How Nate Silver Failed To Predict Trump

[Jefferson23]

The political world’s go-to numbers guy now predicts a series of likely wins for the Republican front-runner.

by Tina Nguyen

If you want to understand how the entire American political establishment failed to predict the hugely successful candidacy of Donald Trump, look no further than the existential bafflement gripping Nate Silver, the election-predictions prophet who, after insisting for months that Trump had no shot at the nomination, is now close to accepting the reality that the billionaire real-estate developer could become the Republican presidential nominee.

Silver, a statistician by trade, gained a sterling reputation in the 2008 and 2012 election cycles by predicting the exact outcome of the presidential elections, nailing every state's electoral college vote in 2012. So it was natural when the wonks of the world accepted his premise last year that Trump was a fad and that Hillary Clinton would coast to the nomination. “Dear Media, Please Stop Freaking Out About Donald Trump’s Polls,” blared one Silver headline as late as November 23.

While Silver admitted in August that he and his FiveThirtyEight team were “going to make plenty of bad predictions over the course of the 2016 campaign,” it was easy to assume that those “bad predictions” would be due to the normal series of poll-dingers: steamy scandals, gaffes and blunders, and flagging fundraising. Yet here we are on the first caucus of the election, and Trump is poised to win not only Iowa, but nearly all the primaries leading up to the nomination.

In the past few days, Silver has published a long article parsing several theories as to why Trump has continued to have staying power and, in a telling move, conceded that perhaps his initial working premise, adapted from the book The Party Decides, is flawed. The 2008 book had become the go-to tome for analysts headed into 2016, in part because of its intuitively simple premise: that the political “establishment,” i.e., a group of ideologically aligned influencers such as politicians, celebrities, and donors, hold the authority to direct Americans to vote for one party and to further their own personal interests. “The book does presume that, in part because of their breadth and diversity, American political parties are strong institutions,” he wrote. “Furthermore, it assumes that strong, highly functional parties are able to make presidential nominations that further the party’s best interest.”

http://www.vanityfair.com/news/2016/02/donald-trump-iowa-nate-silver

[elljay]

1. Statistics

For statistics to be useful you need to have a large enough controlled sample that allows you to derive meaningful conclusions. The problem with this election is that all of the rules have been tossed out the window for both parties. This trend wasn't obvious at the beginning of the election when Jeb! and Hillary were the expected winners. Silver now has a large enough sample to show that the old statistical analysis doesn't apply. I generally respect his work, but suspect that he is just not going to be able to predict as accurately this cycle.

[Jefferson23]

2. His play book is off, where the establishment and cohorts gave a definitive edge..he needs

to adjust, which is what I found interesting.

[pansypoo53219]

3. MITTENS WAS SUPOSED TO WIN! the right SUFFERED MORE OBAMA!

what more do you have to understand.

[bemildred]

4. Statistical predictions are always based on what has happened already.

Because that's what statistics does, summarize past results mathematically, and predict what will happen today and tomorrow based on the assumption they will continue past trends.

They therefore always predict that tomorrow will look pretty much like yesterday, and they always miss real change.

I always ignore all polls. They are often right, but when they are wrong they are really wrong. And they are often wrong, it's just hard to say how beforehand.

[malthaussen]

6. But... but... but...

My team has won on ten consecutive Tuesdays, and they're playing on Tuesday! Go, My Team!!!

Our media is full of pundits who try to attach statistical meaning to things that are absolutely meaningless. It gives them something to say. (Of course, Nate Silver is different; at least he tries to draw conclusions from meaningful data)

I think a lot of it comes down to gambling and superstitions, coupled with the desire to get cred for being a prophet (and profit thereby). The wise gambler bets the actual odds for an event. The superstitious one is always looking for that intangible edge to beat the odds. The wise gambler will make steady, if unspectacular gains (provided he isn't playing in a casino, where the house's cut will eat away his profits). The superstitious one will always lose if he keeps it up, but of course he could get lucky and win big. Funny, though, how often, when the latter does occur, he proceeds to piss it all away instead of cashing in and walking.

