Cardebat and Paroissien (2015) have provided a new methodology to express the scores of different wine expert on the same rating scale. Their method consists in matching the quantiles of the distributions of each experts’s scores in an exhaustive dataset. However, they use the raw empirical distribution functions, which is arguably downward biased. This leads to a frequent upward bias in the converted scores. This paper exhibits the existence of this bias, provides the sign of the latter, and suggests a simple interpolation to mitigate it. (JEL Classifications: Q13, L15, C14).
This article was written by Karl Storchmann