Two-thirds of education production function studies relating learning to sc
hool and parental inputs also include parental income. This confounds deman
d and production functions, since in demand functions income determines the
school inputs used in the production function. In a comprehensive review o
f the literature, we show that with this misspecification significantly pos
itive school input coefficients are 39% less common. Then, with Project TAL
ENT student-level data from 1960 and pooled state data for 1987-1992, we ex
amine the impact of including income with no other change in specification.
This causes most school inputs to become less significant. Hausman tests s
uggest that in OLS regressions there is a correlation between the school in
put measures and the error term, perhaps due to the omission of a good meas
ure of parental time with the student. This appears to bias the school inpu
t coefficients toward zero, but can be corrected with IV methods. [JEL 121,
H52, H42] (C) 1999 Elsevier Science Ltd. All rights reserved.