Is the 'g' factor nothing more than a statistical construct?
By hugo, Friday 17 November 2006 :: Cognition :: #120 :: rss
Scores on a wide range of intelligence tests tend to correlate positively. From a statistical or psychometric point of view this creates a variable, g that merely indicates the strength of this correlation. If there were no correlation at all, there would be no g, but since the correlations tend to be high, people get excited and many of them take the next step of positing an underlying common cause (also called g). For the psychologists who defend this notion, there is a common variable (modulating, say, the way your neurons fire) that influences on the measures of all of these intelligence tests, thus creating the observed correlation. However researchers from the University of Amsterdam are challenging the common wisdom and suggest an explanation for the correlation that doesn't need a common cause.
This picture has no other purpose than to attract the reader towards a boring post on models of intelligence. Note that the band is called the G factor though, that gives me sort of an excuse.
Since a picture can say more than a thousand words (admittedly, also because it's considerably shorter to copy and paste a picture than write a thousands words), here is the bulk of the new model (figure from the paper):
In both schema, xs are psychological mechanisms (measured by intelligence tests). The schema on the left depicts the classical g model: g is seen as an underlying cause, with some extra variance/noise being added to each mechanism (the us). The model on the right is a bit more complex (and they have others even more complex in the paper). The Ks indicate resources that are used by the different mechanisms (brain power if you will). All the fancy arrows are actually the heart of the model: they indicate that all the different mechanisms influence each other. With this model, you have two ways to explain a correlation between the xs: you can say that there is a correlation between the Ks (such that some people have more overall 'energy' than others), but then you're back with a new form of g. What is interesting is the other option: you can explain the correlation by positing that the influence that the different mechanisms have on each other is positive overall (i.e. if you sum the weights of the influence that mechanisms have on each other the result is positive).
If cognitive mechanisms are mutually reinforcing each other, then it explains g without any need for an underlying common cause. To me it seems very plausible indeed that psychological mechanisms should reinforce each other more often than they inhibit each other. The authors do some fancy modeling that I'm not really able to judge, and they claim that their model can explain the main phenomena in the field of intelligence at least as well as the classical g model.
There are several things I like about this model: first, as I said, I find it very plausible from a psychological point of view. Second, there is no need to posit some sort of 'domain general intelligence factor', and for several reasons I'm quite averse to that sort of things. Third it fits in nicely with a view of the mind as a bunch of modules that compete for energy (actually, it is such a view). Fourth and last it promises to bring in the field more dynamical thinking: in this model, the observed correlation is not due to anything but the process of development. If the system is considered in a limited time period, the model is not even possible (and thus, if the model turns out to be on the right track, there can be no good way to explain the observed correlation without a recourse to developmental/dynamical thinking).
A last word. The authors mention in their paper another very interesting paper on the theme of intelligence: an explanation of the Flynn effect. In two words, the Flynn effect is an increase (a big one at that) in IQ observed since IQ tests were designed (approx. a century). This explanation relies on the idea that "people shape or select their environment depending on their IQ. Higher IQ leads one into better environments, causing still higher IQ, and so forth." (quote from the first paper). I like the idea that you could try to integrate the environment (and especially the social environment) into the model presented here. More on this in a later post.
Comments
1. On Saturday 18 November 2006 by olivier
2. On Saturday 18 November 2006 by alberto
3. On Saturday 18 November 2006 by Kenneth Stein
4. On Monday 20 November 2006 by hugo
5. On Monday 20 November 2006 by hugo
6. On Saturday 25 November 2006 by Michael Blowhard
7. On Saturday 25 November 2006 by hugo
8. On Saturday 25 November 2006 by Michael Blowhard
9. On Saturday 2 December 2006 by Alvaro
10. On Saturday 2 December 2006 by Jason Malloy
Add a comment
Comments for this post are not allowed