Some of the problem in the latest election was that the MSM treated polls as asking “Are you going to vote for Hillary, or are you stupid and racist?”
Regards,
Shodan
IMHO this is the crux of the biscuit, so to speak. It’s not that polls shouldn’t be trusted, it’s that polls (and the media, oh god the media!) need to do a better job of EXPLAINING what they are stating, so people understand. Unfortunately, people have a strong tendency to want bullet-point answers. We want the weather report to provide answer for us “Do I need a jacket? Do I need to take an umbrella?” All that stuff about low pressure systems is nice and all, but “Do I need to take an umbrella?”
So, in the political polling universe, hearing all the pollsters (and the media, oh god the media!) repeat over an over that “Hillary is up 4%” is interpreted by the listing public as “Hillary is going to win by 4%”. Why wouldn’t the listening public think that? So, I put it back on the pollsters (and the media, oh god the media!) to do better job explaining what it really means.
The problem in the US election was not with the polls, it was with the innumerate analysts incorrectly interpreting the polls.
The national polls were very accurate: Hillary did indeed get more votes than Trump, and the final answer was within the margin of error.
However a national poll is somewhat useless when looking at a presidential election because the presidency is based on the results of 50 separate votes in 50 separate states.
To adequately predict what will happen you need to look at polls in each state, not look at a country-wide poll and extrapolate state-by-state.
As long as your census is accurate, and your polling company can properly weight the demographics, a poll can be a good snapshot of public opinion.
Barbarian, that’s true, but note that the aggregator 538 (for example) was VERY aware of the state-by-state electoral system – it was the basis for their model. Thus, they correctly predicted that Trump had a sickeningly, exasperatingly, what-the-hell-is-wrong-with-Americans high chance of winning.
Recall hearing from the radio the other day the results of a poll. Something like 40% approval vs 48% disapproval of Obamacare but a 52-38% approval/disapproval for the Affordable Care Act.
Well then I guess I don’t understand polls because there were times when 538 had trump’s chance of winning at 3%. I would never look at those odds as “yikes, he might win!”.
It seems pretty cut and dry to me but I will accept that somehow that means he might win.
And if the election had been held at one of those times, he wouldn’t have won (well, 97 times out of 100). Opinions change over time – this election, largely because of fake news crap, lies, and invented controversies–but they change nevertheless.
That’s what we’ve been saying, you are a good representation of how the public understands what is being reported. It’s not you, it’s the communicators.
No, 538 never said that. You may be confusing it with Wang’s PEC. Even 538’s “NowCast” which was not a prediction of what would happen on election day but an as-if the elections was that day only hit that low (3.7% actually) for one day in August. There were also times in NowCast that were 55% Trump wins. Most of the cycle his “Polls-plus” averaged a 1 in 4 chance of a Trump win (and of the Trump win options a fair number included winning the Electoral College while losing the popular vote).
So not sure if you don’t understand polls, or you remember incorrectly, or you read wrong.
On election night, I remember looking at a map of the U.S., where states were marked as safely Democrat, likely Democrat, toss-up, likely Republican, and safely Republican. I presume that polling data was used at least in part to determine which state fell into which category (perhaps used in conjunction with knowledge of how states voted in past elections). I remember watching in amazement as all likely Republican states went with Trump, all toss-up states went to him, and several “likely Democrat” states went to him. And even the likely Democrat states that went to Hilary were a much closer race than what was projected. Unless I’m wrong and polling data was not used as a factor in determining which state was likely to vote which way, then what happened was polling data was either (a) an inaccurate representation of public opinion or (b) misinterpreted for a whole number of states back in November. So I think it’s reasonable to question the usefulness of polling based on these results. Even if the results fell within the margin of error, the fact that so many of the results skewed in the same direction relative to the prediction indicates the shortcomings of polling. (I hope this post makes sense; I’m writing it in a hurry so sorry if it doesn’t!)
Yes, it did. Leading up to the election I lived on that site and (it might have been shortly after the “grab 'em by the pussy” incident) there was a day when it had Hillary at 97%.
At any rate, like I said, I will accept that I don’t understand polls.
Too late for an edit. That’s probably what I saw.
But on the actual day of the election, 538 gave Trump a 28% chance of winning. That’s taking out your six-shooter, putting two bullets in the revolver, putting the gun to your head and pulling the trigger.
Two bullets in the revolver isn’t exactly the same as no chance you’re going to end up with your brains splattered all over, is it?
It is much harder to poll states than it is to poll the country as a whole. So what happened is that they used state polls which had large margins of error, combined with previous results relative to the national polls. So if a state is usually two points more Democratic than the national polls then they added two points to whatever Hillary was polling and vice versa for Trump.
The problem was although the national polls were actually more accurate than last time they states were out of their historic bounds. So some states were much more pro-Hillary than usual but the states on the fence were much more pro-Trump than normal.
But that’s exactly what it meant. It meant that if there were 100 parallel universes that all ran the election on that day, Trump would win in 3 of them. There was no way to tell if we were in one of those 3 universes or if we were in one of the 97. So yes, “yikes, he might win” - even in those days. There was never a point when it was safe - and toward the end of October, it was close to 30 of those universes.
Polls aren’t useless. It is, however, necessary to understand what they say - and we’re not taught to understand them in schools, the media are not very good at explaining them, and people often fail to look at the actual poling data to see what the poll asked and what it didn’t.
Do you follow Propane Jane on Twitter? Might help your spirits. Here is one of her threads, Storyfied:
https://storify.com/docrocktex26/white-poverty-the-black-middle-class-the-southern-
No, 538 never had Trump’s chance of winning at 3%. Their forecast record is still online, you can see for yourself. The lowest they ever gave Trump was 10.8% on Aug. 14, and then 11.9% on Oct. 17.
Perhaps you are remembering their “now-cast” number, which represented the likely outcome if the election happened on that day. That may have gone as low as 3% for Trump at some point. Their “forecast” allowed for how likely the public opinion was likely to shift before the election.
Sure, there was systematic error in the polling. And some people (e.g. Nate Silver) understood that. That is the nature and limitation for any polls. It always has been, and always will be.
Well, even if that had been the prediction, 3% isn’t exactly negligible. A real world example of something with similar odds is to roll a pair of threes (a hard six) on two normal cubic dice. Not a common occurrence but not a bizarre one, either.
(A ‘hard six’ is a craps term for rolling specifically a pair of threes on two dice, as opposed to a ‘soft six’ which is any combination of the two dice that totals six. Why yes, I have been watching Battlestar Galactica.)
Thank you for that explanation, puddleglum, it does make things clearer.