Simultaneous Statistical Inference in Evaluating Teacher Performance

There is widespread interest in using student achievement data to evaluate the performance of individual schools for accountability. Such evaluations are the capstone of many federal and state education policies. Currently there is also increasing interest in using sophisticated value-added models (VAM) as a part of the evaluations for individual teachers as well as schools, and using the measures for a variety of purposes including merit pay. Evaluation systems based on value-added measures typically involve classifying teachers or schools into several groups according to their estimated performance. These classifications can involve simultaneous decisions for hundreds or even thousands of teachers/schools. Although using standard statistical procedures can control the probability of misclassification for each individual, of concern is the simultaneous error rate of the whole evaluation system across all involved individuals. At the system level, the entire collection of classifications determines whether scarce resources are well allocated or whether the data is useful for other types of decision making. In statistical terminology, the ensemble of classification decisions is referred to as the problem of simultaneous inference. The purpose of this project is to establish a methodological framework for controlling simultaneous errors in classification of teachers/schools on the basis of student achievement and value-added models.