Can flexible non-linear modeling tell us anything new about educational productivity?

The objective of this study is to test, under relatively simple circumstances, whether flexible non-linear models including neural networks and genetic algorithms - can reveal otherwise unexpected patterns of relationship in typical school productivity data. Further, it is my objective to identify useful methods by which "questions raised" by flexible modeling can be explored with respect to our theoretical understandings of educational productivity. This study applies three types of algorithm - Backpropagation, Generalized Regression Neural Networks (GRNN) and Group Method of Data Handling (GMDH) - alongside linear regression modeling to school-level data on 183 elementary schools. The study finds that flexible modeling does raise unique questions in the form of identifiable non-linear relationships that go otherwise unnoticed when applying conventional methods. (C) 2001 Elsevier Science Ltd. All rights reserved. JEL classification: I21.