-- Mal

[bemildred]

7. Don't get me started. It gets ugly.

[malthaussen]

8. I love statistical analyis.

There are so many cool little things that one can learn. But, ye gods, the general public is so ignorant of the fields of statistics and probability. "Law of averages," anyone?

Here's a good one, though, for your amusement: Pitcher Dazzy Vance of the Brooklyn Dodgers had an uncanny ability to win on Mondays in Brooklyn. It apparently surpassed all expectation or possibility of statistical illusion. After examining the problem for some time, it was concluded that Dazzy won so often because... Monday was wash day. The citizens of Brooklyn would hang out their (white) washing where it could be seen from the plate, and this caused the batters to have sufficient trouble seeing the ball, that they hit worse. Fewer hits, fewer runs; fewer runs, more wins for Dazzy. (Which, of course, doesn't tell us why Brooklyn's hitters were not also affected by the wash; presumably the good citizens of the borough did not take in their undies every half-inning. However, playing half their games at home, they may have gotten used to it)

Of course, one could just have gone ahead and bet on Dazzy every Monday at home, and figured some statistical voodoo was at play.

-- Mal

[bemildred]

9. Theorems are a lot easier to come up with than proofs.

We are very good at thinking up reasons why things have happened, that's a good example of that.

What we are not so good at is winnowing all those theories down to the few that will work, can be relied on. There we tend to cling to the past, to our own ideas, and to what makes us feel safe.

And statistics makes it easy to do that, you can support any argument with statistics.

[CTyankee]

13. but how do you know to inform your statistic? It could seem like an outlier...

there might be a helpful model about the rise of fascism in other countries as in what are the characteristics of fascism in society leading up to a fascist government but where do you look to get a replica of you model?

[bemildred]

15. Elaborate on that please.

As it sits I am not sure what we are talking about. What statistic are we trying to inform?

If you are looking for patterns, there is no shortage, we are very good at seeing patterns, that was kind of the point, so good we see them all the time when they aren't really there.

If the object is to predict and prevent Trumps, you can't. You have to watch or take the approach King Herod did a couple thousand years ago, and as we all know that still didn't work.

There is no substitute for paying attention, being a citizen is work.

[CTyankee]

16. I guess that is my point. "Past is prologue" is a good general idea and a worthy warning but

Trump was definitely an outlier.

[bemildred]

17. Yes. Outliers are the problem, and emergence.

If you think about it, none of this around us was evident in the big bang, it just happened for reasons we dimly understand parts of.

We wave our hands and say the math is just too difficult, but the fact is the math doesn't exist, and when it does it has finite precision and cannot predict outside what that allows. (Mal mentioned that, the problem of finite precision.)

Sometimes I look around in my room here and think about how all the things I'm looking at could be different in a thousand arbitrary ways, for some there is a reason I'll never know, of design or art, for others nobody knows why really, the reason is obscure or diffuse. And that makes it more interesting.

[CTyankee]

18. we could start with fascism and its emergence and go from there...we hear so much

from pundits about "anger in the american voter." Factor in the repub. party's penchant for blaming the victims, as Paul Krugman has pointed obout, and you could predict a type of personality that could emerge to rally the disaffected. So there was some handwriting on the wall...the timing is hard as is the "who" of the question. A year ago no one was saying "who is likely to emerge? Oh, yeah, Donald Trump. Perfect personality type. Nope.

[bemildred]

19. Yes. But that's not statistics.

I've been expecting someone Trump-like, in the sense of a free-lancer crashing our political system since Citizen's United passed, but that is educated opinion based on my reading of our history. We've had plenty of Trumps, Citizen's United made them independent of the national party apparatus. Trump does not need Reince Priebus for anything. Or any of them. That is his selling point.

It's not the first time Trump has tried it either, but this time people are listening because TPTB have discredited themselves enough that even the less well-informed get it.

