Network Effects, Switching Costs, and Underlying Preferences in Operating Systems for Servers: A Case of Linux vs. Windows

We seek to investigate to what extent network effects and switching costs affect the decision to adopt Linux or Windows as the operating system for computer servers. To this end, we use detailed survey data of over 100,000 establishments in the United States. To account for unobserved preferences for either operating system, we employ recently developed dynamic discrete choice panel data methods (Arellano and Carrasco 2003). The results from our empirical analysis suggest that among network effects, switching costs, and unobserved preferences, the last two are important factors in the market for operating systems for servers. We find that switching costs are significant, but can be severely overestimated by methods that do not account for unobserved heterogeneity in establishment-specific tastes for operating systems. We also find that once taste heterogeneity is taken into account, network effects are not significant.