Model-Driven Development (MDD) with UML is gaining widespread use in the IT-industry. However, little is known with regards to the actual costs and benefits of MDD with UML. This thesis is a systematic review of 21 selected articles reporting empirical studies on MDD with UML, published between 2000 and 2005, in three leading journals and one conference proceeding.
The objective of the investigation is to get an overview of the state-of-the-art for empirical studies of MDD with UML, and show the typical information found in articles of this topic. The data collected during analysis of these 21 articles, was used to address the following issues: The amount of empirical research of MDD with UML, the extent of different empirical research methods used, what different UML-diagrams that are studied and their benefits, for which application domains UML are studied, to what extent UML is compared to other approaches, the possibilities for meta-analysis and what the authors suggest about future research.
The main conclusion of this study is that it does not exist sufficient empirical evidence to draw conclusions regarding the usefulness of Model-Driven Development with UML. The few existing empirical studies of MDD with UML do mostly suggest that UML is useful, but they are too few and deal with too many aspects of UML-usage, to allow for definite conclusions about the usefulness of the technique. Further, the main findings of this review are the following. Experiments is the most used research method within studies of MDD with UML, and this study found four times as many experiments as case studies. UML usage was found to yield benefits for the comprehension, construction and correctness of a system. However, the study also revealed that the benefits of UML-usage are often dependent on the application domain and the experience and abilities of developers. Furthermore, few studies exist that investigate the cost-effectiveness of UML-usage, and most studies have focus on detailed aspects of the use of single diagrams. Hence, the results could be useful for companies that already make extensive use of UML, but these results do not provide much guidance when UML is introduced in a company.