2016 Postmortem
Related: About this forumNate Silver's new predictions for Nevada Caucus
FiveThirtyEight 2016 Primary Forecast NevadaHillary Clinton has a 69% chance of winning the Nevada caucuses.
Bernie Sanders has a 31% chance of winning the Nevada caucuses.
EDIT: See below - I have included Silver's predictions for South Carolina in reply below
lunamagica
(9,967 posts)hoosierlib
(710 posts)Saturday is going to be fun...
CajunBlazer
(5,648 posts)Go to Silver's site and look at the previous prediction of a tie based exclusively on a poll with a rather bad reputation done within the last few days. Now he has one more poll.
I wouldn't dismiss Silver's prediction unless you are a better statistician than he is. Do you have any background in the subject?
hoosierlib
(710 posts)For the Nevada Democratic caucuses, weve collected eight polls. Here they are, ranked by how heavily they factor in to our latest polling averages.
= NEW A = ALL ADULTSRV = REGISTERED VOTERSLV = LIKELY VOTERS
FEB. 10-15
Opinion Research Corporation
282 LV
Clinton +1
CLINTON 48%
SANDERS 47%
FEB. 8-10
TargetPoint
1,236 LV
Tie
CLINTON 45%
SANDERS 45%
DEC. 23-27
Gravis Marketing
326 LV
Clinton +23
CLINTON 50%
SANDERS 27%
NOV. 10-16
Morning Consult
237 RV
Clinton +29
CLINTON 59%
SANDERS 30%
OCT. 3-10
Opinion Research Corporation
253 LV
Clinton +16
CLINTON 50%
SANDERS 34%
JUL. 12-13
Gravis Marketing
416 LV
Clinton +37
CLINTON 55%
SANDERS 18%
MAR. 27
Gravis Marketing
319 LV
Clinton +46*
CLINTON 61%
SANDERS 7%
FEB. 21-22
Gravis Marketing
324 LV
Clinton +38*
CLINTON 58%
SANDERS 4%
CajunBlazer
(5,648 posts)In a rapidly changing race such as this one good statisticians throw out older polls because the have be come meaningless. Obviously if Silver's were using those older polls, where Hillary was winning by huge margins, he would be predicting a much greater chance of Clinton winning.
But as always, you are entitled to your opinion.
SheenaR
(2,052 posts)Clinton +37
CLINTON 55%
SANDERS 18%
MAR. 27
Gravis Marketing
319 LV
Clinton +46*
CLINTON 61%
SANDERS 7%
FEB. 21-22
Gravis Marketing
324 LV
Clinton +38*
CLINTON 58%
SANDERS 4%
If Sanders gets 4% in Nevada, I'll shave off all of my hair which goes to the middle of my back
CoffeeCat
(24,411 posts)..then he's not really engaging in the deep analytics for which he became famous.
Anyone can look at a couple of polls and draw a conclusion.
I'd like to know more about Silver's analysis.
It looks like the last two polls were given all of the "weight." That's cool.
However, how does he get "Clinton has a 69 percent chance of winning" with those two polls---One which has Clinton and Sanders tied and one that has Clinton leading by 1?
I'd like to learn more...
CajunBlazer
(5,648 posts)Is the fact that Silver's weighting of polls contains propensities of polls to favor one candidate or the other.
hoosierlib
(710 posts)That being said, its just an educated guess based upon a small sample of polls that themselves maybe flawed (i.e. oversampling / undersampling the percentage of men and women or by race).
Its going to be interesting...but as has been the case in IA and NH, the higher the turnout, the better for Bernie.
And thankfully, unlike in Iowa, the total votes for.each candidate will be made public. So if Hillary wins more delegates, but loses the popular vote (which is my guess as to what will occur), we'll know.
CajunBlazer
(5,648 posts)LondonReign2
(5,213 posts)proposition to a 70/30 proposition?
CajunBlazer
(5,648 posts)workinclasszero
(28,270 posts)firebrand80
(2,760 posts)CajunBlazer
(5,648 posts)Hillary Clinton has a 99% chance of winning the South Carolina primary
Bernie Sanders has a 1% chance of winning the South Carolina primary
LexVegas
(6,095 posts)CoffeeCat
(24,411 posts)to get to his final guess, "Hillary Clinton has a 69 percent change of winning the Nevada caucuses."
Is that right?
The only two polls that are given any "weight" are the last two (Opinion Research and Target Point).
Is this correct?
I'm failing to see how he comes up with Clinton 69 percent chance of winning--when the result of those last two polls are the following"
Opinion Research----48 Clinton, 47 Sanders--Clinton +1
Target Points---------45 Clinton, 45 Sanders--TIE
I must be missing something.
CajunBlazer
(5,648 posts)But keep in mind that his weighting system also takes into consideration the propensity of polls to favor one candidate or the other.
CoffeeCat
(24,411 posts)So, there's no way (analytically) that one could some up with "Clinton winning by 69 percent" based on those two polls.
Is there something else that he's using?
I just saw those two polls receiving the total weight.
