EDUCATIONAL OUTCOMES IN OECD COUNTRIES
34
years) of less than $20,000, there is no association between educational spending and
educational outcomes across OECD countries.25
Table 12 probes the cross-country expenditure-outcome relationship in education in
greater detail. Columns (1) to (6) reveal that in all three cycles of the OECD PISA
student achievement test - 2000, 2003, and 2006 - there is a very small positive
bivariate association between expenditure per student and educational achievement that
vanishes once countries below a very basic cumulative spending level of $25,000 are
excluded. In fact, in the latter samples, the adjusted R2 of a regression of test scores on
educational expenditure per student is actually negative in all three PISA cycles. Of
course, many more factors such as students’ family backgrounds enter the determination
of educational outcomes.
The simplest way of addressing bias from unobserved time-invariant country factors is
to ignore level differences and restrict the analysis to changes in expenditure and
outcomes over time. Thus, column (7) reports a regression in first differences between
2006 and 2000, and column (8) reports a fixed-effects regression that pools all three
PISA cycles. In both cases, the association between expenditure per student and
educational outcomes is far from being statistically significant at conventional levels,
and the point estimates are actually negative. These descriptive patterns suggest that
additional resources are not related systematically to improved test scores.26
For policy deliberations, information on the impact of resources from within individual
countries is perhaps more appropriate than cross-country information. Researchers have
studied the determinants of student achievement for more than 40 years. The work was
begun in the United States in the “Coleman Report” (Coleman et al. (1966)), which
introduced the idea of using statistical analysis to relate various inputs of schools to
student outcomes. This work also underscored the importance of including non-school
factors by demonstrating that family differences were very important in explaining
variations in achievement across students. While this original study has been subjected
to considerable criticism, it led to an extensive line of research.
The general picture about school resources in developed countries is now well known
and has been reviewed in a variety of places (see Woessmann (2005a) for Europe and
Hanushek (2002, 2003) for the United States). The available studies concentrate on
various common inputs to schools such as teacher experience or class size. These
factors are both readily available in both administrative and survey data sets and
frequently identified as the focus of policy. The available econometric evidence now
includes literally hundreds of separate estimates within the U.S. and other developed
countries. Quite uniformly, however, there is little strong evidence that any of the
25 With the two outliers, there is a weak positive association as long as other effects are ignored. Taken literally, the gray
regression line that includes Mexico and Turkey depicts an association where a doubling of expenditure in these two countries
is associated with one tenth of a standard deviation in test scores. Note that the test score-growth nexus reported above was
robust to dropping these two countries from the growth analysis.
26 These results also address the concern with the growth analyses reported above that there may be a simple simultaneous
determination of schooling investments and economic growth. Specifically, if a nation becomes richer through growth, it
might put more resources into its schools so as to increase its cognitive skills. However, the current results show that such
expenditure increases are not associated with educational outcomes in the first place. Consequently, additional resources in the
school system are unlikely to generate important reverse causality from growth to test scores.