| Abstract: | Multivariate statistical tests used to assess purity of natural food products, such as fruit juices, do not assume prior knowledge about the most likely causes of adulteration. We describe a likelihood ratio test that has a directional alternate hypothesis so that prior knowlege about the most likely method of adulteration can be utilized. Critical regions for this statistic are based on a nonparametric tolerance interval, which makes it insensitive to an assumption of multivariate normality. This new test is compared to two tests currently used for identifying adulterated samples. It is shown that the new test will have increased power relative to the other tests in the direction of the predicted adulteration. |
