Design process improvement case study using process waiver data

Abstract

The effort to improve the quality of software has produced large amounts of data about development organizations. In seeking answers to software quality problems, a fundamental question is what can reliably be concluded from this data.

We propose a model for the empirical study of software development as a way to begin answering this question. This model defines techniques which range over the spectrum of credibility and generalizability. From the middle of this spectrum, we present an example of a case study in which real project data was used to improve an existing design process. We carefully constructed our study to be credible within the context of the specific software project, without jeopardizing the work in progress or increasing project costs.

Using natural controls with a large volume of data, we were able to test our hypothesis that the quality of a feature was not adversely affected by the removal of certain steps in the design process. This resulted in a shorter process which was scaled to development feature size and which was implemented throughout the project.

This case study also showed the importance of ensuring that the method used to capture and analyze data from large software development projects is indeed highly credible.