Impact of Post Hoc Measurement Model Overspecification on Structural Parameter Integrity

Abstract

In the common two-step structural equation modeling process, modifications are routinely made to the measurement portion of the model prior to assessing structural relations. The effect of such measurement model modifications on the structural parameter estimates, however, is not well known and is the subject of the current investigation. For a five-factor confirmatory factor model, Monte Carlo simulation was used to compare interfactor correlations before and after measurement model respecifications (crossloadings, intrafactor error covariances, and interfactor error covariances) under conditions varying number of indicators per factor, loading magnitude, and sample size. Although results indicate some effect on structural parameter estimates arising under conditions of overspecification in the measurement model, in general, the impact is seen to be quite negligible.