Exploring the Effect of UML Modeling on Software Quality By

Modeling ones design is generally considered good practice in the process of software development. Numerous payoffs are attributed to this practice, in-cluding increased software quality and easier software maintenance. However, empirical studies validating these payoffs are scarce. This validation should nevertheless be considered very important, since models do not represent any value by themselves. Therefore, only when payoffs have been shown to exist, should a company be willing to invest time and resources into creating these models. The de facto standard for communicating a system’s design is the Unified Modeling Language (UML). In this study we explore the relation between the level of detail of a system’s UML diagrams – specifically class, sequence and state diagrams – and the defect density and average defect repair time of the resulting implementation. We did this by performing two case studies, both in an industrial setting, applying different approaches for each of them. For both approaches definitions were formulated on how to record UML level of detail measures of the mentioned diagram types. While the first ap-proach used qualitative rankings, the second applied design metrics to quantify this detail level. Defect samples were manually inspected to try to relate each defect to parts of the UML models. During this process the defects were also typed, so that analyses could be performed using both the entire set of defects and specific defect types when required. To facilitate this defect typing task, defect taxonomies were compiled from the ones found in scientific literature and further tailored to the needs of this study. The results from our case studies show evidence supporting the suggested payoffs: the availability of UML models lowers defect density and average re-pair time. However, some questions regarding validity of this evidence remain.