Belligerati

« Yes, I know it is just an advertisement | Main | Unpleasant truth of the day »

August 23, 2008

Tune the engine all you want, but if you need a lot of power then buy a big one

High-Aptitude Minds: The Neurological Roots of Genius is an article that attempts to elucidate the causes of intelligence through the findings of a series of experiments by neurologists. A couple of things struck me about these papers.

The first is that the article mentions sample sizes in these experiments and they are all very small, with none larger than 50. There is a strong concern that with this small sample that phenomena involving a large number of parameters will not appear distinct from noise. In one experiment mentioned, an experimenter measures 10 brain regions in 47 people. Even if brain weight is a cause and not an effect of intellect is a big if. Arguendo, if there is covariance in the distribution of these regions then a sample this small could have 110 parameters (10 coefficients and 100 entries in the covariance matrix). You can't have more parameters than members of your sample! Even if they are all independent you'd still have 20 parameters and 47 observations. That's at the very lowest bounds of what is enough data to do statistics.

Second the causality arguments seem flawed to the point of establishing nothing. The seem to fall into two categories. The primary one is that somehow correlation is causality, that is, I observe a simple correlation between two data series I observe therefore one causes the other. The alternative is post hoc ergo prompter hoc, which is that I did something and then something happened and therefore I must have caused it.

I recognize that MRI time is expensive and funding is scarce. But if this is the best you can come up with with the budget you have then why bother?

Posted by OneEyedMan at August 23, 2008 7:33 AM

Comments

Post a comment