Latent Growth Curve Analysis with Item Response Data: Model Specification, Estimation, and Panel Attrition
Yang, Ji Seung
MetadataShow full item record
Measuring change in a construct over time in educational or psychological research is often achieved by administering the same items to the same respondents repeatedly over time. When item response data are categorical, a second-order latent growth model (LGM) can be used by incorporating an item response theory (IRT) model as the measurement model (referred to as LGM-IRT). Common item effects can be specified as orthogonal specific factors in the measurement model. This study investigated three issues in using LGM-IRT with common item effects, namely model parameterization, estimation of model parameters, and sample attrition. Selected longitudinal IRT models were first reviewed. The Schmid-Leiman transformation was used to transform the second-order model to first-order formulation so that the model could be estimated in common multidimensional IRT software packages. Simulation studies were carried out to examine different methods of estimating the model, namely using different estimation methods (diagonally weighted least square estimator, Monte Carlo expectation-maximization algorithm and Metropolis-Hastings Robbins-Monro algorithm) and using reduced models. The estimation methods were examined under different test lengths, sample sizes, and panel attrition mechanisms. The reduced models were examined under complete data situation. One empirical analysis was conducted to compare and contrast the different methods using data from the “Multistate Study of Pre-Kindergarten 2001–2003” by the National Center for Early Development and Learning. The results of this research can provide provide guidelines on the utility of the model using aforementioned three estimation methods and the reduced models. The research combines modeling techniques of structural equation modeling and IRT and can make contribution to the literature of this unified general framework.