Statistical techniques that use Akaike’s information criterion ( \(\:AIC\) ) or the Bayesian information criterion ( \(\:BIC\) ) for model fit comparisons are oftentimes applied to incomplete datasets. Multiple imputation is a generally accepted method for handling missing data. The method creates several complete versions of the incomplete dataset. These datasets are all subjected to the same statistical analysis, resulting in several outcomes of the same analysis. Finally, these outcomes are combined into one pooled analysis. Currently no pooled estimators of the \(\:AIC\) and \(\:BIC\) for multiply imputed data exist, other than averaging the measures across imputed datasets. In the current study two different estimators of the \(\:AIC\) and \(\:BIC\) in multiply imputed are proposed, next to averaging. The estimators are studied in three simulation studies under a factor model, a multilevel model, and a latent class model. The results show that the new estimators select the correct model more frequently than averaging does.
