Friday, October 14, 2016

Nate Silver is full of shit

OK, I admit that the title of the post is a little provocative, but I only use it because his models mislead more than they educate.  That's all.

So, I'm a math guy.  I love me some statistics.  But, you may notice that I never link to Silver, who did not, in fact, invent the field of statistics.  (Also, Kenny G. did not invent jazz and Zack Snyder did not invent superheroes).  For my numbers, I just look at the polling averages at RealClearPolitics, and of course, the betting markets.  Right now, Clinton is way ahead.  And, she's a-gonna win unless there's another October surprise.  What's Silver sayin' these days?  Well, let's take a look.  I don't check in with him often because, well, see title.  Trump has a 14.4% chance?  What the fuck?!

And that was just the "polls only" forecast.  Click on "polls plus" and Mr. Silver will tell you that Trump has an 18% chance of winning.  But fuck that.  Let's go with 14.4% just to give Silver a fighting chance.

Let's talk quantum mechanics for a moment.  Subatomic particles behave in ways that appear random.  Instead of following Newtonian laws, they randomly go either left or right at certain points, and we can find no apparent cause for why they do one or the other.  One common explanation for how that crazy shit works is that the multiverse is constantly branching off new universes, so it isn't that a particle randomly decides which way to go-- it is that two superimposed universes break apart and the particle going one way and the particle going the other were actually different particles all along.  Crazy, right?

For the sake of argument, let's accept this model of the multiverse.  14.4% is about 1 in 7.  So, in Nate Silver's model of the current superimposed multiverse, one out of 7 universes branching out from this moment lead to President Trumps.

Bull.  Fucking.  Shit.

How does Silver go so wrong?  The short version is that his method is based on state-by-state polling analysis to construct simulations of the electoral college.  But, isn't that the right thing to do?  We elect presidents through the electoral college, not the nationwide popular vote, right?  Yes, but...

...

A popular vote-electoral college split requires a close election.  That can happen.  See:  2000.  (Sort of.  After all, there was that whole ballot counting issue in Florida).  This thing isn't close.  And when it isn't close, the not-really-swing states swing together.  And this is what is really hard to capture.  What is the right degree of interconnectedness and separateness between the states?  As I've commented before, that's really hard to capture.

What isn't hard to capture is who is way ahead.  That's Clinton.  She is too far ahead nationally to worry about the messiness of state-by-state polls, which is what Silver's method attempts to do, and where Silver's method goes wrong in giving us a multiverse in which 1 out of 7 universes branching out from here have President Trumps.  Coming soon:  the right way to look at things:  the simple way.  The good, old Central Limit Theorem.

Now, Trump could still win, in which case there are universes branching out from here with President Trumps, where I revisit this post.  What will I say in those universes?  Perhaps something like this:  "OWWW!  It really hurt when those monkeys came flying out of my ass!"

Or, more realistically, "wow, that was some crazy shit when they perp-walked Clinton," or a more somber comment on a terrorist attack, because that's probably what it would take at this point to save Trump's candidacy.

Other than that, the polls would have to be completely wrong, even at the national level.  In any case, none of that would save Silver's method.  Trump's chances aren't zero, but they aren't 14.4%.

Trump could win.  But, he needs either a major intervening event, or for all of the polls to be completely wrong.  None of that would save Silver's method of simulating state-by-state contests with no theoretically driven method of dealing with state interconnectedness and independence given how swing states swing together.

And here's some Friday jazz, a bit early.  I'll post something different later, to keep picking on that Dylan nonsense.  Yes, I'll even do that in the jazz series, because really, fuck Dylan too.


No comments:

Post a Comment