Background. Software development is a human-intensive activity. In the pursuit of translating product requirements into executable code, defects are introduced. There are several techniques in current software development methodologies used to find those defects, including peer reviews. This project investigates measurements of the peer review process and provides recommendations on the use of statistical techniques to predict product quality and peer review efficiency.

Methods. Base calculations of defect density and peer review efficiency are performed. Basic statistics of the calculated data are generated and individual distributions are identified. Possible methods for statistical prediction include: multivariate regression, decision trees, logistic regression, cluster analysis, neural network analysis, and statistical process control.

There are approximately 6,100 peer review records. This data are drawn from 7 projects representing over 15 years of peer reviews conducted in a company in the aerospace industry.