« Webinar Thursday: How R is used to optimize tractor production at John Deere | Main | In case you missed it: October 2012 Roundup »

November 07, 2012


Feed You can follow this conversation by subscribing to the comment feed for this post.

Why aren't we just calling this "statistics"?

How exactly did he "win the election"? Obama didn't win because Silver predicted he would win, Silver predicted he would win because his data said that Obama was going to win. Had he predicted a 90% chance that Romney was going to win, Obama still would have won. The irony is that even in that case it could still be said that Silver was "right". This just happened to be one of the 10% of elections where Romney won. You make the very common mistake of mixing up prediction with causation.

I agree that his work here was very, very good and shows the possibilities that using data gives for people that want to see it. I think Silver himself explains quite clearly why people like Rove act as they do even in the face of contrary evidence. The pundits don't make their living off of uncertainty, so I wouldn't hold my breath for pundits on either side to change one bit. I hope that this at least does get more people to read Silver's most recent book since it explains a lot of this very clearly.

And for the record I voted for Romney, but I knew from the data that his chances of winning were very slim. Not everyone on Romney's side rejected Silver's numbers and I definitely don't hold him responsible for the outcome. He had no influence on the actual outcome. He was just the messenger and often too many people conflate the messenger with the message.

@EvanZ This sounds far more complicated than statistics. Statistics can be used to estimate basic probabilities or to express the end result, but to call what he's doing "statistics" probably is a vast oversimplification. His analysis involves running repeated complex simulations that take into account a very large number of variables. Changing any one of these variables likely influences the value, weight, and method of calculation of a large number of other variables. In the end, he expresses the results in statistical terms, but the process involves analyzing a matrix of interrelated factors and distilling each possible outcome into a probability that we can understand. Regrettably, many of his critics chose to attack him directly rather than trying to understand and criticize any possible flaws in his methodology.

@EvanZ is right, one shot does not explain anything in probability. But, Silver has a little series with all winns from 2008 ongoing. Darryl Holman does the same, but fully discloses his method, also predicting right since 2008.http://horsesass.org/?page_id=39659#Q19

Put together we have an argument for statistical model prediction. Not a strong yet but about 6 wins in 6 predictions if I am correct.

@Joseph: I suspect your definition of statistics is too narrow and not what is lived in the real world. Anyway debating over definitions is seldom really useful, as Karl Popper argued.

Thank you for this well-written post. This is an excellent summary of the statistical methods that are essential for modern data analysis. You've presented the salient points with just the right level of detail.

As you know, 2013 is the International Year of Statistics. The idea is to let non-statisticians know what statistics is (and isn't), and how it makes a difference in the world. I intend to write a few blog posts like this one in 2013 that are aimed at a general audience. I encourage you to do the same. With this post as an example, you've set the bar pretty high.

The comments to this entry are closed.

R for the Enterprise

Got comments or suggestions for the blog editor?
Email David Smith.
Follow revodavid on Twitter Follow David on Twitter: @revodavid

Search Revolutions Blog