18 May 2012

This Explains a Lot

Paul Krugman, on Asimov’s Foundation trilogy:

I first read them when I was a teenager. I was really inspired by the psychohistorians, who used statistics and social sciences to predict the future. I knew it was fiction, but what really struck me is the notion that the science of what people do could be important. I wanted to be one of those guys.

I suppose this helps to explain why Krugman is so exceedingly certain about everything he says:  He really believes that he can trust the models. Now, I am not trying to suggest that economics or other social sciences are completely worthless or unimportant, but it is foolish to think that tossing a bunch of mathematic equations and statistical analysis at human behavior is going to be a good predictor of human behavior, and thus the future.  That Krugman has idolized this pretense of knowledge an certainty goes a long way into explaining why he is consistently wrong, consistently contradicting himself, and why he is apparently unable to recognize just how wrong he tends to be.  After all, he’s simply punching numbers in some rather mathematically advanced statistical models.  The problem is that he can never be bothered to question the assumptions upon which the models are built.  And that’s what happens when you trust in science.


  1. Agreed, all except the bit where you say 'mathematically advanced' ;-)

  2. @rweg- well, it's more of a relative term anyhow.

  3. I read the Foundation Trilogy as a youngster and really enjoyed. However, this was before the famous research by Benoit Mandelbrot that dug into chaos theory.

    Also, remember that the predictions failed miserably when a wild card, The Mule, appeared. Models are gross simplifications of the real world and are always wrong on some time scale.

  4. @TGaPO- to paraphrase Clint Eastwood, a man's got to know his model's limitations. That's obviously not the case with Krugman, which is why is his so astoundingly wrong so often. His models can't account for wild cards. But, by definition, models can't.