Collaborative structures and research productivity: an empirical analysis with Bayesian networks

Abstract Collaboration and research productivity are widely recognized as interconnected, yet the directionality of this relationship remains debated. This study examines the interplay between collaborative structures and research productivity in two scientific domains: Life Sciences (LS) and Physics and Engineering (PE), using a Bayesian network approach. By analyzing both publication-based and project-based collaborations, we observe distinct domain-specific mechanisms. In LS, the relationship between collaboration and research productivity is more variable: Degree Centrality in publications shows the strongest association with performance, but in some years Closeness Centrality mediates this link by capturing how proximity to knowledge flows temporarily amplifies the effect of direct collaborative ties. By contrast, in PE, Degree Centrality consistently emerges as the strongest and most stable driver of scientific output. The what-if analysis further shows that in PE, higher productivity increases the ability of universities to attract firm partners in EU-funded projects, whereas in LS it primarily reinforces academic collaborations. These findings indicate that collaboration within publication networks tends to play a leading role in shaping research productivity, while research productivity appears to have a stronger influence on collaboration patterns within project-based networks. This mutual reinforcement is shaped by disciplinary and temporal factors, highlighting the importance of tailoring collaboration strategies to specific scientific domains.