The self-pity is what you have to watch out for, people sufficiently feeling sorry for themselves do all sorts of nasty things, as we see every day now.

[KoKo]

27. Yes...he finally was the "Man of the Moment" after trying for years...

But, in his late Mid-Life Crisis...he has finally found his time of glory in the sun. This is what I see of him. He was bored. The kids were grown and running the business...his challenges were getting fewer and, a man with that big an ego, just needs to "Go For It" ...one more time. Maybe Rupert Murdoch's lifestyle just didn't suit him (he has had enough wives) and his ambitions were greater than being head of a vast Media Empire which would be just too confining.

There's much about him (his OTT language and strong-arming tactics) that smacks of desperation. However, I don't fault him for trying. We all have times of challenge in our lives and we deal with it in many ways without the abundance of connections and funds that Trump has access to. Look at Bill and Hillary who are also in late Mid-Life Crisis and they are now competing with "The Donald" over the spoils of what's left of failed policy both Foreign and Domestic, Post WWII for Decades now in the USA.

What I do fault both candidates for, though, is that it is all about THEM and not really about the well being and future prosperity and concerns of the people who are suffering and left behind in a sagging economy. Even those who bided by all the rules, (including going into massive debt to get advanced degrees), who are now still dealing with the aftermath of 2008 Wall Street Crisis and the money drain in our economy for our Endless Wars.

Neither side is running to "Serve "The People." It is and will be all "about Business as Usual"...and "War As Far as we Can Imagine" in our future. UNLESS THEY "F...UP! And, that is highly likely.

[bemildred]

31. The best way to lead is to figure out where everybody is going and then hurry to get in front.

I'm surprised more of our politicians don't understand that.

[bemildred]

35. Open primaries have undercut the power of the two main parties too.

Without the captive audience it is much harder for them to control who gets nominated etc., as we see today. I've been waiting for someone to go overt about it and directly address the other parties voters, Trump is doing that some but not that overtly yet.

[egold2604]

5. models work until the situation changes and the past is no longer a predictor of the future

[CTyankee]

12. and in knowing basics like how the Margin of Error is greater in a small poll size and

over how much time...just math...someone tried to explain that to Joe Scarborough the other day and he laughed if off, saying sounded like something walter mondale would say...sheesh...

[CTyankee]

14. then the model must be adapted to the current situation and that is difficult...

[egold2604]

26. The models are built by people who understand statistics then turned over to interns

Last edited Thu Mar 10, 2016, 10:38 PM - Edit history (1)

Nat builds a structural logistic regression model which is generally run by interns. He is too busy giving interviews to actually run the models any more. You have to be totally aware of current situations and how it will affect the model. Then he publishes a "number" with no variance around it and the number take a life of its own. The user has no understanding of statistical modeling and assumes, incorrectly, that the number published is the real number.

Remember, MBA students only have to take 3-6 credit hours of business statistics. They rely on the experts to give them an answer and they vaguely remember coefficient of variance, or any other statistical tests or what they really mean.

We have external events that have never been seen before. We have the GOP clown car running for president and you have Hillary and Bernie. I would never want to forecast this mess

[CTyankee]

29. I am no statistician (fine arts major!) but I mentioned outliers upthread which of

course, Donald Trump has been. He must have figured out that his outlier approach could work. IDK how he arrived at that conclusion but he hasn't fizzled like Pat Buchanan. Of course, Trump has his own fortune to pay for it which Buchanan didn't have.

I have a pretty good political sense based on a long experience in these campaigns and when Trump burst upon the scene I said we should take a "wait and see" attitude because it looked like he spotted a trend. Ever since my "fling" with John Edwards I have sworn not to be fooled again. That was tough...

[Nitram]

10. Better yet, HOW EVERY BODY FAILED TO PREDICT TRUMP.

[Jefferson23]

11. Yea, but not everyone makes their living off of being a political prophet.

He has his limitations which included relying on the establishment's support trends
of the past. When you look at the percent of GOP voters who feel their party
betrayed them and its at 60%...he doesn't seem to be such a genius.

