How do you possibly get Bayesian forecast models this wrong, Veeky Forums?

-Don't include electoral weight
-Based on very unreliable information
-Polling online those that are even using websites that offer polls

gee I wonder

rigged by kikes

These two are on the right track. If you're a political strategist one of the things you look at closely are the polls to gauge region and demographic data. That way you can best figure out where to invest your candidate's time and resources.

The flaw with this is of course you then must expect some level of "fair play" by your opponent (with some margin of error). It's highly likely Trump supporters be it the official party or a group connected to them figured out the Clintons depended on such data (by way of big media) to a high degree and just merely told voters to either say they are undecided, won't vote or will vote for Hillary instead.

Then comes the election and all of a sudden Trump is winning a bunch of close call states (several which were supposed to go to Clinton). So truth is the model wasn't the problem, the raw data received was the problem.

But of course anything involving statistics will suffer from these kind of issues if you let it. Which is why you shouldn't put too much faith in it.

Oversampling democrats had nothing to do with it, polls sampled similar amounts of democrats in previous elections (especially 2004 and 2008) and the polls still predicted the election

The problem was, like mentioned, garbage in -> garbage out

I still find it strange, before the turn of the century, that exit polls used to be better than 99% accurate. They were, in fact, used as the indicator as to the level of corruption in the election system. (And, in other nations, still are.)

So when/how did we get so bad at making polls? What are we doing wrong, and how do we fix it?

"So when/how did we get so bad at making polls? What are we doing wrong, and how do we fix it?"

One thing that you have to realize is that polls are wildly inaccurate in contemporary times because most polling agencies (literally almost all, I worked for a major one a few years back) only do landline phone calls. Almost none of them index or do random calling based on cell numbers. They are ancient dinosaurs in terms of their methodology.

Online polls are also basically worthless because they are not location specific at all, and are easy to game.

Statistics, motherfucker, do you speak it?

20% chance of x means that out of a million draws, 200k will hit. It means its less likely to happen than the other option, not that it's impossible.

anybody with a brain knew that trump was going to be president at the RNC and DNC. exceptionally smart people knew almost a year ago.

this was a legitimate surprise to most people, because they're clueless, and the old methods that they relied on are worse than useless.

'exceptionally smart'

that's when I knew you were trolling

This was obvious though. You didn't need too much brainpower to figure out the corrupt murderer was gonna lose over someone who wants to make America great again.

Good guys win, bad guys lose. It was a good day.

>So when/how did we get so bad at making polls? What are we doing wrong, and how do we fix it?

It's when the far left started saying

"
If you vote for Trump, you're racist
If you vote for Trump, you're xenophobic
If you vote for Trump, you're sexist
If you vote for Trump, you're Islamophobic
If you vote for Trump, you're uneducated
If you vote for Trump, you're homophobic
If you vote for Trump, you're a white supremacist
If you vote for Trump, you're a bigot

There is no middle ground. The is no acceptable excuse. Trump supporters should go fucking die.
"

Only SJWtards believe attacking someone with superficial bullshit labels does anything. The last retard who did that got jailed due to libel and trying to defame someone with lacking evidence.

Bayesian stats isn't magic. If the data isn't good and the priors are uninformative, it isn't going to tell the future for us out of thin air.

It's not about "stop voting", rather it changes voter behaviour. Same thing happened in the last Canadian election: one opposition party was ahead by a wide margin, polls started to show it was shifting to another opposition party, so people voted strategically for the party that stood the best chance of knocking out the incumbent party.

Polls showing Trump actually stood a chance in the last two weeks garnered Republican support, money, and morale. People who were considering voting for third parties instead supported Trump.

Polls have a huge influence on voters.

it was very close dumbass, no you can't fucking predict that MI and PA and WI and FL will go red by a % point.

Really goes to show how disastrous it has been with the republicans allowing the dems to establish voting plantations

/pol/esmokers live in their parent's basements and when the pollsters call it's mom or dad who answers the phone.

