A qualitative investigation of UML modeling conventions

Analogue to the more familiar notion of coding conventions, modeling conventions attempt to ensure uniformity and prevent common modeling defects. While it has been shown that modeling conventions can decrease defect density, it is currently unclear whether this decreased defect density results in higher model quality, i.e., whether models created with modeling conventions exhibit higher fitness for purpose. In a controlled experiment¹ with 27 master-level computer science students, we evaluated quality differences between UML analysis and design models created with and without modeling conventions. We were unable to discern significant differences w.r.t. the clarity, completeness and validity of the information the model is meant to represent. We interpret our findings as an indication that modeling conventions should guide the analyst in identifying what information to model, as well as how to model it, lest their effectiveness be limited to optimizing merely syntactic quality.