Saturday, February 22, 2003

Even More Bell Curve

From "Lessons from the Bell Curve", by James Heckman, in the Journal of Political Economy, October 1995.

Heckman is no left-wing hack. He gives the authors too much credit for their purity of intent and ignores the not so subtle racism that Digby discusses below, but here is what he concludes:
The Book fails for five main reasons.

1. The central premise of this book is the empirically incorrect claim that a single factor - g or IQ - that explains linear correlations among test scores is primarily responsible for differences in individual performance in society at large. Below I demonstrate that a single factor can always be constructed that "explains" all correlations in responses to a test or correlations in scores across a battery of tests, but in general this g is not constructed by conventional linear methods. There is much evidence that more than one factor -- as conventionally measured -- is required to explain conventional correlation matrices among test scores. Hernstein and Murray's measure of IQ is not the same as the g that can be extracted from test scores available in their data set. They do not emphasize how little of the variation in social outcomes is explained by AFQT or g. There is considerable room for factors other than their measure of ability to explain wages and other social outcomes.


2. In their empirical work, the authors assume that AFQT is a measure of immutable native intelligence. In fact, AFQT is an achievement test that can be manipulated by educational interventions. Achievement tests embody environmental influences: AFQT scores rise with age and parental socioeconomic status. A person's AFQT score is not an immutable characteristic beyond environmental manipulation.


3. The authors do not perform the cost-benefit analyses needed to evaluate alternative social policies for raising labor market and social skills. Their implicit assumption of an immutable g that is all-powerful in determining social outcomes leads them to disregard a lot of evidence that a variety of relevant labor market and social skills can be improved, even though efforts to boost IQ substantially are notoriously unsuccessful.


4. The authors present no new evidence on the heritability of IQ or other socially productive characteristics. Instead, they demonstrate that IQ is more predictive of differences in social performance than a crude measure of parental environmental influences. This comparison is misleading. It fails to recognize the crudity of their environmental measures and the environmental component that is built into their measure of IQ, which biases the evidence in favor of their position. Moreover, the comparison as they present it is intrinsically meaningless.


5. Finally, the authors' forecast of social trends is pure speculation that does not flow from the analysis presented in their book. Most of the social policy recommendations have an ad hoc flavor to them and do not depend on the analysis that precedes them. The appeal to Murray's version of communitarianism as a solution to the emerging problem of inequality among persons is a deus ex machina flight of fancy that is not credibly justified.