Per-student expenditures are an extremely poor measure of education resources applied to students' learning. How money is spent is far more important than how much is spent. Moreover, the accounting of expenditures has become so complicated in recent years that it is difficult to focus on its classroom applications. The many services provided by schools in response to the demands of a changing society require increasing percentages of school budgets without any concomitant return in student achievement. In particular, a large percentage of the increases in real expenditures over the past 20 years has gone into special education and much of that into special procedures for health and other related services that special education students are entitled to by law (Chaikind et al., 1993). Often, the more difficult the teaching conditions, the larger the expenditures; in many production function analyses this may have led to the erroneous inference that expenditures have a negligible or even a negative effect on achievement. 7
In our view the independent variables presently in use are deficient. Instead of abandoning production function work altogether, however, we need to look at schooling inputs differently. In particular, we need to think about how to measure variation in inputs that we expect from other research to have some plausible relationship to student performance.
On the other side of the equation, the dependent variable in most production function research is also problematic. The measure most commonly used is a score on a national standardized exam. These tests are purchased from private publishers who devise them to be as widely applicable as possible so as to sell the most copies. The generic quality of the tests tends to make them insensitive to variations in the quality of any particular curriculum used by a school or school system. In other words, the most common dependent measure in education production function research is largely independent of the instructional and curriculum content and quality in any particular school, district, or state. Tests that are developed to suit many different curricula and instructional approaches will be inappropriate for any single approach. In the language of systemic school reform, if the dependent variable is not aligned with the teaching and learning going on in the school, it is no wonder that it does not pick up variations in school resources.
Yet the standardized norm-referenced tests have to be sensitive to variation in student performance, for they are required to be psychometrically reliable instruments. What variation do they pick up if they do not systemically assess variation in curricula or teaching quality or engagement of students? There is a clue in many studies (Coleman et al., 1966). These studies show far larger variations in achievement within classrooms and schools than across schools,