Can We Count on AIC (Akaike Information Criterion), BIC (Bayesian Information Criterion), and Likelihood Ratio Test in Model Selection?

• In Event: Roundtable Session 117
  In Roundtable Session: Complexities Encountered When Modeling Multilevel Data

Abstract

In Multilevel modeling (MLM), likelihood ratio test (LRT), the Akaike Information Criterion (AIC; Akaike, 1974) and the Bayesian Information Criterion (BIC; Schwarz, 1978) statistics are often used to compare models. However, there have been long-standing arguments about their inconsistent performance in model selection. This study shows that the low effectiveness of LRT, AIC and BIC in model selection within the context of repeated measures is still true within the context of non-repeated measures, and their performance not only depends on level-1 and level-2 sample sizes, but also is related to the type of model under investigation.

Authors

• Meihua Qian, Indiana University

• Jonathan A. Plucker, University of Connecticut