CajunBlazer
(5,648 posts)his weighting contain propensities of polls to favor one candidate or the other.
CoffeeCat
(24,411 posts)that he is using to arrive at his conclusion: "Hillary has a 69 percent chance of winning the NV caucuses"
These two polls are: ONE TIE, and one with Clinton winning by 1.
So, how does he arrive at 69 percent? Especially when both polls are within the margin of error.
Is it your understanding, CajunBlazer, that Silver is also factoring in those older polls (just weighting them less)? Or is it your understanding that he's only factoring in those latest two polls?
mythology
(9,527 posts)Please understand what he's saying. He's saying Clinton has a 69% chance of winning, not the he will get 69% of the vote. That is if the race happened 100 times, Clinton would be estimated to win 69 of those runs. That could be by one vote or a million votes.
Disagreeing with his analysis is fine, but if you don't know what he's arguing, it's hard to do that effectively.
hoosierlib
(710 posts)But their model includes the older polls (though weighted less)...
We got this...Nevada will go for Bernie
CajunBlazer
(5,648 posts)Note that he made no predictions until after the last poll. After the last poll which was a tie, Nate's prediction was a tie. Now with the latest poll, he is predicts a 69% of a Clinton victory. The prediction change was dependent on the last poll.
By the way, most of the weight is going to be on the last poll - the previous GOP organization is very low on the ratings on his website.
CoffeeCat
(24,411 posts)Is that your understanding?
Fawke Em
(11,366 posts)Because math and numbers don't seem to be considered in this prediction given the last two polls.
CajunBlazer
(5,648 posts)Fawke Em
(11,366 posts)based upon recent polling and no one seems to know - at least not in this thread.
Two ties don't a 69 percent change of winning make. That's not statistics, that's Uri Geller.
restorefreedom
(12,655 posts)i am not going to try and stop him
CajunBlazer
(5,648 posts)restorefreedom
(12,655 posts)i don't care if he predicts a 20 point bernie victory and it happens, i have no interest in what he has to say.
i will wait for the people to speak.
CoffeeCat
(24,411 posts)When the only two polls he used to arrive at his decision--show a TIE and Clinton with a 1-point advantage.
She's lost 20 points in NV in the past few months.
Again--Is there anything else, besides these two polls, that Nate used to get that "Clinton has a 69 percent change of winning the NV caucuses"?
CajunBlazer
(5,648 posts)And I would laugh at Silver's prediction if he predicted Hillary was going to lose.
vdogg
(1,384 posts)They briefly pulled him out, and politely dusted the tire tracks off his back, but little did he know another bus was awaiting.
CoffeeCat
(24,411 posts)I'm interested in the data.
And truth be told--Silver doesn't get into the heavy analytics that he used to. Nothing wrong with that. No one is criticizing him...or putting him "under the bus" as you stated. LOL!
Since ESPN purchased 538, he's been mainly in the sports arena. I think he prognosticates quite a bit in the area of politics, but not with the mathematical rigor and analysis that catapulted him to stardom.
I'm just wondering how he arrived at his conclusions.
That's all.
I'm wondering how these last two polls that he has given 100 percent weight to (ONE a TIE and ONE with Clinton +1, produced a "Clinton has a 69 percent chance of winning the NV caucus).
Just wondering on what he is basing that 69 percent.
vdogg
(1,384 posts)These 2 polls aren't much to go off of. I was just having a little fun after that breathless post yesterday about how Bernie had overtaken Clinton on 538. At the time it was based off one poll that had them tied at 45%
CoffeeCat
(24,411 posts)...because he's not engaged in the amazing analytics and analysis that he has in the past.
It's obvious that he loves politics and offers his opinion and surface-analysis at times. However, his opinions certainly aren't as steeped in the deep analysis that made him famous.
I don't really pay attention as much. I don't really hang my hat on what he says.
His earlier writings were absolutely riveting.
I think anyone can look at the last two Nevada polls (Clinton/Sanders TIED; Clinton +1) and draw a conclusion based on the poll data. That seems to be what Silver is doing. And that's cool. It's just not that big of a deal.
CajunBlazer
(5,648 posts)... he always used to predict elections. Once has a computer system for evaluating the accuracy and tendencies of various polling organizations, all he has to do is input poll results along with their associated statistics and turn the crank.
So why should his prediction be less accurate now?
CoffeeCat
(24,411 posts)does analyze poll data and he ranks pollsters based on their methodology. If the polls are sound, he rates those pollsters with high marks. I love that he does this, because there are many junk polls out there, and who has time to wade into their methods--which are usually pages and pages long.
I do concur that Silver does this and does this well. I'll often look up a poll and see what he has to say, before I get to excited (or bummed!).
However, I followed Silver, back in his early days when he was doing very deep analysis of not only poll data but demographics in elections. I remember him making predictions about Obama winning in Pennsylvania and he had maps pulled up and he was analyzing them, down to precinct level. He was able to demonstrate why he thought Obama would win, despite the majority favoring Clinton. So, not only was he looking at polls, he was looking at stats about support that was hidden in pockets. He was doing this for many areas of the country.
