Comparing Metaheuristic Algorithms for Structural Equation Modeling Sensitivity Analysis

• In Event: AERA Research in Progress Roundtable Session Two
  In Invited Roundtable: Research-in-Progress RT002 (Thursday)

Abstract

Structural equation modeling (SEM) is widely used to evaluate complex theories, yet conclusions may be vulnerable to omitted confounders. We examine sensitivity‐analysis methods that search for a minimal set of relations between a phantom confounder and model variables that would alter study conclusions. Building on the SEMsens R package, we compare three metaheuristic algorithms—ant colony optimization (ACO), Tabu search (TS), and simulated annealing (SA)—using a Monte Carlo design informed by ten published educational SEM studies. We vary model size and define objective functions based on p-values, path coefficients, and global fit (RMSEA). We anticipate algorithm-by-model interactions in convergence accuracy and runtime, and provide practical guidance for applied researchers conducting sensitivity analyses in SEM.

Authors

• Xinyu Liu, University of Minnesota

• Katerina M. Marcoulides, University of Minnesota