For Whom the Bell Curve Tolls

I was catching up on Dr. Nick's blog and ran across a posting from TJIC that he links to about the subject of race and IQ. I've long known that TJIC's opinions were pretty far to the right on this subject, but wow. Read the post here.

To be honest I was pretty shocked.

We all have a tendency to read books and articles which confirm our inner biases and self-image, espeically when that self-image sets us apart from the rest of the population. That's understandable. OTOH, this is a question of science, and science, when it's done correctly, is a model of an objective reality. After the inital shock of reading TJIC's post I had to think through the question, "Shocking, but is it true?" I had read the Bell Curve when it came out and as some of you know, I spent a lot of time in college training as a teacher to work with gifted kids. I ultimately chose not to follow that career, but it involved a lot of training around human intelligence, it's measurement and it's manifestation. Also, as some readers know, I had a full bore neuropsych evaluation last year, so my own inteligence has been measure as best as can be done with current technology and some of the results surprised me (and the testers). Given this, after a few hours of thinking, I came to the conclusion that the Bell Curve, even after almost 15 years, is still a bad piece of science. Here's why:

There are a bunch of cumulative assumptions the authors of the Bell Curve (Herrnstein and Murray) make which lead to their results, any one of which if not true results collapses their hypothesis:

Intelligence may be depicted as a single number. This is based on Spearman's hypothesis that "g", a measure of general intelligence that synthesizes multiple aspects of exists and is dominant. While a useful concept, psychometrics moved way beyond the concept of g in the 80s (early critiques of this date back to the '30s). Howard Gardner and Bob Sternberg both have proposed different, measurable models of multiple intelligences (Sternbergs theory is actually very useful from a practical point of view). The use of multiple tests to catalog mental processing in different areas is pretty well established at this point in time, and is certainly the norm moving forward. Online IQ tests are fun and sometimes (if you score well) can make you feel good about yourself, but they really don't have a lot of value as a tool of science. The Bell Curve is based almost exclusively on the approach that g is the central measure of intelligence and is reflective of a more complicated reality. It is however, too simple a measure for them to deconvolve their entire theory. Kind of like judging your entire physical health based on your bowling score. Bowling requires coordination, speed, strength, good vision, fast reflexes and mental discipline, all factors measured in physical exam. It is reflective of those aspects, but loses a lot of information in the integration. A better, more accurate approach would involve designing an intelligence test which controlled for multiple environmental and cognitive module biases. However, that would be work and the authors of the Bell Curve are economists, not psychometricians. Simply put, a single number IQ score is too coarse a measure on which to base these kinds of sweeping conclusions.

Intelligence is primarily genetic. While there is absolutely a genetic component to intelligence, the degree to which your potential intelligence as dictated by your genes' control of your brain structure is translated into your functional intelligence which you use every day is determined by lots of factors in your environment many of which no one has any control over. To draw the conclusions Herrnstein and Murray have come to, i.e. that one group of people has a overall lower genetic potential for intelligence than another, they would have to de-convolve all the relevent environmental factors. They don't do this, in part because no one could do this with the data they are using. One possible explanation for the Flynn Effect is the improving environmental and educational environment is allowing people to realize more of their genetic potential (another is that improved access to education is making a larger number of people better readers and they are doing better with the cultural bias of the test).

Intelligence is a fixed, unchangable value after a certain age. I don't even know where to begin. The general effect of the physical environment (disease, pollution, trauma, etc.) is to lower functional intelligence over time for populations exposed to these hazards. In the absence of external factors, IQ scores do tend to be stable, however the Herrnstein and Murray do not control for these factors to any significant degree in their data. I think that have cooked into the data the very effect they are looking for.

Testing Bias. A lot of the data they use is old, and while they attempt to correct somewhat for this, it's lipstick on a pig. I dont think you can meaningfully compare data in the 60s with data from the 80s. The tests are different, built with different assumptions and testing for different effects. It's simply not possible to integrate results in the way they do, not without adding apples and oranges.

Correlation and Causation: All that said, there is one more beef I have with the Bell Curve, at least as work of science, they bury the regression analysis in the back of their book. Their entire premise is based on an R2 value of 0.4. While interesting statistically, no one in a physical science would publish a result like that, much less base a public policy on it. Further, that is their *best* result. Read Appendix 4 of their book, then read this analysis.

Race: Finally, they define a "race" fairly loosely. Their functional definition is, basically, dark people with lots of ancestors from Africa. Unfortunately for them, modern genetics doesn't actually bear out on this, although it turns out there *are* distinct races within the human family. It's a very complex (and very interesting) subject, but again the Bell Curve simply integrates over this complexity. For a researcher to really support the Bell Curve's conclusions, they would need to deconvolve intelligence scores along genetically important racial lines. Obviously this was not possible to do in 1994, so I don't fault the authors, however in science all conclusions are tenative and need to be re-evaluated in the light of new evidence and new models. Depending on the Bell Curve's definition of race too much is unsupportable at this late date.

And that's why I disagree with the Bell Curve. :)