• Bipartite network analysis models associations between statue-stelae types and sites. • Typological co-occurrence exhibits stable, non-random relational configurations. • Monte Carlo and QAP/ERGM tests assess robustness under secondary deposition. • Network structure is driven by typological affiliation rather than spatial proximity. • A reproducible framework for relational analysis of prehistoric material culture. The Lunigiana statue-stelae represent one of the most substantial assemblages of European prehistoric art, both in terms of the density of finds and their symbolic significance. This study employs network analysis to examine the relationships between find locations and morphological typologies, modelled as a bipartite network. Centrality measures and community detection are used to describe structural properties of the resulting network, highlighting configurations of typological association and co-occurrence. Projection of the bipartite network onto locations enables the exploration of indirect links between sites based on shared typological categories. To address the potential effects of secondary deposition and to assess the stability of observed patterns, Monte Carlo simulations and QAP/ERGM-based approaches were applied as validation procedures. Network construction was carried out using the open-source software Gephi, while statistical analyses were performed in R, ensuring transparency, reproducibility, and methodological extensibility. Overall, the application of computational network methods allows the statue-stelae to be examined as elements of a formal relational model, rather than as direct proxies for prehistoric interaction, transmission, or social behaviour. Rather than proposing new typological or chronological interpretations, the study introduces a formal and reproducible framework to test whether established typological distinctions exhibit stable, non-random relational structures under explicit network constraints, even in the absence of primary depositional contexts and finer-grained attribute-level distinctions.
Nincheri et al. (Wed,) studied this question.