Evaluation of approaches for blockmodeling symmetric dynamic networks with incomers and outgoersCugmas, Marjan (Avtor) Žiberna, Aleš (Avtor) dynamic networksblockmodelingmonte carlo simulationlocal mechanismsevaluationThe paper builds upon the 2023 study by Cugmas and ˇZiberna that empirically evaluated blockmodeling approaches for dynamic networks, where the dynamic stochastic blockmodeling approach (Matias and Miele, 2017) and stochastic blockmodeling for multipartite networks (Bar-Hen et al., 2022) were recommended. This study delves deeper into evaluating blockmodeling approaches by focusing on symmetric networks and considering incomers and outgoers. The findings indicate that several factors affect the outcomes of blockmodeling approaches. Dynamic blockmodeling generally outperforms the blockmodeling of each time point separately, especially when the partitions remain relatively stable over time. In terms of specific approaches, the dynamic stochastic blockmodel is recommended when the blockmodel type does not change in time. However, if the blockmodel type changes, either stochastic blockmodeling for multilevel networks (Chabert-Liddell 2022) (in the absence of incomers and outgoers) or stochastic blockmodeling for multipartite networks (with incomers and outgoers) should be considered. For smaller networks with unstable partitions and random links within the blocks, the k-means blockmodeling for linked networks (ˇZiberna, 2020) is recommended. Researchers should select approaches based on in-depth knowledge of the networks in question. They should also consider both the partitions derived from blockmodeling each time point separately and the default partitions as the initial partitions in dynamic blockmodeling to achieve better results.20262026-09-16 14:27:14Članek v reviji187940sl