Estimating Statistical Power With Incomplete Data

Software developments increasingly facilitate inclusion of incomplete data, but relatively little research has examined the effects of incomplete data on statistical power. Seven steps needed to conduct power analyses with incomplete data for a variety of commonly tested hypotheses are illustrated, focusing on significance tests of individual parameters. The example extends a growth curve model simulation presented by Curran and Muthén (1999) to the incomplete data situation. How to estimate statistical power for a range of sample sizes from a single model, as well as how to calculate the sample size required to obtain a desired value of statistical power, is demonstrated. Effects of data being missing completely at random (MCAR) or missing at random (MAR) across a range from 0% (complete data) to 95% missing data are considered. SAS and LISREL syntax are provided in this paper with syntax for other software available from the authors.