Dowsing tests have the "control group" built-in to the protocol. The test compares dowsing results with that of random chance guessing, and the random chance guessing is the control group. For anyone who does not believe that guessing produces a certain statistical distribution, then what you could do is add a bunch of guessers to the protocol and then compare the dowser's results to the guessers'. Yes, the dowser would know that they are the dowser and the guessers would know they are guessers, but the purpose of the test is to determine a particular person's dowsing claims. There is no way to test that without the dowser knowing they are the dowser, nor is there any need.