Learning from Experience in Software Development: A Multilevel Analysis

Wai Fong Boh, Sandra A. Slaughter, J. Alberto Espinosa

Management Science · 2007 · 353 citations · 50 references

Concepts

TL;DR

Learning curves are well documented in manufacturing and services, but it is unclear whether they apply to knowledge work such as software development. This study investigates whether individuals, groups, and organizational units learn from experience in software development and whether such learning improves productivity. Using multilevel modeling on more than 14 years of data from a telecommunications firm’s incremental development process, the authors assess the productivity effects of specialized, diversified, and collaborative experience across individual, group, and organizational levels. The results show that specialized experience most strongly boosts individual productivity, diversified experience in related systems better supports group and organizational productivity, and unrelated diversified experience has minimal impact, confirming learning curves in software development.

Abstract

This study examines whether individuals, groups, and organizational units learn from experience in software development and whether this learning improves productivity. Although prior research has found the existence of learning curves in manufacturing and service industries, it is not clear whether learning curves also apply to knowledge work like software development. We evaluate the relative productivity impacts from accumulating specialized experience in a system, diversified experience in related and unrelated systems, and experience from working with others on modification requests (MRs) in a telecommunications firm, which uses an incremental software development methodology. Using multilevel modeling, we analyze extensive data archives covering more than 14 years of systems development work on a major telecommunications product dating from the beginning of its development process. Our findings reveal that the relative importance of the different types of experience differs across levels of analysis. Specialized experience has the greatest impact on productivity for MRs completed by individual developers, whereas diverse experience in related systems plays a larger role in improving productivity for MRs and system releases completed by groups and organizational units. Diverse experience in unrelated systems has the least influence on productivity at all three levels of analysis. Our findings support the existence of learning curves in software development and provide insights into when specialized or diverse experience may be more valuable.

References

50