Software development: processes and performance

TLDR: Two paradoxes for practice are suggested: that teams of software developers are brought together to create variability and production methods are used to reduce variability, and that team-level social processes may be a better predictor of software development team performance than are production methods.

Abstract: This paper presents data that describe the effects on software development performance due to both the production methods of software development and the social processes of how software developers work together. Data from 40 software development teams at one site that produces commercial software are used to assess the effects of production methods and social processes on both software product quality and team performance. Findings indicate that production methods, such as the use of software methodologies and automated development tools, provide no explanation for the variance in either software product quality or team performance. Social processes, such as the level of informal coordination and communication, the ability to resolve intragroup conflicts, and the degree of supportiveness among the team members, can account for 25 percent of the variations in software product quality. These findings suggest two paradoxes for practice: (1) that teams of software developers are brought together to create variability and production methods are used to reduce variability, and (2) that team-level social processes may be a better predictor of software development team performance than are production methods. These findings also suggest that factors such as other social actions or individual-level differences must account for the large and unexplained variations in team performance.