97-3 |
|
R. XIONG, Dept. of Food Science, Univ. of Arkansas, 2650 N. Young Ave., Fayetteville, AR 72704-5585 and J.-F. Meullenet, Department of Food Science, University of Arkansas, 2650 N. Young Avenue, Fayetteville, AR 72704. Replication is useful in discrimination testing, but replicated data are difficult to analyze because they may be dependent. Currently, beta- and other distributions are used to account for replication, which are parametric approaches requiring testing the goodness-of-fit of the assumptions. These models are more complicated than the binomial models and not appropriate for all discrimination tests. Non-parametric approaches may be useful to analyze replicated discrimination data. The objective of this study was to develop a non-parametric method for analyzing data from non-replicated and/or replicated discrimination tests. The method developed is solely based on the binomial distribution. For replicated experiments, the expected cumulative probability of the various scenarios (i.e., numbers of correct answers for a given number of trials) is used to define “pseudo success” and “pseudo failure”, then calculates the adjusted chance probability for “pseudo success” using the chance probability for the discrimination test, and finally uses the binomial test for testing the null hypothesis. Numerous published data sets were used to test the method. In one of the examples, duo-trio tests were conducted in duplicate using 15 consumers. Ten of the subject gave one correct answer in 2 trials, while 5 subjects gave correct responses in both trials. When ignoring replications, the binomial test shows a significant difference between the products (p-value=0.021). The proposed method, which accounts for replicated trials, shows no significant difference (p-value=0.314). It was found that for all the data sets used that the results of this alternative method were in agreement with those from the beta-binomial, linear mixed or other models. This method is non-parametric, easy to use, can be conducted in spreadsheet and unifies the models for non-replicated and replicated data. The proposed method implies that non-parametric approaches can be used to account for replication in discrimination testing and that data from replicated trials can easily be analyzed.
Session 97, Sensory Evaluation: Analytical testing
|