Hacker News Reader

Comment by cocktailpeanuts | Hacker News Reader

How is it not what I describe? You just described exactly what I described.

Everyone was wondering (both pro-trump and anti-trump) where the "90%" came from. Because it was pretty clear the race was not that obvious. People keep saying nobody saw it coming, but that's their emotion speaking. People just didn't want Donald Trump to win because they thought he's a joke, but objectively speaking people knew clearly that there were tons of his supporters around the U.S., and it was definitely NOT 90% chance that Hillary would win.

To me, THAT's the ultimate "fake news". Slapping a percentage number on it and putting it on the front page is just not right, especially when it's based on very subjective (and eventually false) speculative data.

View on HN

You said:

"The front page had this interactive realtime story of how Hillary is supposedly winning"

I said:

"it absolutely wasn't showing false data that indicated Hillary was winning when she was not"

They're pretty clearly contradictory. That data, at no point, showed Hillary "supposedly" anything. You're making my point about readers not understanding statistical probabilities. 90% doesn't mean she's winning. It means she's got a 90% chance of winning, and there is a 10% chance she will lose. She lost.

> People keep saying nobody saw it coming, but that's their emotion speaking.

Quite the opposite. The data (that same data you're saying came from nowhere) showed that Hillary had a 90% chance of winning. There was variance in the polls, but pretty much all of them showed Hillary with an edge. Yes, there were a great many reports of people saying that there was a lot of Trump support across the country, but those reports are the very opposite of "objective" - they were thinly spread and anecdotal. Pretty much every newspaper ran stories interviewing those Trump supporters, but there was no hard data to quantify it.

Polling data isn't "subjective" and "speculative". There is a clearly defined methodology behind it. That methodology was incorrect in the 2016 election, and didn't accurate gauge support for two unprecedentedly unpopular candidates. That makes it "incorrect news", not "fake news".

You could argue that the news industry has relied too heavily on this sort of statistical prediction stuff, and I'd agree with you. But in 2012 FiveThirtyEight correctly predicted all but one state, and everyone absolutely lauded Nate Silver for it ("Is Nate Silver a witch?"). But really, what that meant is that each of Nate's predictions were the right side of 50%. And in 2016 they most definitely weren't.

Polling data wasn't even wrong. It was almost spot on in measuring popular vote totals. The mistakes were solely with some state level polls (and really not even mistakes, just high uncertainties) and poll aggregators.

Really, I think most of the incorrect predictions came from one error: assuming a higher level of independence between state results than is actually the case. Nate Silver/538 talk about this some, and they gave Trump the highest (~30%) chance of winning on election day.

Polling can be incredibly hard. If you ask the wrong type of questions you introduce a bias. If you ask the wrong types of people you have sampling error.

All of these things could have occurred, but something definitely was wrong with the polling data. Nate Silver just like NYT gave Hillary a 90% chance at the start of the day, and this was all based on using the polling data as inputs to their models.

>Nate Silver just like NYT gave Hillary a 90% chance at the start of the day

He didn't though, and that's my point. What you're saying is factually untrue, they haven't updated their prediction since election morning, and it sits at 71% Clinton [0] (Updated Nov. 8, 2016). If you don't believe that, then you can take a look at archive.is, which provides timestampped copies. This [4] is an archived copy from well before polls closed (3:30 PM), there are others from earlier in the day.

I don't know what led you to believe that it was a sure clinton win, but that just wasn't the case.

Clinton won the popular vote by somewhere between 2 and 3%. National polls from just before the election (also per 538, [1]):

    +4, +2, +3, +4, +4, +4, +4, +4, +2, +3

An arithmetic average of those gives C + 3.4.

Take Florida, a state trump won, here's polls taken in Novemeber [2]:

    +1, -3, +0, +2, +1, -4

An arithmetic mean gives -.5, Trump won by about a percent.

Take Pennsylvania [3], another close state. November polls again

    +1, +2, +4, +6, +3, -2, -2, +0, +0, +2, +2

An arithmetic mean gives Clinton + 1.18, Trump won by about .7, so this was perhaps the greatest polling blunder, and its still a margin of under 2%, which isn't unheard of.

[0]: https://projects.fivethirtyeight.com/2016-election-forecast/ [1]: https://projects.fivethirtyeight.com/2016-election-forecast/... [2]: https://projects.fivethirtyeight.com/2016-election-forecast/... [3]: https://www.nytimes.com/elections/results/pennsylvania [4]: https://archive.fo/4EQGB

I stand corrected, it was 71.4% The 90% from NYT stood out in my mind.

I see what you're saying now.

But I don't think you got my point either. The detail doesn't matter. Whatever that number was about, it was pretty clear that 90% was not the right number, I'm not even pro-trump and I was confused, imagine what others thought.

