Regression modelling of software quality: empirical investigation

Abstract

The use of software complexity metrics in the determination of software quality has met with limited success. Many metrics measure similar aspects of program differences. Some lack a sound theoretical foundation. Attempts to use these metrics in quantitative modelling scenarios have been frustrated by a lack of understanding of the precise nature of exactly what is being measured. This is particularly true in the application of these metrics to predictive models. The paper investigates some basic issues associated with the modelling process, including problems of shared variance among metrics and the possible relationship between complexity metrics and measures of program quality. The modelling techniques are applied to a sample data set to explore the differences between modelling techniques with raw complexity metrics and complexity metrics that have been simplified through factor analysis. The ultimate objective is to provide the foundation for the use of complexity metrics in predictive models. This, in turn, will permit the effective use of these measures in the management of complex software projects.