|Day & Time||2nd December, 2016 15:00-|
|Venue||Room C816, Building of the Faculty of Science Graduate school of Science|
|Lecturer||Prof. Kazuhiko Hayakawa (Hiroshima University)|
|Title||Corrected Goodness of Fit Test in Covariance Structure|
|Abstract||Many previous studies report a simulation evidence that the goodness of fit test in covariance structure analysis or structural equation modeling suffers from over-rejection problem when the dimension of manifest variables is not small compared with the sample size.
In this paper, we first investigate the reason and show that the test statistic involves a bias term which becomes large as the number of manifest variables grows. Then, we propose a test statistic by subtracting that bias term. Incidentally, it is demonstrated that the corrected test coincides with one of the tests considered in Amemiya and Anderson(1990). We also propose a simple modification of Sattora and Bentler’s mean and variance adjusted test. Monte Carlo simulation is carried out to investigate the performance of the corrected test in the context of confirmatory factor model and cross-lagged panel (panel vector autoregressive) models. Simulation results reveal that the corrected test overcomes the over-rejection problem and outperforms existing tests in almost all cases considered.