Software Engineering

S ystem testing followed by a product release decision are the last guards in assuring software quality—insufficient testing or the wrong release decision can lead directly to the delivery of low-quality software to users. At the same time, relying too much on system testing to guarantee quality is dangerous because it occurs too late to correct poor-quality software. Moreover, previous studies have shown that bug fixing is much costlier during system testing than in earlier phases.1 Therefore, we must not only be aware of factors that increase defects but also seek possible process improvements to reduce defects before system testing. To identify and justify process improvements in individual organizations, where processes, data, and context are varied and unique, we explored using a multivariate modeling technique to analyze past development data collected in organizations. However, unlike some academic approaches, we employed a basic linear regression approach with a limited number of independent variables, each associated with what we call software engineering (SE) beliefs. These are short statements that are attention-getting, understandable, and obviously practically useful, such as “about 80 percent of the defects come from 20 percent of the modules,” or “peer reviews catch 60 percent of the defects.”2 SE beliefs are a kind of practical hypothesis that