台大心理系

Structural equation modeling (SEM) is a commonly used multivariate statistical method in psychological studies. Under SEM framework, researchers can flexibly specify their models based on available psychological theories. In this dissertation, a penalized likelihood (PL) method for SEM was proposed. Compared to the usual likelihood, PL includes an additional penalty term to control the complexity of the hypothesized model. When the penalty level is chosen appropriately, PL can yield a model that balances model goodness-of-fit and model complexity. The proposed method is especially useful when limited substantive knowledge is available for model specification. An expectation-conditional maximization (ECM) algorithm was developed to maximize the PL estimation criterion with several state-of-art penalty functions. Four theorems on the asymptotic behaviors of PL were derived, including the local/global oracle property of PL estimators and the selection consistency of Akaike/Bayesian information criterion (AIC/BIC). Two simulations were conducted to evaluate the empirical performance of the proposed PL method. The practical utility of PL was demonstrated through two real data examples.