Measuring Human Capital and Teacher Value-Added with Transcript Grades: A Factor Model in Network Data (JMP)
Download JMPAbstract: This paper develops an approach to recover human capital from course grades while allowing for selection into courses and flexible grading heterogeneity. I model course grades using a factor model and recover the relative grading standards by relying on students who have overlapping course enrollments. These students induce a network structure, and I characterize sufficient conditions for identification in terms of the connectivity of the network using graph theory. I use this result to estimate teacher value-added on student human capital for a novel set of teachers–not just those whose students take standardized tests. I then calculate the gap between the value-added of tested and non-tested math teachers at the school-year level. A larger gap is associated with higher levels of economically disadvantaged and non-white students, indicating that the socioeconomic and racial gap in teacher value-added is understated by considering only tested teachers. Higher proficiency rates on Math standardized tests is negatively associated with the gap between teachers in tested and non-tested subjects, but this association flips upon controlling for school fixed effects. Consequently, while schools with higher math proficiency rates tend to allocate their higher value-add teachers to non-tested subjects, a within-school increase in Math proficiency positively predicts a greater within-school allocation of high quality teachers to tested subjects.