Education-related changes are often argued as the main reasons for changes in earnings distribution. However, omitted variable and measurement error biases possibly affect econometric estimates of these effects. Brazil experienced a sharp fall of individual labour income inequality between 1996 and 2014. Coincidentally, in the Brazilian National Household Sample Survey ( PNAD) there are special supplements on family background in these two years that allow us to better address the role played by falling education returns. This paper takes advantage of this information to provide new estimates of the level and evolution of the returns to education in Brazil using variable premiums by education level, quantile regressions, and pseudo panels. Regarding measurement error, the empirical strategy is to make use of the information of who responded to the PNAD questionnaire but controlling for availability biases. We find evidence of attenuation bias which reduces mean returns from education between 14 and 31.5 per cent. On the other hand, omitting parents' education information also accounting for selectivity issues reduces the premium estimates by 24 per cent. Perhaps more importantly, the fall of education premium is heavily underestimated when we do not take family background into account. The highest fall of returns occurred in intermediary levels of education and income. Cohort effects also show that the reduction in the educational premium has been going on for several generations. Finally, we assess how parents' education affects the educational outcomes of their children and how the intergenerational mobility of education has evolved over the last years. We find a reduction on the intergenerational persistence of education from 0.7 to 0.47 between 1996 and 2014. Cohort effects regarding intergenerational mobility also show that the fall in the persistence of education is also stronger for younger cohorts, which coincides with the fall of education premiums.
Brazil earnings inequality schooling returns intergenerational mobility omitted variable bias