Rating data arise in several disciplines and the class of Generalized Linear Models (GLM) provides a consolidated methodology for their analysis: such structures (and a plethora of variants) model the cumulative probabilities of ordinal scores as functions of subjects’ covariates. A different perspective can be adopted when considering that discrete choices as ordinal assessments are the result of a complex interaction between subjective perception and external circumstances. Thus, an explicit specification of the inherent uncertainty of the data generating process is needed. This paradigm has triggered a variety of researchers and applications, conveying in the unifying framework of GEneralized Mixtures with uncertainty (GEM ) which encompasses also classical cumulative models. Some critical discussions conclude the paper.