Proceedings of the 25th International Conference on Software Engineering, pp. 570-577.

See listing at ACM Guide to Computing Literature

Abstract:
Improvements in the software development process depend on our ability to collect and analyze data drawn from various phases of the development life cycle. Our design metrics research team was presented with a large-scale SDL production model plus the accompanying problem reports that began in the requirements phase of development. The goal of this research was to identify and measure the occurrences of faults and the efficiency of their removal by development phase in order to target software development process improvement strategies. The number and severity of problem reports were tracked by development phase and fault class. The efficiency of the fault removal process using a variety of detection methods was measured. Through our analysis of the system data, the study confirms that catching faults in the phase of origin is an important goal. The faults that migrated to future phases are on average eight times more costly to repair. The study also confirms that upstream faults are the most critical faults and more importantly it identifies detailed design as the major contributor of faults, including critical faults.