[CTyankee]

20. I'm sure you are right, he seized the opportunity given his ego and power...he thinks

this will overcome all, but history shows that eventually, these people fail...some with dire results...

[Jefferson23]

23. He saw an opportunity and did well with it for that period of time. I posted the OP I felt

highlighted that human nature can't be analyzed as well as he seemed to think.

People will absorb just so much bullshit, to be blunt...how they react to it is
not always predictable, as Nate has learned. Unfortunately the establishment
within the GOP are still lying to them when they tell them why they should
not vote for Trump. That is concerning to me, it is true Trump is hurling every fear
they have, but the GOP did that too..just somewhat milder. Romney did not
say, hey..WE will stop being the party that only looks after the investment
class..which leaves their base still ripe for a guy like Trump or much worse.

[KoKo]

22. Caught up in the wonky numbers...total math genius..but...

forgets to include "Variables/Deviations" into his models?

I think he focuses on the numbers and doesn't get into the "personalities" of the people he is checking stats for?

All Good Minds sometimes go Astray...I've read.

As a Bernie supporter I've hoped that Nate might have his figures wrong with Bernie, this time, because he can't really get into "Tides of Change?" Or, he is correct and we peon brains just can't get his statistical brilliance.

Anyway...he had a great run with Obama predictions. I'm not sure that this election hasn't screwed with everyone's brains as to what goes forward. Not blaming Nate.

[Jefferson23]

24. People are under a great deal of stress and they can react in many ways, it seemed clear

Nate does not factor that in his model. It's not that he is not intelligent it is
how he uses his intelligence..he needs to adjust to some degree that which
reflects the American psyche...I feel.

[KoKo]

21. Very Interesting...Thanks for Posting...

Nate was the Guru for Obama Polls...but, no one can be perfect all the time. And, he has moved on since then...but, still interesting to watch how he predicts these days.

[Relentless Liberal]

25. Nostradamus couldn't have predicted Trump

Cut Nate some slack, man!

[ian cameron dromore]

28. 88% of statistics are made up

[CTyankee]

30. "Lies, damn Lies, and statitics."

[bemildred]

32. How to Lie with Statistics

5.0 out of 5 starsvery popular account of how statistics can be misused
By Michael R. Chernick on February 12, 2008
Format: Paperback

Statisticians hate the old adage "Lies, Damned Lies and Statistics", but statistical methods do have that reputation with the general public. There are many excellent accounts, some even understandable to laymen that explain the proper ways to analyze, study and report the analysis of statistical data. Huff's famous account is illustrative and well written. It gives the average guy a look at how statistics is commonly misused (either unintentionally or deliberately) in the popular media. Graphical abuses are particularly instructive. Readers should recognize that statistical methods are scientific and with proper education anyone should be able to recognize the good statisticians from the charletons. For now Huff's book is still a good starting place. As a statistician I hate the public image portrayed in the quote above. However, I do sometimes have fun with it myself. As I write this review I am in my office wearing a sweatshirt that reads "When all else fails manipulate the data."

A modern book by a consulting statistician on the same topic is "Common Errors in Statistics and How to Avoid Them" by Phil Good. If you enjoy this book take a look at Good's book also.

http://www.amazon.com/How-Lie-Statistics-Darrell-Huff/dp/0393310728

[CTyankee]

33. ep. You can phrase the question carefully and deviously...got that...

[bemildred]

34. Very popular.

Removing "anomalies" is trickier, people are onto that. I remember when I was first doing software engineering (programming) at JPL in the 80s and these supervisor types were talking about software "anomalies" because they were not comfortable with admitting the software had bugs. I had a lot of fun with that, made a few enemies.

But putting all the slant in the questions and the language is considered very clever.

And the other thing is unobtrusively selective methodology, which is tricky because a sample is a selection, so there is a lot of room for "error" there.

In academic circles you don't get away with that much, even today, but in politics anything is good to go.