Abstract
A latent variable modeling approach for examining population similarities and differences in observed variable relationship and mean indexes in incomplete data sets is discussed. The method is based on the full information maximum likelihood procedure of model fitting and parameter estimation. The procedure can be employed to test group identities in mean and mean contrasts, variances, covariances and correlations of studied variables. The outlined approach can also be used to address concerns of meta-analysis for correlations across multiple studies.
Notes
1This article assumes throughout that studied subjects provide unrelated data (i.e., there is no nesting of subjects in higher order units and thus there is no data hierarchy; Raudenbush & Bryk, 2002).