Network Analysis for Factor Score Estimation: Supervised and Unsupervised Scenario

• In Event: Modern Psychometric Methodologies and Applications

Abstract

Network analysis offers an alternative approach to estimating latent scores by analyzing the local characteristics of the elements in a network. This study investigates the use of network scores, which employ centrality measures from network analysis, as a promising way to mitigate the indeterminacy issue associated with factor scores under both supervised and unsupervised scenario. The results show that, when theories are provided, the network scores show substantial robustness under misspecification conditions compared to four traditional CFA scoring methods. In the second simulation, network scores also perform better than EFA score when structure knowledge is unable to obtain. Overall, this study suggests that the network approach should be favored over conventional estimates in many circumstances.

Authors

• Jinying Ouyang, Peking University

• Zhehan Jiang, Peking University

• Lingling Xu, Hangzhou Normal University

• Fen Cai, Peking University

• Haiying Liang, University College London