Technology adoption in input-output networks

This paper studies how network structure can affect the speed of adoption. In particular, we model the decision to adopt Python 3 by software packages. Python 3 provides advanced features but is not backward compatible with Python 2, which implies adoption costs. Moreover, packages form input-output networks through dependency relationships with other packages, and they face an additional adoption cost if the dependency packages lack Python 3 support. We build a dynamic model of technology adoption that incorporates such a network and estimate it using a complete dataset of Python packages. Estimation results show that the average cost of one incompatible dependency is roughly three times the cost for updating one's own code. We conduct counterfactual policies of community-level targeted cost subsidies and show that network structure is crucial to determine an optimal policy of cost subsidy.