Higher education value added using multiple outcomes

Milla, Joniada [UCL] San Martin, Ernesto [Pontificia Universidad Catolica de Chile, CORE (UCL) and Measurement Center MIDE, UC, Chile] Van Bellegem, Sébastien [UCL]

We build a multidimensional value added model to analyze jointly the test scores on several outcomes. Using a unique Colombian data set on higher education within a seemingly unrelated regression equations (SURE) framework we estimate school outcome specific value added indicators. These are used to measure the relative contribution of the school on a certain outcome, which may serve as an internal accountability measure. Apart from the evident estimation efficiency gains, a joint value added analysis is preferable to the unidimensional one. First, unless modeled in a multidimensional framework, the comparison of value added estimates for different outcomes within a school is not well defined; our model circumvents this issue. Second, even in the case of a separate major field of study analysis there still exists unobserved heterogeneity due to institutional diversity. This makes it more compelling to employ a rich set of outcomes in computing value added indicators. In the end, we aggregate the outcome-specific value added estimates to produce a composite value added index that reflects the combined value added contribution of all the subjects for each school.