Power struggles: Estimating sample size for multilevel relationships research

Conducting research on human relationships entails special challenges of design and analysis. Many important questions benefit from the study of dyads and families, and studies of relationships in natural settings often involve longitudinal and/or clustered designs. In turn, power analyses for such studies require additional considerations, because multilevel statistical models (or structural equation modeling equivalents) are often used to analyze relationships data. Power calculations in multilevel models involve the difficult task of specifying hypothesized values for a large number of parameters. Planning studies can also involve power trade-offs, including whether to prioritize the number of dyads sampled or the number of repeated measurements per dyad. Unfortunately, the relationships literature provides limited guidance on how to deal with these issues. In this article, we present a data simulation method for estimating power for commonly used relationships research designs. We also illustrate the method using two worked examples from relationships research