Regression Analyses and Education Production Functions: Can They Be Trusted?

Abstract

Many studies have purported to demonstrate that schooling has little independent impact on achievement and that administrators can do little to boost students'test scores. Daniel F. Luecke and Noel F. McGinn question such results and use variations of a computer simulation model to generate data sets similar to those collected by educational researchers. They subject the data generated to several kinds of aggregation procedures and regression analysis, and compare the statistics thus yielded with their knowledge of the causal relationships programmed into the data. They conclude that many "no significant effect" findings may be artifacts of statistical techniques used to analyze cross-sectional survey data.