A Graphical and Statistical Analysis of the Judgment of Princeton Wine Tasting

Abstract

The wine ratings from the a wine tasting comparing wines from a young wine region, New Jersey, to wines from Bordeaux was held June 8, 2012 in Princeton, NJ. Graphical analysis revealed substantial differences in the use of the rating scale between judges, both in the centroids of their scores and the variability in their scores. Analysis of variance failed to detect any significant differences in the white wines regardless of data transformation or statistical model. Analysis of raw data from the red wines suggested significant differences, but severely violated ANOVA model assumptions and was invalid. Rank transformation, standardization, a model with heterogeneous variances, and Friedman's test all indicated no significant differences among red wines. Based on the confidence interval on the difference between all New Jersey and all Bordeaux wines in each flight it was estimated that differences larger than 1.5 on the 20-point scale would have been declared significantly different for either white or red wines. Therefore, the tasting was powerful enough that any meaningful differences between the wines from the two origins would have been detected. (JEL Classification: C19, Q19)