Growth Modeling: Akaike and Bayesian Information Criteria and Sample Size Considerations

• In Event: 52.070 - Multilevel Modeling: Estimation, Model Selection, and Sample Size

Abstract

The decision between competing statistical models can be complex, so researchers often turn to fit statistics, including information criteria, to support theoretical considerations. In multilevel modeling, multiple sample sizes are available for the computation of criteria that incorporate sample size penalties, and previous literature has not consistently supported which sample size to use. Overall, this study examined the AIC, BIC, and SABIC for performance accuracy in the context of competing growth models, finding that the AIC performed best, followed by the SABIC, and then the BIC. Researchers also addressed which sample size to use in calculation of the BIC and SABIC. We preliminarily suggest that incorporation of the level-2 sample size might be more-accurate for true model selection.

Authors

• Sarah D. Newton, University of Connecticut

• D. Betsy Mccoach, University of Connecticut