Analyses proceeded with model building. Multilevel model building employed a modified version of Hox’s (2002) group approach. Hox (2002) recommends beginning multilevel model development at the lowest level of analysis and then proceeding with the introduction of variables at higher levels. Following this procedure, one arrives at the optimal model if adding a variable adds little or no improvement between two consecutive models. One evaluates model improvement by examining the chi-square drop between two consecutive models and the probability associated with chi-square, and by comparing the models’ fit indices. The authors intended to begin multilevel model development at the classroom stage, but it was evident during the single-level model development stage that multicollinearity issues would arise if we promoted all classroom-level variables to the school level. Consequently, our multilevel model development began at the school level. The authors used the final school-level model obtained during single-level analyses (see Figure A1 in the Appendix on p. 322) as the base for the multilevel model. The initial model contained separate path coefficients for each level of analysis.