Also, Silver would write lengthy articles. He wrote one about the paradox of doing very well in early primary states while having high national polls. He asserted that this was VERY bad news for a primary candidate. And he drilled down into the statistics and what that meant for that candidate's chances. This stuff was so fascinating. He was a master at leveraging a political theory (that seemed almost counter intuitive) and then using his data to support the theory. Then, you were left believing his theory without a doubt, due to his supporting evidence.
This is the Nate Silver I loved.
I don't think he's any less credible today. I still think his opinions are interesting. However, I don't see him engaging in the same type of rich, data-based analysis.
Much of what he is doing now, is not 'trademark Silver." Really, anyone can look at a group of polls and call out averages...or dismiss a poll or two because they are biased or faulty. Or weight specific polls and not others--and dispense a number. It's not the magic of Silver that we saw several years ago.
I wish he would return to his glory days!
CajunBlazer
(5,648 posts)And I see what you meant about Silver's analysis originally being "Deeper". I really wasn't very familiar with his work until recently. Most of my insights are based on my understanding of what constitutes good statistical practices and...well...just good ole common sense.
Still my point is the his method analyzing polling organizations based on the accuracy of their past polls and their know bias toward specific parties or candidates probably follows the same methodology he used early on. In fact I strongly suspect that the entire process is now automated so that the system is not only using current poll information to make predictions. I think that elections results are also being feed into the system to be compared with poll data for those elections and the results are being used to adjust rating of the polling organizations on an ongoing bases.
If as you believe he is spending most of his time on sports analysis, this whole system could be operating with out much of his input in much the same way that a talented group of engineer sets up a complicated assembly line and then backs off and just lets it run.
If I am right about this, and the initial methodology is self adaptive enough, there is no reason why Silver's website's predictions based on polls should not be as accurate as they ever were, maybe even better as more data on polling organizations is gathered for specific situations over time.
If that is the case, even without the deeper analysis you described, I would rather go with his website's current predictions than any other other data that I know to be available.
Does that seem reasonable to you based on your obviously superior knowledge of the subject?
CajunBlazer
(5,648 posts)I'm assuming that the last CNN poll is viewed by his statistics as one that usually favors Bernie.
And you're obviously no one to throw Nate under the bus so they probably weren't talking about you.
SheenaR
(2,052 posts)as opposed to polling both sides question
http://www.democraticunderground.com/12511256524
CajunBlazer
(5,648 posts)SheenaR
(2,052 posts)If they were looking to get a 50/50 split they would not post those odds... Any square bettor looking at them says Hillary is a lock there.
CajunBlazer
(5,648 posts)stopbush
(24,396 posts)to the rafters that the election was over.
Today, not so much.
Look - 538 is making a PREDICTION based on poll numbers and other factors. Yesterday's 51% chance for BS and today's 69% chance for HRC are not poll numbers. They are a prediction of the likelihood of a win BASED on 1. poll numbers and 2. trends.
And, 538 is but one prediction entity out there.
And no, Nate Silver is not a corporate shill because his prediction swung toward HRC today.
Sheesh!
morningfog
(18,115 posts)stopbush
(24,396 posts)Not privy to his methodology, but I'm guessing it has to do with trends he's sensing.
morningfog
(18,115 posts)Alfresco
(1,698 posts)A sixth forum, a Town Hall event, is scheduled for February 18, 2016, at the "KMA event center" in Las Vegas, Nevada. It is planned to be aired on MSNBC and Telemundo.
Tom Rinaldo
(22,913 posts)Polls no longer effect the race within days of an election other than to motivate whoever is slightly behind. But they do help frame how the outcome is perceived. If Clinton goes in with the expectation that she will win and she loses anyway it would make a loss there sting even harder.
CajunBlazer
(5,648 posts)... despite the fact that they are more likely to be aware in LV, the betting capital of the country, than anywhere else.
VMA131Marine
(4,149 posts)Silver is saying that HC has a 69% of winning the Nevada caucuses but he's not saying anything about the likely margin of victory, which could be fractions of a percentage point. Based on the polling to date, if it is accurate, then statistically speaking the likelihood that HC is ahead is 69% whereas it is only 31% likely that BS is ahead.
JI7
(89,269 posts)Based on the polls we have seen.
CajunBlazer
(5,648 posts)Silvers methodology has rating for every poll organization based on their accuracy of their polls and where their is a tendency to over favor one candidate over the other.
There are only two recent polls - the first showing a tie and the second Clinton ahead by 1%.
That first poll (the tie) was done by a GOP organization which has a poor rating on Silver's scale so it is going to be lightly weighted compared to the second poll.
Now this is speculation, but I strongly suspect organization that did the second better weighted poll (where Clinton was ahead by 1%) usually creates polls whose methodology favors Sanders in some way or another. If I am right, that means Silvers system would automatically assume that Clinton is ahead by more than the 1% indicated on the second poll.
Hence the increase of Clinton probability of winning to 69%.
However, while I think the current prediction is decent (it's all we have), but it would be more dependable with another reliable poll or two mixed in.