Application Of Multilevel Models For The Program Outcome Evaluation

Objectives: To provide a new hot area and methodological reference for the outcome evaluation of social program in the present study, multilevel models were applied to filter the influencing factors on the program outcome, explore the effect of compositional variables, and make the value added analysis of high-level unit effects.Methods: Based on the data of Gansu Basic Education Project and considering the hierarchical structure, one-variate multilevel models and multivariate multilevel models were applied to filter the influencing factors on the program outcome, measure compositional effects, and make the value added analysis of general high-level unit effects by residuals analysis. Combining the results from the models, the more credible evaluation of the program outcome was got.Results: (1)The results from one-variate and multivariate multilevel models were mainly consistent. (2)Compositional effects were measured. (3)By analyzing the residuals of high-level units, their general effects were got. (4)The program had some effects, however the effects of various areas were different.Conclusions: Multilevel models can offer more credible results than traditional regression models, because it considers the characteristic of hierarchically structured data. The setup and application of compositional variables can provide wide space for factors in high-level units. Analysis of high-level unit residuals can offer richer information, which cannot be gained by the fixed part and traditional regression models, so that it is a new evaluation thought with promising prospect.