Design architecture, developer networks and performance of open source software projects.
详细信息
文摘
In this dissertation I seek to understand the factors differentiating successful from unsuccessful software projects by investigating how design architectures and developers' positions within collaborative networks affect software project outcomes. There is a growing body of evidence that a product's architecture -- defined here as decisions made regarding how to decompose design tasks and the resulting design structure -- is a critical issue to both researchers and managers to understand. However, prior research studies on software development tend to emphasize design without taking into account the collaborative development environment, or have little power in predicting the phenomena with which the design constructs are associated because of insufficient empirical evidence. This dissertation develops and tests a model measuring the impact on software project performance of (1) software products' design architectures and (2) developers' positions within collaborative networks. Three indicators of project success are used: product quality, the signal of market success, and project velocity. Two dimensions of design architecture---degree of decomposition and coupling---and two characteristics of developer network structures---degree centrality and structural holes---are investigated for their impact on project performance. Using data gathered from SourceForge.net and the data warehouse including its monthly dumps, empirical analysis is conducted on the top 100 ranked projects and compared to a randomly chosen 100 projects. The main findings are that (1) the degree of decomposition has an inverted U-shaped relationship with project performance, (2) When tested on the sample of top 100 projects, average degree centrality of a project team has a positive and significant effects on project performance, and (3) the effects of network metrics on some of the indicators of project performance are positive in the top 100 projects and some of them are negative in random 100 projects. We employ some control variables such as number of projects each developer works on and organizational sponsorship to provide insights into the direction of the effect of degree centrality and structural holes on project success. Theoretical and practical implications are discussed, and several future research directions will be outlined.