Level-Two Design Effects for Data From the National Center for Educational Statistics

• In Event: Multilevel Modeling Methods and Applications II

Abstract

This research introduces example indices that should be considered when using two-level models with probability samples from NCES. Currently, researchers may ignore the sampling design beyond the levels that they model, which can result in incorrect inferences regarding hypotheses but the degree of bias depends on the informativeness of the stratification and clustering in the sampling design. Some software packages accommodate sampling design information for two-level models, but not all. Researchers using software that cannot accommodate the sampling design may wonder to what degree their estimates may be biased. For five example datasets, design effects of ignoring the sampling designs in unconditional multi-level models will be presented for specific variables. Equations for calculating these design effect indices will be provided.

Authors

• Laura M. Stapleton, University of Maryland

• Yoonjeong Kang, University of Maryland - College Park