Education Outcomes and Spending
By Arnold Kling
An OECD report on comparative performance of high school students in different countries is receiving a lot of attention.
Overall, wealthier countries tend to do better in educational terms than poor nations, but there are exceptions: Korea’s national income, for example, is 30 per cent below the OECD average but its students are among the best performers in OECD countries. Nor is high expenditure necessarily a key to success: a number of countries do well in terms of “value for money” in their education systems, including Australia, Belgium, Canada, the Czech Republic, Finland, Japan, Korea and the Netherlands, while some of the “big spenders” perform below the OECD average.
The United States is a prime example of a big spender with poor performance. Still, I think one should be careful about jumping to conclusions based on that. One of my pet peeves is that education research almost never measures “value added.” That is, it is rare to take two similar groups of students, assign them randomly to different educational processes, and observe the differences in outcomes. Instead, we measure differences in outcomes.
The reality is that statisticians and educators believe that outcomes are determined to a large extent by factors outside of the education system. Genetic endowments are known to play a large role. Parental income usually correlates with educational attainment, but I wonder if income is a separate causal factor or a proxy for genetic factors.
In Montgomery County, Maryland, where I live, school officials routinely excuse poor outcomes at some schools on the basis of the low incomes of students. This is a politically acceptable excuse, but it is really a disguised expression of genetic determinism.
It will be a great day when some of the billions of dollars that the government spends on education go into research that measures value added rather than tantalizing but unscientific data on outcomes.
For Discussion. Could value added be measured effectively using statistical methods, or are actual controlled experiments necessary?