SOCNET is a service of INSNA, the professional association for social
network researchers (http://www.insna.org). To unsubscribe, send
an email message to
As a followup to the session at Sunbelt XXXIX in June, we are organizing a second session on Inference and Generalisability in Modelling Samples of Networks and Multi-Level Network Data at the EUSN2019 conference, held in Zurich, September 9-12.
The goal of this session is to bring attention to methodological issues related to generalizability or inference to populations of networks and to
propose methods and diagnostics for joint estimation of models
multiple networks or for networks with multi-level structure. A further
motivation is given below as a well as a number of relevant questions.
contributions on any of these, or related questions, or
applications in which generalizability or inference to populations of
networks play a role.
Please do no hesitate to contact us with further questions.
Marijtje van Duijn and Pavel Krivitsky
Sociometric data that we
collect are increasingly rich, and we increasingly analyse not single
networks but ensembles of networks. Data using the same name generator
on disjoint sets of actors in disjoint but similar settings have been
collected about classrooms, schools, households, firms, legislative
bodies, and other such replicable scenarios.
data, we often wish to pool the information from these multiple
networks, and to draw conclusions generalisable to a broader population
of networks in those settings. Methods to do so range from post-hoc
meta-analyses to full hierarchical multi-level models.
joint analyses raise a number of methodological questions, however. Some
of them are questions that are asked in any situation that involves
sampling from a population:
* What does it mean to draw a representative sample of networks?
* Can networks selected using different procedures be analysed together, and how?
* What "population" quantities are actually being estimated when metanalyses are performed or multilevel models fit?
Others are specific to social networks:
* Can the same model be fit to all of the networks in the ensemble?
* How can parameter estimates from networks that vary in size and/or composition be compared?