Arguments for Randomizing

Extract

I have organized my remarks about randomizing under four headings: computation, communication, causal Inference, and complexity. It Is hard to think of a wore controversial subject than that of randomization. My remarks are simpler and more extreme than they ought to be. I have put them in a rather bald and definite way in order to draw the lines more sharply and to make my message as clear as possible. I do not doubt that under extended debate it would be necessary to qualify some of the things I have to say, but I would insist on the point that I would be offering qualifications, not retractions.

It is often said by pure Bayesians that once the likelihood function is available knowledge of any randomization scheme used is superfluous Information. It seems to me that this argument misses an Important point which I want to illustrate by a simple artificial example.