Rethinking the Universal Use Hierarchical Linear Modeling

• In Event: Robert L. Linn Distinguished Address Award

Abstract

Since the publication of Hierarchical Linear Models and its associated software nearly 35 years ago, the use of hierarchical linear models (HLM) or multilevel models (MLM) is universal in educational research owing to the clustering of students into schools or other educational units. It is almost mandatory such models be used. However, like any statistical model, the right model depends on purpose as well as data structure. This talk will question the indiscriminate use of HLMs and MLMs in educational research. I will review the advantages of such models over other methods that account for clustering and the limiting assumptions behind HLMs and MLMs and discuss examples where these models can lead to biased results. HLM or MLM is an essential tool for the toolkit, but not the only one.

Author

• Daniel F. McCaffrey, Educational Testing Service