In this paper we develop a private-collective model of voluntary public knowledge production, where group-based social preferences have an impact on coalition formation. Our theoretical model builds on the large empirical literature on voluntary production of pooled public knowledge goods, including source code in communities of software developers or data provided to open access data repositories. Our analysis shows under which conditions social preferences such as 'group belonging' or 'peer approval' influence stable coalition size, as such rationalising several stylized facts emerging from large scale surveys of Free/Libre/Open-Source software developers (David and Shapiro, 2008), previously unaccounted for. Furthermore, heterogeneity of social preferences is added to the model to study the formation of stable, but mixed coalitions.