Endogeneidade da educação naprevisão da taxa de retorno: avaliação metodológica e aplicação para regiões brasileiras e estados selecionados

Abstract

In view of the endogeneity problem in the education variable and the available data sources for Brazil, two estimation methods for the mincerian earning equation are compared to infer about which one has the best predictive power – endogenous model or the procedure with instrumental variables (IV). It is found that the method of Heckman applied to the first model has the lowest mean squared error (MSE). This happens due to the quality of IV, which alert to the need for greater interaction between the agencies responsible for the design and data collection and the scientific community about the information to be generated in a survey to meet local political actions with diagnoses more accurate. Furthermore, given better predictive power of the endogenous model, alternative specifications in the earning equations are performed in order to allowing average returns to be variable as human capital and regionalization interact. It is observed that the education rates of return of education vary according to regions and Brazilian States, where the Northeast region and the State of Ceará present the lowest returns, equivalent to approximately 14.68%, which compared to the Mid-West region and the State of Santa Catarina the differential between the returns reaches 26% and 31%, respectively. This corrects part of the skill bias, and shows that individuals from regions with distinct degree of economic development do not accumulate equally the same level of human capital.