Due to the need to formalize models relating latent concepts, Partial Least Squares Path Modeling has been widely used in marketing research for the quantitative analysis of customer satisfaction. However, in marketing applications latent concepts are expressed as a synthesis of variables that cannot be measured strictu sensu. Typically, in fact, the consumer is asked to express the level of agreement to a statement, or a judgment about particular characteristics of the offered product or service, choosing one out of a set of ordered response levels. Variables observed in such a way, however, cannot be considered numerical, as they are not measured on an interval scale. Nearly always, in order to directly obtain quantitative values, the interviewer asks the interviewee to associate the agreement level to one of the values on a certain scale (e.g. 1-10 or 1-100). As a matter of fact, this procedure implies an a priori quantification of non-metric variables, which follows two rules: • Quantifications are different for each level and equally spaced • Metric is the same for all the variables in the model. Through a sensitivity study, we investigate how the choice of different quantifications affects model quality and parameter estimation.