After estimation, using criteria listed in Table 4, researchers evaluated fit indices to assess the model’s adequacy. When the authors judged a model as adequate, they introduced the best predictor of TDS at the next hierarchical level. If the model did not fit the data adequately, modification indices and theory guided the addition and removal of paths or variables. The researchers evaluated modification indices first for the classroom level of analysis, followed by the school level, and finally by the district level. As suggested by Rowe and Hill (1998) and by Hox (2002), multilevel modeling stopped when adding a new variable did not improve model fit (i.e., if the chi-square of the model with the new predictor variable was larger than the model without the variable).