Why 100%? Everyone seems to think that they can run a couple of subjects, or even themselves only, in a double-blind test and come up with some sort of answer. Yeah, sure. A positive result in a DBT speaks for itself. However, the converse, a negative result, is another creature. What kind of preliminary data and power analyses were performed to determine if the N (sample size) were appropriate?

Some extremes:

1) A person runs a DBT 100 times. He scores 53 out of 100 correct. Would anyone doubt that this is chance?

2) 1000 people run a DBT 100 times. Each of them scores between 51 and 55. If all 1000 people scored above 50%, any analysis you care to perform will tell you that the difference, though small, is real.

Now put that first person back into the second scenario. What does that score of 53 really mean? Is it still chance? If you haven't done the appropriate design, and used an appropriate N, who knows?

DBT as performed in audio shows an appalling lack of understanding what the test is actually sensitive to. It has a reasonably low level of Type I error, that is, saying a difference is present when in fact it is not. However, the probability of Type II error, saying a difference is NOT present when in fact it is, is astonishingly high...so much so that the test is useless without an N orders of magnitudes higher than those normally used in this kind of study.

A useful test must show low probability of both Type I AND Type II error. This is where the DBT/ABX crowd have their little blind spot.

If somebody can show me a DBT in audio where Type II error was even considered, much less controlled for, I'd be interested in the reference.