ImOne, you are making an attempt at data searching.
That 22% success rate was for 50 tests, which really is not enough trials to achieve the expected 10% rate due to random chance. I would want to see at least 100 trials and 1000 would be enough to satisfy me. The other two tests had a 0% in 26 attempts and 12% rate for 35. All together, this gives a success rate of 12% for 111 tries.
Looking at the water test, it would seem the dowsers beat the odds. But the dowsers were claiming 80-100% accuracy, so by their own standards, they failed the test. A skeptic expects a 10% accuracy, but the 10% number only applies when enough tests have been conducted.
Similarly, we could look at the brass-finding test, which had a 0% success rate. Some will say this is proof that James Randi somehow interfered with their abilities. Even someone familiar with statistics will see a 0% success rate as odd, but still, this is not unexpected for only 26 tries.
As for looking at the top guessers, I mean, dowsers, that is also selecting data. Suppose I got 10,000 people to take the dowsing test. The overall success rate is going to be 10%, but some people are going to score much higher than 10%. In fact, I can get someone to score 100% if I test enough people for long enough. Of course, if I get that person to try and repeat the test, then the success rate is going to fall back down to 10%. That was a valid point about repeating the test, but there was no evidence that repeating the test would give results that would have given any other conclusion.
This is why the overall results were used to determine the validity of dowsing. There needed to be more than 50 tests to arrive at something approaching the expected 10% success rate. The overall success rate of 12% backs up the skeptical claim that dowsing is no more effective than random guessing. The dowser's claim is 80-100%. I have to say the test was sufficient to back up the claim of the skeptics, and debunk the claims of the dowsers.
