Modular Design and Option Value: The Impact of System Architecture on Developer Productivity

Abstract

How do system design decisions affect the ability of the system to adapt to uncertain future demands? While a variety of studies explore the link between a system’s design and its technical performance (e.g., the number of defects experienced in future) few empirical studies provide robust quantitative data on how design decisions create (or destroy) value through their impact on a systems ability to adapt to future needs. Despite strong theoretical and descriptive evidence that modular (i.e., loosely-coupled) systems are easier to adapt than tightly-coupled systems, the lack of empirical confirmation of the size and direction of such differences represents a serious gap in our knowledge with respect to how modular designs create “option” value. We address this gap by analyzing the relationship between design decisions and developer productivity for a large, successful commercial software system. Our analysis relies upon measuring the level of coupling of 14,000 components in the system, and using these measures to predict the productivity of developers over 8 successive six-month periods. Critically, we adopt a panel-data approach, to control for differences in developer skills, which are likely to dwarf other predictors of productivity. We show that a developer contributing to the most modular parts of the system is 75% more productive than a developer contributing to the least modular parts. Our findings are a critical first step in highlighting the option value of modularity, in that adapting to future demands (i.e., creating new features) takes less time and effort in modular parts of the system