In addition to significance tests for coefficients in the models,effect sizes are often reported for multilevel models. As the predictors of interest in the models were at level 2 (the MOOSESTM codes for teacher talk), the appropriate effect size is a proportion reduction in variance at level 2 (Raudenbush & Bryk, 2002). This is calculated as the reduction in variance in the random intercepts realized by the addition of one or more level 2 predictors to the model. Unfortunately, likely as a result of the small sample size,these estimates were unstable, frequently yielding either negative or misleadingly large positive values. Therefore, we do not report these proportion reductions in variance values.