An investigation of fit measures for longitudinal network and attribute co-evolution models

co-evolution network models

Dohleman, Bethany S.

关键词: co-evolution network models;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/18645/Dohleman_Bethany.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Modeling the processes underlying social network and attribute change allows researchers to better identify and understand dependencies present amongactors — people, places, or things. The connections that exist among these actors change over time, depend on the presence or absence of other connections,and depend on the characteristics of the actors present. Advanced modeling techniques have been proposed that are designed to capture changes in structureand actor attributes.Fit measures have been developed for actor-based models of network structural evolution (Schweinberger, 2007; Snijders, 1996). Snijders et al. (2006) extended the actor-based structural evolution model to an actor-based co-evolutionarymodel that includes actor attributes. Despite recent methodological advancesin the estimation of co-evolution models, measures have neither been developed nor evaluated to assess how the inclusion of actor attributes contributes to themodel.Four measures are developed in this paper, including an extension of Snijders’ t-test to a Pseudo-Wald test statistic, an extension of Schweinberger’s score testfor use in a co-evolution model, an entropy-based goodness-of-fit measure, and an R^2 goodness-of-fit measure for actor-based co-evolution models.Following the theoretical development of each measure, the behavior and performance of each are compared in a large simulation study. Results from this simulation show that each proposed measure displays expected behavior,especially as the number of actors increases, and shows promise for future use as fit measures in co-evolution models.