A mutilevel decomposition of school performance using robust nonparametric frontier techniques ∗

espanolEste documento de trabajo evalua el desempeno educativo desde una optica multinivel, utilizando fronteras parciales. Se consideran variables idiosincraticas a nivel de escuela, clase y alumno. El estudio se ha aplicado a estudiantes de cuarto ano de educacion basica en Chile. De acuerdo con los resultados, menos del 30% de la variabilidad del rendimiento academico de los estudiantes puede ser atribuido a sus escuelas. Se corrobora tambien que un modelo formado unicamente por variables a nivel de estudiante daria lugar a elevadas ineficiencias no atribuibles a la gestion de la escuela, sino mas bien a una inadecuada asignacion de recursos; esto es, si se ignoran las variables especificas de recursos asignados a las escuelas, su desempeno es infravalorado, pues las ineficiencias causadas por una asignacion de recursos suboptima o las dificultades debidas a un entorno socioeconomico desfavorable se atribuyen a una gestion deficiente del centro. EnglishThis working paper proposes a methodology for evaluating educational performance, from a multilevel perspective. We use partial frontier approaches to mitigate the influence of outliers and the curse of dimensionality. We consider idiosyncratic variables at the school, class, and student levels. Our model is applied to a sample of students in fourth year of primary school in Chile. Results are consistent with previous ones that found that less than 30% of the variance in students’ educational attainment was attributable to schools. Our application also corroborates that a model considering only student-level variables would yield high inefficiencies not attributable to the school management, but rather to inadequate resource-endowments. In other words, when disregarding specific variables concerning the resources allocated to schools, the performance of those schools is undervalued, largely because inefficiencies caused by suboptimal resource endowments or difficulties that arise from the socioeconomic environment are instead attributed to poor management