2016 national polls may end up being more accurate than the 2012 national polls.
I’m amazed at how often the landline will ring but hang up at the answering machine. Not sure if they’re polls, “Rachel from card services” or another scam, or a creditor looking for someone who’s not me but I usually treat the phone like spam email; by replying/answering they know they’ve got a ‘live’ one.
The last time I was feeling magnanimous a few years ago & was willing to take a poll, I ended up hanging up partway thru it because I couldn’t give my answers. I forget the specifics, but it boiled down to I’m not a single issue voter; if a candidate holds the opposite position from me on one of guns/abortion/gay rights/immigration I might still vote for them anyway if the other three are the same as what I believe in Their multiple-guess answers were very leading & strongly opinionated & didn’t reflect my meh position, yet I couldn’t answer anything but A, B, or C. Yes, I realize it’s easier to run those answers thru a Scantron to come up with your results rather than grading essays but don’t make the choices such polar opposites that I felt like I was being led by the nose to their desired answers, especially when I didn’t really know who was conducting the poll.
Jill Stein pulled enough votes from Clinton in Florida to let Trump win that state. If it had gone to Clinton, she’d be President elect…
What about the Johnson votes… I suspect removing him from the picture would have made trumps win even bigger
Aggregating polls together reduces the error due to the fact that a random sample isn’t a perfect representation of a population. That’s the “margin of error” that the polls generally acknowledge.
But it doesn’t do anything to reduce systemic error.
If your “likely voter” model is off, or if people more likely to vote for a given candidate are less likely to answer your poll for whatever reason, or for a thousand other reasons, then your result is going to be off, and it doesn’t matter if you poll 100x as many people… your result is still going to be off.
It’s not the overall that’s bad. It’s more at the individual level, with states being off as much as 5 or 6%.
I think the problem is that the older methods of phone polling are dying (since you can’t call cell phones) and the newer methods haven’t matured and worked out all their problems.
The odds of NOT rolling a 1 on a six sided die is 83%.
Would you bet $500 that you won’t roll a 1 on a six sided die?