The effect of school spending on student achievement: Addressing biases in value-added models

The estimation of education production models used to evaluate the effect of school inputs and past skills on test scores, often called value-added models, can be biased by three main econometric issues: unobserved child characteristics, unobserved family and school characteristics and measurement error. We propose a two-step estimation technique which exploits the availability of test scores across time, subjects, families and schools in a unique administrative data set for England to correct for these potential biases. Our empirical results suggest that omitting school characteristics biases the estimation of the effect of school expenditure, whereas omitting unobserved child endowment biases the estimation of the effect of past skills but not the effect of school expenditure.