Projecting software defects from analyzing Ada designs

Models for projecting software defects from analyses of Ada designs are described. The research is motivated by the need for technology to analyze designs for their likely effect on software quality. The models predict defect density based on product and process characteristics. Product characteristics are extracted from a static analysis of Ada subsystems, focusing on context coupling, visibility, and the import-export of declarations. Process characteristics provide for effects of reuse level and extent of changes. Multivariate regression analyses were conducted with empirical data from industry/government-developed projects: 16 Ada subsystems totaling 149000 source lines of code. The resulting models explain 63-74% of the variation in defect density of the subsystems. Context coupling emerged as a consistently significant variable in the models