Scientific fragmentation, i.e., the division of research fields into increasingly isolated silos of knowledge, has traditionally been viewed as problematic. Although it has been raised that fragmentation can also be beneficial, criteria for distinguishing favorable from detrimental fragmentation remain elusive. Leveraging citation network analysis, we present a proof-of-concept study investigating whether patterns of field self-referentiality can serve as empirical indicators differentiating these fragmentation types. Analyzing citation patterns of 7,146 publications spanning 50 years (1974-2023) from translational psychological treatment research, we examined how network density and modularity effects on scientific outcomes (i.e., productivity, collaboration, and impact) change over time. Generalized Additive Models of time-varying relationships revealed a phase transition around 2000 where network effects on outcomes reversed from negative to positive. Critically, this reversal occurred at comparable density levels across time periods, demonstrating that temporal context rather than absolute network structure values drives the observed effects. These findings suggest that fragmentation effects are better understood as context-dependent: early fragmentation enabled productive exploration, while integration became beneficial once the field developed an internal knowledge core. Self-referentiality patterns — the proportion of citations to highly influential internal work, emerged as empirical indicators tracking this transition, independently of the network structure measures. We selected translational psychological treatment research as our use case because domain expertise enables validation of the detected transitions against independently documented field developments. While based on a single field, the analytical framework is designed for replication across diverse scientific domains.
Petrule et al. (2026) studied this question.