Also, it is not ok to be irresponsible just because you're using a "scientific method". This is what cigarette companies did to fund research that "prove" that smoking is healthy. And this happens increasingly more, because writing polarizing articles generate more shares and page views and money.

Lastly, I am not trying to talk politics here, and it's not even just about NYT. Nowadays for a lot of writers it's all about generating more page views. Again, I'm fine with it (even the "fake news" viral sites), but I was pointing out that it is hypocritical to claim that you're "the truth", when you do things that are biased.

> Whatever that number was about, it was pretty clear that 90% was not the right number

Was it? Maybe 90% didn't match up with your intuition, but NYT wasn't making that number up out of whole cloth. That's what the polls were saying, as analyzed by the NYT polling staff.

Maybe they were wrong (maybe they weren't-- things that have a 10% chance of happening happen all the time), but you certainly haven't demonstrated, or even really attempted to demonstrate, bias. All you've really said is:

* Trump won

* Your intuition told you he had a greater than 10% chance of winning before election day

> throughout the entire election cycle the narrative was NYT vs. Trump. News organizations and journalists should report stories but should not become the story.

It's important to remember that this narrative largely came (and continues to come) from Trump, and he benefited greatly from it.

You seem very confident that "90% was not the right number", and I'm curious how you reached that conclusion. Do you have research that indicates the odds were actually way different, or does it just that it feel incorrect?

If I win the lottery, is it misleading for people to state that my odds of winning were one in a million?

Dude, that guy who was supposed to lose by 90% chance actually DID win. And you're telling me to prove this by showing you some research result?

It's not misleading to say your odds of winning the lottery is one in a million because it IS one in a million, mathematically. NYT saying Hillary would win by 90% chance is NOT mathematics. Do you really think your analogy was adequate?

My point was that people believe statistics, just like you believe that because NYT did it, it would be scientific and you should believe it. But that's where the danger lies. And an influential publication such as NYT should be ashamed of themselves for misleading people by trivializing things down to just a percentage and feeding it to the world, in realtime.

I'm sure a lot of Trump fans who saw that poll on NYT that morning who didn't care for voting decided to go vote.

> Dude, that guy who was supposed to lose by 90% chance actually DID win.

Again, that's how probabilities work. Unless the prediction was 100% Hillary win then a Trump win is possible. This isn't a new concept, it happens in sports all the time. Odds are offered for betting, and every now and then there are upsets. That doesn't mean the odds are nonsense.

What's the alternative? News organisations never even attempt to gauge whose campaign is more popular? I don't think that's realistic.

> Dude, that guy who was supposed to lose by 90% chance actually DID win. And you're telling me to prove this by showing you some research result?

Actually yes, that is exactly what I'd like to find out more about. I'm asking why you are so sure that the NYT and other media publication odds were so wrong? Is it just that he won, therefore his odds must have been better than 10%? Most betting markets (which the HN set seems to put a lot of faith in) had Clinton as a 80ish % favorite (http://www.cnbc.com/2016/11/07/betting-sites-see-record-wage...), do you think they were a product of a misled public?

> Everyone was wondering (both pro-trump and anti-trump) where the "90%" came from

No, they weren't, because everyone paying even a little bit of attention to the issues around aggregating polling numbers new that 538 had already dissected the problem with that long before the election—most prediction models were falsely assuming that variations in outcome from.the poll averages between different states were independent events when history shows that such deviations are strongly correlated. Which is why 538, working with essentially the same polling data as everyone else, had Clinton's chance of losing much higher than NYT and some other predictors.

> To me, THAT's the ultimate "fake news".

Predictions of the future are (and were presented as) analysis, not news. To be "fake news", it has to firsr pretend to be news, that is, reports of things that have occurred.

> Because it was pretty clear the race was not that obvious.

I think that's where the confusion lies. The majority of published polls and pundit calculations pointed that way, and it was earnestly interpreted by well-intentioned journalists that 90% of polls meant 90% probability. Perhaps I'm cutting a little too deep with Hanlon's razor here, but I think systemic biases are a more likely culprit than some kind of media conspiracy.

I don't recall the New York Times reporting a 90% chance of victory on election day. I recall her chances according to the Times decreasing several percentage points (down to the high 70s IIRC) during the last few days leading up to the election. If the interactive graphic you are talking about was the 'paths to victory' graphic that I'm thinking of, it probably reflected the most current polling data - which was probably already obsolete - that they had on hand at the moment.

My own observation during the days leading up to the election was that her chances of winning were decreasing at a faster rate than could be reported with reliable polling data, though I still believed that she'd eke out a win.

http://web.archive.org/web/20161108070757/http://nytimes.com... "Hillary Clinton has an 84% chance of winning"

Reply on news.ycombinator.com