The Influence of Number of Clusters and Level-2 Residual Distribution on Multilevel Parameter Estimates: A Latin Hypercube Design Perspective

Abstract

Monte Carlo methods with Latin Hypercube Design (LHD) were used to explore the impact of increasing numbers of clusters considering four cluster residual distributions in a two-level model. The implementation of LHD is illustrated and the results update existing guidelines for the number of clusters needed to obtain accurate estimates.