Comparison of Statistical Power From Multivariate and Univariate Latent Growth Models With Missing Data

• In Event: Marriott Roundtable Session 73
  In Roundtable Session: Precision, Power, and Small Sample Size Issues in Growth Curve Modeling Variations

Abstract

A multiple domain latent growth model (MDLGM), which incorporates associations among several latent growth trajectories simultaneously, is conceived as a powerful multivariate analysis technique compared to univariate latent growth models (LGMs). The MDLGM would be even more beneficial with the presence of missing data. However, no study examined the effects of missing data on statistical power discrepancy between the MDLGM and a set of LGMs. Hence, this study compared the power of the MDLGM to that of a set of LGMs for detecting group differences in growth rates when there is missing data. The results from this study indicated that the MDLGM was more powerful than a set of LGMs with relatively low inter-correlation between domains with missing data.

Authors

• Kejin Lee, University of Nebraska - Lincoln

• Tiffany Ann Whittaker, The University of Texas - Austin