Obtaining sound intraclass correlation and variance estimates in three-level models: The role of sampling-strategies

Three-level clustered data commonly occur in social and behavioral research are prominently analyzed using multilevel modeling. The influence of the clustering on estimation results is assessed with intraclass correlation coefficients (ICCs), which indicate fraction variance outcome located at each higher level. However, ICCs prone to bias due high requirements regarding overall sample size In Monte Carlo simulations, we investigate how these characteristics statistical power components robust ML-estimation. Results reveal considerable underestimation Level-3 importance combination ICC sizes. Based our results, derive concise sampling recommendations discuss limits inferences.