Factor Score Estimation Under Model Misspecification: Network Score Using Hybrid Centrality

• In Event: Machine Learning

Abstract

Despite the significance of factor scores in assessment practice, its estimation is often ambiguous, facing even more challenges when model misspecifications exist. This study adopted hybrid centrality within a network analysis framework to estimate factor scores. A simulation study comparing the CFA scores (regression or Bartlett scoring method) with the network scores was conducted under both proper specification and misspecification conditions. For a correctly specified model, the network scores performed similarly to CFA scores in terms of RMSE and correlation of true scores and latent scores. Significantly, the network scores showed substantial robustness under misspecification conditions with cross-loadings compared to CFA using regression or Bartlett scoring. When model misspecifications are possible, the network approach is preferred over traditional estimates.

Authors

• Jinying Ouyang, Peking University
  Presenting Author

• Zhehan Jiang, Peking University
  Non-Presenting Author

• Yuting Han, Beijing Language and Culture University
  Non-Presenting Author

• Lingling Xu, Peking University
  Non-Presenting Author