Causal Structure Learning Using Composite Scores: The Case of Dichotomous Indicators

• In Event: AERA Graduate Student Research in Progress Roundtable Session One
  In Roundtable Session: Education Research Modeling and Measurement (Table 6)

Abstract

Establishing cause and effect between variables is of vital importance in the social sciences. A novel causal model selection method, Direction Dependence Analysis (DDA), utilizes higher moments of variables to examine cause-effect properties of the data generating mechanism. In the continuous data domain, one typically makes use of composite scores to approximate latent abilities, however, it is unclear how DDA is affected by utilizing composite scores. The present study aims to address this gap in causal structure learning research with a focus on dichotomous indicators. The simulation results suggest a dramatic loss in statistical power when applying DDA on raw or EAP scores instead of true scores. A larger number of items to measure the predictor, larger causal effects, and strongly skewed true scores can help to mitigate this power loss.

Author

• Xiuhan Chen, University of Missouri
  Presenting Author