ABSTRACT
This chapter presents and compares the use of model-based procedures for analyses of incomplete longitudinal data. It illustrates model-based procedures for analyses of incomplete longitudinal data on adolescent alcohol consumption. The chapter demonstrates how the chi-square test statistic and model fit indices are calculated from the model log likelihood values in the presence of missing data. Multiple imputation is quite different from the Full Information Maximum Likelihood approaches outlined in previous sections because the missing data are handled in a step that is entirely separate from the main analysis. Although numerous approaches to handling missing data are available, the chapter demonstrated and compared the utility of different model-based procedures for analyses of incomplete longitudinal data within a latent growth modeling framework. From a statistical point of view, the best missing data procedures do several things. Although the missing data procedures presented are, in general, statistically sound, they weigh convenience and availability over statistical precision and efficiency.
