An experimental evaluation of three teacher quality measures: Value-added, classroom observations, and student surveys

•Using a random assignment experiment, we evaluate the predictive validity of value-added, classroom observations, and student surveys.•Like prior research, we find that value-added measures are unbiased predictors of teacher performance.•Classroom observation scores are predictive of teachers’ performance after random assignment while student surveys are not.•These results provide support to teacher evaluation systems that use value-added and classroom observations. Nearly every state evaluates teacher performance using multiple measures, but evidence has largely shown that only one such measure—teachers’ effects on student achievement (i.e., value-added)—captures teachers’ causal effects. We conducted a random assignment experiment in 66 fourth- and fifth-grade mathematics classrooms to evaluate the predictive validity of three measures of teacher performance: value-added, classroom observations, and student surveys. Combining our results with those from two previous random assignment experiments, we provide additional experimental evidence that value-added measures are unbiased predictors of teacher performance. Though results for the other two measures are less precise, we find that classroom observation scores are predictive of teachers’ performance after random assignment while student surveys are not. These results thus lend support to teacher evaluation systems that use value-added and classroom observations, but suggest practitioners should proceed with caution when considering student survey measures for teacher evaluation.