Theoretical versus practical complexity: Complexity measure development for the Unified Modeling Language using a Delphi study

Systems development methods have become more complex, concurrently, perhaps even as a result of the added complexity that has come to characterize many of today\u27s systems and applications. As a recently constructed modeling language, UML (the Unified Modeling Language) has been criticized for the complexity it presents to its users, and those seeking to learn to use the modeling language. Using Rossi and Brinkkemper\u27s complexity metrics, Siau and Cao completed a complexity analysis of UML and 36 other modeling techniques, finding that UML ranged from 2 to 11 times more complex than the other techniques. Erickson and Siau proposed that Rossi and Brinkkemper\u27s metrics present the theoretical maximum complexity, as opposed to a practice based complexity, that must be less than the theoretical maximum. This dissertation proposes that a subset of UML, identified as a core or kernel, and composed of the most commonly used constructs, can be equated with practical complexity. A Delphi study was conducted using practitioners as the experts, in an attempt to identify a practice based kernel of UML. The Rossi and Brinkkemper metrical algorithm was then applied to the resulting kernel, and is presented as an alternative measure of the complexity faced by those using UML. The Delphi participants identified a kernel of UML that resulted in an approximate 61% reduction in complexity of the overall language, when the Rossi and Brinkkemper metrics are applied to the kernel. While the reduction still leaves UML as an extremely complex language, the results indicate a number of implications that can be drawn from the research. First, it appears that developers in practice use UML more in the Analysis stage than in later stages of the development process. This leads to the second point, that developers are using UML at least somewhat differently than the creators of the language intended. Finally, for educators, the results can provide a pedagogical guide map for teaching UML in the classroom