Tree based techniques in statistics are gaining a renewed interest in the Big Data era because they constitute a pervasive tool that entails effective interpretation of results. In this setting, we have addressed the goal of providing a tool for building trees for ordinal responses in a model-based framework. The procedure relies on a class of mixture models for evaluations and preferences whose characteristic feature is the probabilistic specification of uncertainty. As a result, an added value of the proposal is that every node is associated with a distinctive level of indeterminacy. A twofold advantage is then achieved: first, explanatory variables for the response can be assigned an importance degree according to the deepness of the implied binary split along the tree, and second the terminal nodes directly correspond to alternative profiles of respondents. Then, exploiting a graphical feature of the chosen modeling, similarity of the clusters so determined can be further investigated. We present and discuss the results of an application to a very mainstream topic: the perception of happiness, that show how the integration of tree methods with the chosen modelling is appealing and effective.