Statistical Analysis of Double-Blind Tests for Multiple Audiences
Statistical analysis of listening test results suggested to date have all been based on the binomial distribution. In the case of multiple audience tests, this assumption does not allow for differences in hearing acuity (i.e., varying ability to detect differences among equipment). The result is an overly optimistic estimate of the confidence interval for the p value, or premature rejection of the null hypothesis that there are no differences among equipment. This paper proposes the use of a compound distribution called Polya-Eggenberger distribution, which allows for variation in hearing acuity across the audience. A preliminary analysis of some real world data indicated that it is able to replicate the observed mean and variance of real data to a much better degree than the binomial distribution can.
This paper costs $20 for non-members, $5 for AES members and is free for E-Library subscribers.