In and out of projects: a simple model and an experiment on the SourceForge development community

The allocation of time and effort within the communities of Open Source Software developers is an interesting yet relatively unexplored area. How can coordination be achieved, absent monetary rewards? How do developers choose where to direct their efforts amongst the thousands of existing software projects? How come the vast majority of Open Source projects is a failure, i.e. does not go beyond the announcement phase? The paper proposes a simple dynamic stochastic model that addresses these issues. The model is a non-strategic N-player dynamic interaction, in which players are asked to open, join or leave projects that are assigned a random probability of survival, a proxy for the project-launcher skills. The number and nature of open projects varying with time, players face a continuously chang- ing landscape. The model is simulated using a simple agent-based code, and tested in a lab with human subjects. Results show that the model can replicate most of the stylized facts of the SourceForge.net dataset, namely the high number of lurkers hopping from project to project and the highly skewed distribution. The experimental evidence supports the main behavioral hypothesis of the model, interestingly showing that human subject tend to consistently launch more projects than maximizing behavior would imply