The Effect of Estimation Selection in Multilevel Modeling With Small Sample Sizes

• In Event: Multilevel Modeling: Estimation, Model Selection, and Sample Size

Abstract

This simulation study examines how selection of estimator affects parameter estimates with a small number of groups in multilevel modeling. Three estimation procedures were used to test the accuracy of multilevel model estimation with a variety of manipulated conditions: iterated generalized least squares (IGLS), restricted iterative generalized least Square (RIGLS), and Markov chain Monte Carlo (MCMC). Relative biases of parameter estimates were found to be sensitive to the relatively smaller number of level two units across the three estimation methods. While negative relative biases were found in both IGLS and RIGLS estimations, positive relative bias was detected in MCMC. In general, use of RIGLS estimation resulted in less bias across all conditions as compared to IGLS estimation.

Authors

• Hyewon Chung, Chungnam National University

• Jiseon Kim, University of Washington - Seattle

• Ryoungsun Park, Wayne State University