Observed characteristics and teacher quality: Impacts of sample selection on a value added model

We measure the impact of observed teacher characteristics on student math and reading proficiency using a rich dataset from Florida. We expand upon prior work by accounting directly for nonrandom attrition of teachers from the classroom in a sample selection framework. We find evidence that sample selection is present in the estimation of the influence of teacher characteristics, but that failure to account for it does not appear to substantially bias estimation. Further, our procedure produces some evidence that more effective teachers are more likely to exit the classroom.