Statistical Software Engineering

Abstract : This report spotlights problem areas in software engineering to which the application of modern statistical methodology can be fruitfully applied. It was produced by an expert cross-disciplinary National Research Council panel. The panel deliberated, discussed, and made recommendations based upon their experience and on information gathered at a two-day public forum at which there were presentations by 12 software and industrial engineers, scientists, and statisticians. The most important findings are: What is needed to address the challenge of cost-effectively building huge high quality software systems is productive interactions between software engineers and statisticians. Essential catalysts for this interactions include: Following a collaborative model that partners statisticians, software engineers and a REAL software process or product; following a model for data collection and analysis that ensures availability of high quality data for statistical approaches to software engineering issues; and paying attention to relevant issues in education, such as in the areas of designed experiments, exploratory data analysis, modeling, risk analysis, attitudes toward assumptions, visualization, and statistical computing tools.