Statistical Power in Structural Equation Modeling

Latent Growth Curve Models (LGCM) simultaneously structure changes in means, variances, and covariances. Using approximation methods, we investigated the power to detect covariance in change in a bivariate LGCM (Hertzog et al. 2006). Power was surprisingly low throughout for reliabilities below 0.9 and less than 4 to 5 measurement occasions. In another study (Hertzog et al., in press), we used Monte Carlo simulations to investigate the power for detecting slope variance using different statistics. Power again was low and depended on the specific significance test and the covariance between slope and intercept.