Chapter 4: Power Analysis in Structural Equation Modeling

Within the structural equation modeling (SEM) literature, methodological examinations of what constitutes adequate sample size, and the determinants thereof, are plentiful (see, e.g., Gagné & Hancock, 2006; Jackson, 2003; MacCallum, Widaman, Zhang, & Hong, 1999; Marsh, Hau, Balla, & Grayson, 1998). Predictably, practitioners gravitate toward those sources providing evidence and/or recommendations supporting the acceptability of smaller sample sizes. Such recommendations, however, tend to be based upon issues such as model convergence and parameter bias; they do not typically address statistical power. Thus, while there might be relatively large models for which, say, n = 50 yields reliable rates of model convergence with reasonable parameter estimates, such a sample size might be entirely inadequate in terms of power for statistical tests of interest.