I just chalk it up to more RW delusional nonsense. I was kinda amused by some of the other stories, though.

"GWU PROF DECORATES OFFICE WITH PHOTOS OF COMMUNIST DICTATORS RESPONSIBLE FOR THOUSANDS OF MURDERS"

"UNIVERSITY OF CA SPENDS $80K ON SOCIALISM PROJECT"

"STUDENT SPEAKS OUT OVER HARVARD UNIVERSITY’S ANNUAL ‘INCEST-FEST’ PARTY"

Why should i trust it?

http://www.dailykos.com/story/2012/08/23/1123481/-Nate-Silver-Calls-B-S-on-U-of-Colorado-Election-Prediction-Model

And according to ABCNews these were "retroactive" predictions.

I call bullshit.

to suggest that it may not be entirely trustworthy?

missed the "Snork"

It is a right wing site.

Here's the first paragraph of their mission statement:

"CampusReform.org is designed to provide conservative activists with the resources, networking capabilities, and skills they need to revolutionize the struggle against leftist bias and abuse on college campuses."

I (obviously) didn't accept the story at face value & now I know why.

... That's kind of like "future history".

retroactive resignation from Bain.

Both of which have longer track records of accurate predictions say Obama wins...

has never predicted anything correctly...only retrodicted past results. With multivariate models, it's very easy to manipulate to one's liking. Weight a variable here, omit a variable there. They're also notoriously unstable. Include a seemingly innocuous variable, or omit one, and the results can be dramatically different.

Social scientists also commonly abuse or simply misunderstand statistical results, especially of the multivariate variety. As a social scientist who works commonly with multivariate statistics, I am always wary of the pitfalls of seemingly "good" results. I wouldn't worry about this.

Brought to you by Colorado University?

a stronger warning...now I am going to have to shower twice with lots of exfoliant and then repeat...yuck. SOMEBODY GET ME A LOOFA!!

As noted above, you can fit anything to a set of data in retrospect. Give me ten variables, a set of historical outcomes, and I'll come up with a formula that fits those variables to those outcomes.

But what strikes me as funnier than that is the notion of a formula that provides a probability of a binary event with "perfect accuracy". What is that even supposed to mean? If I predict an outcome of X with a probability of 90%, then is my model "inaccurate" if X does not occur? The non-occurrence of X was merely a manifestation of the 10% probability that X would not occur.

The model predicted a Romney PV win. The vote didn't match the prediction. This is proof that the Socialist Muslim Uppity Kenyan forces stole the election.


Q.E.D.

For any curve, you can fit a precise regression line if you allow enough variables.

There's nothing wrong with retroactively fitting a curve as long as you classify the result as an unverified hypothesis and then conduct prospective research to confirm it.

Sid

Because there are a lot of fucking goobers in states where it won't matter how much Rmoney wins by.

But that doesn't win you the election.

And every time you try to mention 2000......"STOP BLAMING OUR LORD AND SAVIOR GEORGE DUBYA BUSH FOR EVERYTHING YOU COMMIE BASTARD"!!!