• Quick note - the problem with Youtube videos not embedding on the forum appears to have been fixed, thanks to ZiprHead. If you do still see problems let me know.

How did Nate Silver get it so wrong?

Brainster said:
Anybody who's played a fair amount of poker knows that a 75% chance of winning is nothing to take to the bank.

Essentially, Hillary lost all the toss-ups.
Click to expand...

I bet racehorses. I can't tell you how many times I've been set up for a big pay day only to have a 25-1, 50-1 or whatever come and destroy me in the end.
 
Nate Silver didn't get it wrong. He is, once again, a hero of election forecasting. Nate's final polls-plus forecast was 72% probability Clinton, 28% Trump. When you say things have a 28% chance of happening, then they should actually happen 28% of the time. Well, this was a 28%er. Nate was the only election forecaster to correctly account for the true uncertainty in the polls and the historical fact that the errors in polls are correlated across states.
 
I think I can guess what is being discussed in every Statistics/Probability 1A class in America today.....
 
Brainster said:
Anybody who's played a fair amount of poker knows that a 75% chance of winning is nothing to take to the bank.

Essentially, Hillary lost all the toss-ups.
Click to expand...

A bit like your opponent needing one more heart after the flop for their flush.* Those last two cards are tense, not relaxed.





*My gut was 30%, the actual odds are closer to 35%.
 
As they say, going with the odds for a small return is "investing"; going against the odds for a large return is called "gambling." In the long run, you would not want to consistently bet against the polls.

Tobacco companies would always trot out lil' old ladies who were in their nineties despite smoking two packs a day and having a voice like Vin Diesel. Meanwhile, there are otherwise healthy people who drop dead in their mid-40s while exercising.
 
we have shy conservatives, afraid to tell neighbours and pollsters of their political leanings.
bashful brexiteers, ashamed to tell pollsters they don't like sharing buses with coloured types, now confused to find out syria isn't in europe.

last night the bbc coined timid trumpeteers which i think is apt.
 
Pollsters are not wanted for being accurate but for the value they bring to advertisers journalism. They provide nice simple and above all easy click bait. Beyond that the "instant" polls have very little utility.
 
imodium said:
we have shy conservatives, afraid to tell neighbours and pollsters of their political leanings.
bashful brexiteers, ashamed to tell pollsters they don't like sharing buses with coloured types, now confused to find out syria isn't in europe.

last night the bbc coined timid trumpeteers which i think is apt.
Click to expand...
I suspect it is more a case of shy liberals unwilling to state they could not stomach their party's candidate. Of course, we could get into the No True Liberals discussion.
 
There is a "conspiracy" I have heard that there was a large segment of alt-right folks who intentionally kept quiet ("don't let your libtard friends know you are a white nationalist") to skew polls.
 
Selection bias.

IIRC, pollsters use land lines to sit-at-homes. Single digits of bias towards welfare recipients, unemployed, retirees living on SS, stay at home moms would account for it. And minus the fact that workers just don't take the time to respond to phone calls.

Then allow for the "I'll vote for who looks like the winner" factor, and a better poll would have caused a Trump popular vote win.

Plus the pollsters adjust the raw results to suit "models".

Perhaps in the future, less than 10% advantage will be considered "too close to call"?
 
This election was a low turnout one. Lots of people stayed home, and that skewed things. I'd guess that a good number didn't want to vote for Clinton even though they preferred her to Trump, saw that she was the heavy favorite, and stayed home. 'Whew, she will beat Trump, but my conscience can be clear because I didn't vote for her' style thinking.

Clinton won the popular vote, and it really wouldn't have taken much to have tipped enough of the states she lost to her.
 
Darat said:
Pollsters are not wanted for being accurate but for the value they bring to advertisers journalism. They provide nice simple and above all easy click bait. Beyond that the "instant" polls have very little utility.
Click to expand...

I think that's a bit sweeping. There are a number of pollsters who do indeed try to get click-bait. I think that PPP have been criticized numerous times by Nate Silver for putting in silly poll questions such as "Do you prefer lice or Donald Trump?"

fivethirtyeight had a regular feature called "Good Use of Polling/Bad Use of Polling" in which they looked at particular polls and asked if the polls that people have been noticing were good uses of polling. A lot of the time they flagged up these "clickbait" polls and said that they should just be ignored.

Similarly, there are pollsters who deliberately try to get the result they want because they are partisan. I think Rasmussen is considered one of those polls by fivethirtyeight. Some of these use tactics such as push-polls in which the questions are framed in a way that invites the respondents to agree with a conclusion that the pollsters want to get.

Then of course, the online polls or "clickers" as fivethirtyeight started calling them, which have no standards of who can vote and how often and whose results can easily be manipulated by particular readerships, are essentially worthless.

But not all polls are like that. Some pollsters have their reputations on the line. It doesn't matter for online clickers who ask questions such as "Do you believe you did NOT evolve from a monkey?" because there can be no event such as an election against which their accuracy can be tested, but for pollsters who want to be taken seriously, then simply getting advertising revenue is not a priority. Or rather, if they want to get advertising revenue they may find that being as correct as possible is important.

This throws up its own problems, as Silver has pointed out, which is that pollsters tend to "herd" close to an election. If their own polls start to look like outliers then they fear being considered wrong after the fact and thereby adjust their figures. In this election's case, Silver has recently said in a podcast that pollsters were deliberately skewing their data until they arrived at a Clinton +4 figure (I think some of this is conjecture on his part, but he shows that polls are usually much further apart at other times in the race).

fivethirtyeight recognize the fact that there are going to be various problems with the polls and thus their model is designed to correct for those errors. This is why they assign different weights to their polls, and also why they include other indexes in the polls-plus model. They also set out the parameters of the model ahead of time and largely refuse to add anything into it. This stops them from acting like pundits. For example, they decided to leave out "Early voting" as an important factor despite the Clinton's campaigns claims that early voting would sway the election more in her favour. The only major addition that I think they added was Evan McMullin's run in Utah which had a fair chance of snatching some of the electoral college votes.

Anyway, I think there is actually a lot of cynicism around polling, but fivethirtyeight's takeaway is that it gives good data if it is examined carefully.
 
imodium said:
we have shy conservatives, afraid to tell neighbours and pollsters of their political leanings.
bashful brexiteers, ashamed to tell pollsters they don't like sharing buses with coloured types, now confused to find out syria isn't in europe.

last night the bbc coined timid trumpeteers which i think is apt.
Click to expand...

I don't think it's timidity, I think its more that those who voted Brexit/Trump included a lot of people who simply regarded the pollsters as part of the 'establishment' that had ignored their concerns and simply refused to cooperate.
 
You must log in or register to reply here.

Back
Top Bottom