Comparing t and Sandwich Robust Statistical Methods of Coping with Level-1 Outliers with Hierarchical Linear Models

• In Event: 29.036 - Marvelous Multilevel Modeling

Abstract

The hierarchical linear model (HLM) has been widely used in educational studies. Violations of model assumptions have non-ignorable impacts on the model. One issue is the sensitivity of HLM to outliers. Three simulation studies were performed to evaluate the sensitivity of HLM to outliers. The first one investigated the biases of parameter estimates with different numbers and types of outliers given different sample sizes. The second one examined the performance of the Huber Sandwich estimator in correction of standard errors. The third one employed a t-distribution to correct the biases of estimates. The preliminary results indicated that the biases of parameter estimates varied with different types and numbers of outliers. The Huber Sandwich estimator and t-distribution estimator are compared.

Authors

• Jue Wang, University of Science and Technology of China

• Zhenqiu Lu, University of Georgia