The contribution presents a pilot exercise on the analysis of differences in ranking data combining non-parametric and parametric methods. Suitable hypothesis testing procedures for discrete data are considered to identify significance differences in response profiles determined through benchmark classification trees. For marginal rankings, these response profiles can be further parameterized with a mixture of discretized Beta distribution to model polarization towards the extremes positions and floatation in between. As a by-product of the exercise, a new classification tree for qualitative outcomes is advanced that allows to disclose both global and local differences in response distributions.