Posts Tagged‘Bayes’

The Reverend Bayes needs a new prosthesis

by reestheskin on 10/07/2014

No Comments

I like Gerd Gigerenzer’s writings (see for instance The Empire of Chance and Simple Heuristics that make us smart) and I am sure it is not his fault that the same stories keep coming round again and again. This story on the BBC web site treads over old ground but of course the lessons remain the same (even if the book is different). Doctors don’t like working using Bayes’ theorem in clinic — at least not if we have to use algebra, rather than real numbers (as Gigerenzer makes clear). And I still think we do a poor job of teaching medical students statistics. But something niggles me about his line of argument, and in part it it is not a million miles away from some of Gigerenzer’s other work on heuristics and ‘quick and dirty’ computation.
One view of expertise is that doctors somehow work from ‘basic principles’ and then work out what to do. This used to be the dominant view of medical expertise: we had to understand the physiology, so that we had a live model in our brain of what was happening to the patient. This may well be true in some instances, but more often it seems to me that the burden of knowledge to do this is so great, that we just follow simple shortcuts or heuristics— or we read it off look-up charts. I actually think this is sensible. We don’t need to fret about the molecules, just as I don’t need to worry about machine code or C+ when I write this blog. What Gigerenzer is drawing attention to is the absence of the relevant cognitive prostheses that takes care of the number crunching for us. Of course if the prosthesis existed, we would play with it, and actually become more at ease with the algebra.

Correlation is not causation, but…..

by reestheskin on 17/05/2014

No Comments

Nice graph, via John Naughton.  Well, indeed correlation is not proof of causation, but a la Judea Pearl and others, correlation may tell you a bit about causal structures.

spurious-correlations