A correct method to calculate the IDD value-added indicator in the Brazillian higher education quality assurance framework

Abstract

This paper investigates the value-added indicator used in the Brazilian higher education quality assurance framework, the so-called IDD indicator for undergraduate programmes (“Indicator of the difference between observed and expected outcomes''). The two main claims are that since 2014 this indicator is calculated incorrectly and that this mistake has relevance for public policy. INEP, the educational statistical agency responsible for educational quality indicators in Brazil, incorrectly uses multilevel modeling in their value-added analysis. The IDD indicator is calculated by estimating a varying intercept linear mixed model, but instead of identifying the intercepts with the value added of courses, INEP uses the mean of the student residuals. That this was indeed the error made is shown by reproducing exactly INEP’s published values using the incorrect method with the microdata for the 2019 assessment cycle. I then compare these values to the ones obtained with the same model and same data, but using the correct value-added measure. A comparison of reliability estimates for both methods shows that this measure of internal consistency is indeed higher for the correct method. As an example of policy relevance, I calculate the number of courses that would change from “satisfactory” to “unsatisfactory” and vice-versa, using the usual criteria established by INEP, if the correct method is applied.