| In 1988, McArdle identified issues modeling multivariate growth using what he termed "second order latent growth curve" models. Specifically, he raised questions about which type of structural equation model to use for longitudinal data analysis. For example, "Should the growth model be fit to common factors extracted from the measured variables (CUFFS)?"; or "Should a growth model be fit to the measured variables and then the intercepts and slopes of the common factors be considered (FOCUS)?" Both SEMs can be fit to the same set of data, however, as pointed ou in McArdle (1988), these two alternative models not strictly nested and therefore statistical comparison becomes more difficult. Most recent works have utilized only the CUFFS model (Ferrer, Balluerka, & Widaman, 2008; Grimm, Pianta, & Konold, 2009; Hancock, Kuo, & Lawrence, 2001; Lui, & Flay, 2009), however, because of the close relationship of the FOUCS models to the newer dynamic multivariate modeling issues (e.g., McArdle, 2008), it is an option worth further consideration.;In the examples presented by McArdle (1988), data from the Weschler Intelligence Scale for Children (WISC) were used. The longitudinal WISC dataset was completed in 1965 by R. T. Osborne (Osborne & Suddick, 1972). Children were measured on 11 subscales of the Weschler Intelligence Scale for Children (WISC). The selected sample consists of 204 children who were repeatedly measured under standardized conditions at approximate ages of 6, 7, 9, and 11. In this research, the same data and models are used but more focus is placed on the appropriate choice of SEM. Classic and contemporary techniques for analyzing repeated measures date are compared. The focus was on the specification of second order LGMs and the advantages of these methods over the more classic MANOVA and first order LGM techniques. The key results and their implications are discussed. |
