Thursday, August 06, 2009

On statistical significance

Saw this nice little essay by Tim Harford some time ago, I don't have much to say about it but others might like to read it too. One thing I've long been aware of is how much of statistics has been driven not through the "trickle-down" theory of ivory-tower analysis eventually permeating practical applications, but rather by a ground-up process where people are driven to develop methods primarily to address specific practical problems that they are dealing with. As well as "Student" Gosset at Guinness (that Tim mentions), Fisher himself did most of his important work at Rothamstead, a sister lab to the place I worked some years ago. It did occur to me a few years ago that some certain climate scientists might have been developing important breakthroughs in their methods for probabilistic prediction, but unfortunately it soon became clear that they were just reinventing well-established fallacies.

Meanwhile, here is Andrew Gelman on things being significantly non-zero or not...

1 comment:

Hank Roberts said...

> reinventing ...

Oy ess.
And so are the s(k)eptics.

From Gelman's thread:

Howard Wainer | February 9
This happens so often it has a name:
"The margin of error folly: If it can be, it is."