[matt4077]

In regards to the predictions, it's important to understand that their role in it was the same as it is with the weather forecast: they're just reporting what people come up with.

Those people, the pollsters and aggregators, were indeed wrong with regards to the winner. It's however important to note that the polls were less than 3% off. It just happened to make quite a difference in the winner-take-all system.

538 was arguably more right than others: their model sensed the uncertainty and gave Trump a 30% chance of winning.

Compared to the NYT/WaPo/WSJ, the self-styled outsiders like The Intercept are incredibly biased, or just bad. It's sometimes hard to see how information flows, but barely any actual news starts at The Incept/Breitbart/HuffPost/etc.

[code_duck]

They could have said it was a 95% chance, or that it was a 10% chance of Trump winning and been just as 'right'. The only thing that could be 'wrong' in predicting a probability like that is if they say there is no chance of something, and it happens.

[crdoconnor]

>It's however important to note that the polls were less than 3% off

Wisconsin last poll 41% for trump, Wisconsin result 49% (~8% off)

Michigan last poll 42% for trump, Michigan result 48% (~6% off)

North Carolina last poll 45% for trump, North carolina result 51% (~6% off)

Florida last poll 45% for trump, Florida result 49% (~4% off)

Source: http://www.electoral-vote.com/evp2016/Pres/Maps/Nov09.html

http://www.electoral-vote.com/evp2016/Pres/Maps/Nov08.html

You can't excuse the pollsters. That's an epic fuck up.

[dragonwriter]

Polls aren't predictions (some people draw predictions from polls, but those are different from polls.)

They, particularly, don't even in theory sample the universe of "people who are actually going to vote", the sample registered voters or some model of likely voters. Those are known (to anyone who has more than a casual understanding of political polling) sources of nonsampling error when treating polls as a measure of the actual vote, since the universe sampled is different than the universe of interest, and there was particularly a lot of publicly-aired certainty about the utility of backward-looking likely voters models in this election.

I don't think there is a particular problem with the polls so much as with people's (including, unfortunately, many of the people talking about polls in the media) understanding of what polls measure.

To the extent there was an epic fuck up, it was in poll-based predictions that the effects of these type of things are correlated across states, so they gave Trump a near-certainty of winning. (Better predictors, like 538, had Trump an underdog but with a sizable likelihood of winning -- you can't say a 70%/30% chance is wrong given that the result is the one given a 30% chance is the one that materializes.)

[hjffyy66]

Don't you have to account for their margin of error though?

I don't see them on your link. But usually when a poll says 45% there's a spread in both directions indicating the level uncertainty in their data.

So for example a pollster could find that 45% of people prefer X over Y with a +/-5 point margin of error. Meaning it could be 50% prefer X over Y or 40% given our model

Frequently I've seen something like 2-3% margin of error either way. So with your numbers Florida for example could only be off 1% given the margin of error. I dunno what the margin of error was though.

Remember polls are best guess statistical models. There is a lot of jitter in human modeling. Not least of which is due to the simple fact that people lie. We lie a lot to fit in versus be honest. and we encourage this feeling in people. There's no realistic way I can see for pollsters to account for that

[crdoconnor]

This bias was consistent across all polls (swing state polls, anyway, didn't look at the others).

I'm not sure if there was actually even a single poll where Trump's support was overestimated.

[cmurf]

Well it's looking like they assumed more people were going to show up, based on their polling, and a good chunk didn't show up. That is, Trump voters showed up approximately as the poll models predicted, while Hillary voters apparently weren't as committed to actually showing up as they were when polled. So the polling fouled up only in terms of capturing a meaningful nuance whether the voter was likely to show up, which is a really hard thing to do.

And the polls didn't ask, and may have a lot of error capturing for, the Q&A along the lines of "if there's another controversy regarding candidate X are you a lot/a little more or less likely to vote?" It's plausible the polls don't capture last minute antipathy.

[matt4077]

You're doing a bit of cherry-picking but I guess it's only fair, since reality conspired to do it similarly. Not sure how far off the polls were in aggregate.

But my point was mostly that the media doesn't have much to do with the polling. And secondly, I think it's important to distinguish between ideological bias and just simply mistakes. There are many pollsters from all over the political spectrum and I'm convinced that they made a best effort to be accurate, and that it has just become difficult to do accurate polling. Or that the polling may have even given a very accurate picture of reality, but that volatility has just increased dramatically.

[saynsedit]

My original point said nothing on the media's involvement with polling, outside of simply reporting it. As for the pollsters, whether it was ideological bias or simply mistake is also irrelevant to my original point.

My claim is that the "real news" committed the journalistic equivalent of manslaughter. Whether or not they intended to mislead the public doesn't negate the fact that the public was misled. Blaming "fake news" doesn't justify their own role in misleading the public. There is no court of journalism, but the consequences regardless will be that people will trust the traditional news sources less.

[crdoconnor]

I just picked the states which actually decided the election. For all I know it was 8% off in Oregon too but that wouldn't have mattered.

It did appear that there was a systemic bias against Trump in the polls, whether intentional or not.

[masonic]

It's even worse than you state -- those are just percentage points off, not full percentage off.

To restate:

Wisconsin last poll 41% for trump, Wisconsin result 49% (~8 percentage points off, or 20% off)

Michigan last poll 42% for trump, Michigan result 48% (~6 percentage points off, or 14% off)

North Carolina last poll 45% for trump, North carolina result 51% (~6 percentage points off, or 13% off)

Florida last poll 45% for trump, Florida result 49% (~4 percentage points off, or 9% off)

[bubbleRefuge]

Don't forget HuffPo's models giving HRC an 90% chance a few days before.

[code_duck]

That could have been completely correct... and the card was drawn for one of those 10%. To test that probability empirically, we'd have to have the election at least a couple dozen times, right?

[MatthaeusHarris]

Simultaneously and independently, with exact copies of all registered voters.

[matt4077]

Yeah, HuffPo is just one of the leeches feeding of actual journalism. I really enjoyed the little fight they got themselves into the day before the election – their failure is/was to consider the errors in state polls to be independent.

[saynsedit]

What is The Intercept's bias?

[underbluewaters]

They are harshly critical of unchallenged power, so of Obama, US foreign policy, and Clinton. People perceive that as a bias against Democrats but I'd expect we're about to see some good reporting against the Republicans now.

[saynsedit]

I don't see being critical of people in power as a bias, unless you're saying that they don't spend enough time being critical of people not in power.