Relational event models for bipartite networks with applications to collaborative problem solving in organizations

The increased availability of information on the exact timing of interaction episodes linking pairs of social and economic agents is sustaining a renewed theoretical interest in the micro-relational dynamics underlying social and other kinds of networks. New statistical models for relational event sequences have been proposed to take full advantage of the information contained in large samples of time-stamped observations on a variety of interaction episodes linking social agents. Relational event models promise to provide novel insight on processes of emergence, change and demise of social networks starting from the observation of interaction sequences. The study described in this document contributes to these contemporary modeling efforts by extending available relational event models to sequences of bipartite associations in the context of organizational problem solving processes. The project pursues two related objectives. The first is methodological and involves an extension of the relational event model (REM) to represent the dynamics of bipartite associations – associations that are defined by relations between agents belonging to different (and disjoint) classes. If successful, the project will make available to researchers interested in the evolutionary dynamics of bipartite associations (a.k.a. “two-mode networks”) powerful statistical models that may be used to support inference about very large and complex samples of relational events. The second objective is empirical and involves a demonstration of the value and general applicability of the new bipartite relational event model (B-REM) to problem solving processes within organizations. The benchmark for success in terms of this second objective is the novel insight that the new model affords on the network structure of organizational problem solving in open productions. The model is built interactively with actual empirical data in order to test its ability to support inference. The empirical opportunity to specify, calibrate and test the new B-REM model is provided by data that we have collected on problem solving actions within an established Free/Open Source Software (F/OSS) project observed throughout its complete history (2002-2013). The data involve the bipartite association between all the software problems (or “bugs”) ever reported during the development of the project (15,556) and all the individual problem-solvers ever engaged in the development and maintenance of the software throughout its history (1,188). The sample space includes all the 68,398 problem solving actions that were actually observed during the sample period. F/OSS productions like the one selected for analysis are particularly useful for the purpose of this project because information on organizational problem solving may be accessed directly by mining on-line bug repositories, and because each individual action is associated with timing information that is complete and exact to the second. As a consequence the complete historical sequence of problem solving actions can be reconstructed in continuous time and analyzed via the relational event model for bipartite association that this project wants to establish.