>he didn't factor in the brexit effect

Any non brainlet should be able to realize this. Maybe more people when asked would say they prefer Clinton, but how many people fervently support Clinton, to the point where they'll go far out of their way to vote for her? Compare this to Trump supporters, of which many would do anything to vote for him. So maybe 55% of people support Clinton, but maybe only 50% of them actually end up waiting in line for hours to vote for her. Compare this to the much more motivated Trump supporters, of which maybe 65% of his 45% of the population would go out of their way to vote for him, thus resulting in more Trump votes than Clinton votes.

Maybe the polls were right, we just happen to live in the lucky 1/5 of timelines where he won?

>template shitpost
LMAO

Beyond the fact that the media was bent on showing highly favorable numbers for Clinton (as evidence by some polls that got it right but weren't publicized/outweighed by the other BS ones), pollsters really just fucked up on this one by making assumptions that worked in the past.

The biggest upsets came from the rust belt states which haven't swung red in a long time. Polling is fairly laborious so polling services strategize their focus outreach to states they deem "important." Because of this, states that weren't expected to swing were scarcely polled compared to the typical ones like Ohio, Florida, etc. Combine this with and the Brexit sentiment and you've got a winning recipe for a batch of interns getting fired on Wednesday morning.

>but how many people fervently support Clinton, to the point where they'll go far out of their way to vote for her?

But everyone ALREADY knew this was going to be a problem once Bernie Sanders dropped out of the race.

The issue was trying to gauge how much support Trump had against Clinton post-convention. The polls indicated "not too much" but of course that assumes people were being truthful when in actuality they were not.

Thus why the model to measure the data was fine but not the data itself.

>using words you don't understand
Never change, /pol/

- Polls favor urban voters
- Trump fans are more passionate, and thus more likely to vote
- There is a chilling effect that even in complete anonymity, Trump supporters will pretend they're not voting for him
And most importantly:
- Pollsters are usually not the brightest people

>SJW-tier comeback
yawn

That's exactly why the polls failed this time. They assumed that the turnout of Democrats and Republicans for an exciting Democrat candidate and a boring Republican candidate in 2012 would be the same as the turnout for a boring Democrat candidate and an exciting Republican candidate in 2016.

They were either stupid or willfully wrong.

>literally forced into replying to every (You) like a chatbot no matter how incoherent
Dat autism :^)

By intimidating voters, very simple.

The left went nuts, we did not trump 2017

enjoying your new president cucky?
:^)

> How do you possibly get Bayesian forecast models this wrong,
The current results show Trump winning WI by 1.0%, MI by 0.3%, PA by 1.2%, and FL by 1.3%.

If you flipped 1% of the electorate from R to D, the result (number of electoral votes) would be almost exactly reversed (307-231 rather than 232-306).

So any model is essentially trying to predict the result of flipping a coin 4 times (which, if the coin could have a bias, isn't the same thing as flipping 4 coins).

You're never going to get polls which are accurate to within 1%.

Overall, the systemic bias was D+2% (national poll average D+3.5%, final result D+1.5%). It turned out to be larger in those four states (polls D+4%, result R+1%).

> That's exactly why the polls failed this time. They assumed
Polls try to avoid making assumptions. You collect numbers, extrapolate those numbers from sample to population, publish results (and methodology).

If the people reading the polls want to re-interpret those numbers according to certain assumptions, that's up to them.

Subjective weighting factors (i.e. likely voter) are also based upon numbers. That involves using historical trends, which change. But there's no credible alternative; you can find a million plausible justifications (some of which are likely to turn out to be true) for "adjusting" the numbers one way or another, but then you've left the polling business and taken up punditry.

Polls need to banned to be honest.

Talking about candidates too. Report on what their policies and platform is as they said it is and not how you feel it is. That's it.

>The last retard who did that got jailed due to libel and trying to defame someone with lacking evidence.
This needs to happen more often.

I don't think we could ever create a proper mathematical model for /pol/. There's a thesis in there for somebody in dynamical systems.

I do believe there was probably a higher bias than calculated and an extreme shy tory effect. This is due in part because of the logical deduction
Trump Racism
was pushed incredibly hard by every single news outlet besides fox.