The aim of this article is to conduct a comparative methodological essay of two estimators of age-period-cohort models: the conventional estimator obtained by the generalized linear restricted models (MLGR) and the so-called intrinsic estimator (EI). The objects are contributions of age, period and cohort effects to temporal changes in the progression probability to the 5th grade of elementary school for Brazilian women. The APC modeling is justified because age, period and cohort effects may significantly affect the probability of grade progression: age effects reflect both the minimal age of school entry and the trade-off between study and work; period effects are associated with different economical and political conjunctures, as well as with current educational policies; finally, cohort effects reflect social attributes unique to a group. Both are juxtaposed in terms of the efficiency, significance and parameter estimates. The results reveal the potentiality of the solution to the age-period-cohort model based on intrinsic estimator, which presents parameters with smaller variance than those estimated from generalized linear restricted models. Therefore, projections of grade progression probabilities based on the extrapolation of standard errors of the intrinsic estimator can be promising.
Age-period-cohort models; Intrinsic estimator; Grade progression probability
