A Composite Estimator of Effective Teaching

States and districts are collecting multiple measures of teaching to evaluate teacher effectiveness, but there is limited information about how indicators can be combined to improve inferences about a teacher's impact on student achievement and about teaching. We derive a statistical model and estimate the parameters of an optimal combined measure of teacher effectiveness using data from the Measures of Effective Teaching (MET) project. We contrast the optimal composites to composites created using equal weighting of indicators and to weights based on existing state policies. Our explorations consider multiple scenarios for data collection to determine tradeoffs between collecting more data and combining multiple indicators to improve the accuracy of inferences. We find evidence that there is a common component of effective teaching shared by all indicators, but there are also substantial differences in the stable component across measurement modes and across some indicators within a mode. The implication from our model is that composites that place relatively equal weight on all indicators will tend to capture the component of effective teaching that is common across indicators. We also find that optimal weights strongly depend on the target criterion and the optimal predictor tends to put most of the weight on the indicator corresponding to the target criterion. Composites formed based on state policies are moderately to highly correlated with optimal predictor of teacher contributions to achievement on the state test. Due to the relatively high reliability of the indicators in the MET project dataset, there are small differences in composites created under different data collection scenarios.