Reply #218: Here is the blog entry by Krugman that I was remembering:

[eomer]

...

I think this is an acceptable practice, as long as you keep your perspective. The models clearly aren’t literally true, and in no sense are you testing your theory. What you’re basically doing is elaborate thought experiments that are somewhat disciplined by the data, and which you hope are more informative than just plain guesses.

The point is that if you have a conceptual model of some aspect of the world, which you know is at best an approximation, it’s OK to see what that model would say if you tried to make it numerically realistic in some dimensions.

But doing this gives you very little help in deciding whether you are more or less on the right analytical track. I was going to say no help, but it is true that a calibration exercise is informative when it fails: if there’s no way to squeeze the relevant data into your model, or the calibrated model makes predictions that you know on other grounds are ludicrous, something was gained. But no way is calibration a substitute for actual econometrics that tests your view about how the world works.

Yet the freshwater guys have in large part exalted calibration into a substitute for econometrics. (I recall a story about one freshwater department that wanted to eliminate econometrics from the graduate curriculum and replace it with analysis, on the grounds that the models weren’t yet ready for testing. If anyone knows more about this, can you let me know?)

...

http://krugman.blogs.nytimes.com/2011/10/18/calibration-and-all-that-wonkish/

His point describes better than I could what I think about the Bazant model: it's an interesting thought experiment but gives very little help in deciding whether you are more or less on the right analytical track.