Policy implications stemming from empirical analyses that evaluate national federal compensatory programs are clouded because negative findings could be due to either the nature of the educational process or the inadequacy of the data collection and analysis technique. This paper constructs a framework within which the separate effects of each are assessed. Two general sources of direct evidence are considered. The primary source is the result of multivariate analyses of national and state educational data such as found in Project Talent or the Equal Opportunity Survey. The secondary source is the results of analyses of national and state data on ESEA Title I project evaluations. It is found that standardized tests are useful for some purposes such as overall student comparisons, but their high interaction with student background factors blunt their usefulness in determining the most effective alternative school inputs, policies, and practices. Given the present output measures, it is concluded that variations in educational expenditures do in fact have little independent effect upon educational output. The main problems lie in the proper understanding of the educational process, the usefulness of the output measure, and the soundness of analysis techniques. A strong interdisciplinary attempt should be made to correct these problems. (Author/AM)