The economics discipline has been based on the abstraction of human action into units, which can be quantified and used in mathematical models. This may even be viewed as one of its definitional and methodological characteristics. Yet the handling of certain concepts has always been delicate, when they are of a qualitative nature and also core concepts of other social sciences (e.g., institutions, norms). The paper argues that in its venturing further into neighbouring social sciences, mainstream economics departs from the elementary premises of analytical rigour. Specifically, it shows that some concepts are mathematisable, i.e. built in order to be measurable, but other concepts are inherently not mathematisable, i.e. cannot be transformed into variables that can be used in mathematical models, in particular because they do not have the required properties for being modelled, such as the stability of the reference (e.g., norms, etc). In its use of such concepts mainstream economics practice more an absorption than conceptual exchanges across social sciences. Aggregate behaviour can be predicted by mathematical forms: yet, regarding a non-mathematisable concept such as that of norm, the possibility of building a variable that accurately represents it in a model remains indeterminate. Mainstream economics in fact transforms ex post observations of regularities of a given aggregate behaviour into a holistic ontology of human beings (i.e. all modalities of behaviour can be variables in mathematical models). Despite claims of superior scientificity by mainstream economics, this transformation is unjustified.< Réduire