Sunday, April 22, 2012

The Dickens-Flynn Model

[Edit 02/11/2017: The final version of this essay can be found here.]

The Dickens-Flynn model is an attempt by William Dickens and James Flynn to explain how environment and genes can interact to account both for environmentally-driven increases in raw intelligence scores over time (the Flynn effect) and for the persistent influence of genetically-driven individual intelligence differences (Spearman's g). Admittedly, there is much that is appealing about the Dickens-Flynn model, for it does directly address the observable tension between the Flynn effect and g (something that many explanations for the Flynn effect fail to do). And yet the model is also troubling in a rather conspicuous way, for it appears to sport the overall mechanics of a Rube Goldberg machine.

The original description of the Dickens-Flynn model is provided in Dickens & Flynn (2001)—unfortunately that paper seems not to be freely available. The authors do however provide an accessible summary at this location. Another source for the details constituting the Dickens-Flynn model is Flynn's book What is Intelligence? (Flynn 2007), which at various points outlines both the workings of the model and its explanatory power vis-a-vis the Flynn effect. There have also been several published criticisms of the Dickens-Flynn model, including this one from Linda Gottfredson, and Dickens has responded to some of these criticisms in outline form here.

Leverage plays a key role within the Dickens-Flynn model. For instance, small genetic advantages are described as being amplified through the means of a natural attraction to environmental surroundings that serve to enhance those advantages. Environmental influences on population-wide intelligence characteristics are said to be boosted through such factors as social multipliers and rolling triggers. And finally, genetic and environmental forces are depicted as mutually propelling each other by means of reciprocal feedback mechanisms. Indeed, there is something particularly mathematical about the use of such concepts as multipliers, triggers and feedbacks—instead of defining these terms through real world descriptions and empirical examples, the model more frequently casts them in the role of free variable, serving to balance out and drive the equations where required. This tendency towards calculational necessity can also be witnessed to some extent in Dicken's clipped responses to the model's criticisms, where arguments against social multipliers and circular feedback mechanisms are answered through an appeal to simultaneous equations, estimated network effects and functional form.

None of this is to say that the Dickens-Flynn model is inherently wrong or essentially inconsistent. It remains perfectly conceivable that the model's many mechanisms and various factors can be fitted (with a little more mathematical tweaking perhaps) to the observable characteristics of human intelligence behavior. But even if this is so, in parsing through such ideas as social multipliers, rolling triggers and reciprocal feedback mechanisms, one is left with the distinct impression that the Dickens-Flynn model can in no way be described as straightforward—and this of course is in contrast to the phenomenon which the model is intended to explain, a phenomenon which can be captured in very few words (human raw intelligence scores increase consistently and universally over time). A charitable way of depicting the Dickens-Flynn model would be to say that it is intricate and complex; a less charitable depiction would be to say that it is convoluted.

Which raises a question: why would Dickens and Flynn insist on providing a model of such intricacy and complexity? Why did they not gravitate instead to something more direct, with fewer moving parts? Flynn, ever the paragon of candor, provides a thorough and revealing answer in the pages of What is Intelligence?. There Flynn notes that during the height of the intelligence race debates, Richard Lewontin had offered a compelling description for how an environmental influence could uniformly impact an entire population while still leaving undisturbed any genetically-driven individual characteristics:

Lewontin tried to solve the paradox. He distinguished the role of genes within groups from the role of genes between groups. He imagined a sack of seedcorn with plenty of genetic variation randomly divided into two batches, each of which would therefore be equal for overall genetic quality. Batch A is grown in a uniform and optimal environment, so within that group all height differences at maturity are due to genetic variation; batch B is grown in a uniform environment which lacks enough nitrates, so within that group all height differences are also genetic. However, the difference in average height between the two groups will, of course, be due entirely to the unequal quality of their two environments.

Although Lewontin's description proved mostly untenable within the context of the intelligence race debates, Flynn later recognized that it seemed to form an ideal explanation for the Flynn effect:

So now we seemed to have a solution. The present generation has some potent environmental advantage absent from the last generation that explains its higher average IQ. Let us call it Factor X. Factor X will simply not register in twin studies. After all, the two members of a twin pair are by definition of the same generation. Since Factor X was completely missing within the last generation, no one benefited from it at all and, therefore, it can hardly explain any IQ differences within the last generation. It will not dilute the dominance of genes. Since Factor X is completely uniform within the present generation, everyone benefits from it to the same degree and it cannot explain IQ differences within the present generation. Once again, the dominance of genes will be unchallenged. Therefore, twin studies could show that genes explain 100 percent of IQ differences within generations, and yet, environment might explain 100 percent of the average IQ difference between generations.

And yet as Flynn attempted to apply Lewontin's idea specifically to the Flynn effect, he quickly found himself frustrated:

However, Lewontin offers us a poisoned apple. History has not experimented with the last two generations as we might experiment with plants in a laboratory. Consider the kind of factors that might explain massive IQ gains, such as better nutrition, more education, more liberal parenting, and the slow spread of the scientific ethos. It is quite unreal to imagine any of these affecting two generations with uniformity. Certainly, everyone was not badly nourished in the last generation, everyone well nourished at present; everyone without secondary school in the last generation, everyone a graduate at present; everyone raised traditionally in the last generation, everyone raised liberally at present; everyone bereft of the scientific ethos in the last generation, everyone permeated with it at present. If the only solution to our paradox is to posit a Factor X or a collection of such, it seems even more baffling than before. We should shut this particular door as follows: a solution is plausible only if it does not posit a Factor X.

It is in this manner that Flynn places a strange restriction around what he might accept as a plausible explanation for the Flynn effect: the explanation would have to carry all the desirable characteristics of a Factor X, but without actually being a Factor X. And it is into the space of that restriction that the Dickens-Flynn model comes to be. Flynn is quite forthcoming about what he perceives to be the model's most valuable characteristic:

Best of all, our solution posits no Factor X.

It is out of this historical context that we can get a better sense for why the Dickens-Flynn model must be necessarily complex and intricate, and not straightforward. It is as though Flynn has set forth the task of building a device designed to capture mice, and yet the device must in no way resemble a mousetrap. Therefore it's not all that surprising that soon we find ourselves inundated with the equivalent of levers, pulleys, waterwheels, whistles and bells, and if the objection is made that in the end it's still just a mousetrap, the architects can point to the enormous amounts of clever machinery and insist no one would mistake it for any such thing. And if the objection is raised that the mice are after all scurrying away, the assemblers can suggest that perhaps a few more pieces of duct tape and chicken wire should take care of the matter. Whether or not the many parameters of the Dickens-Flynn model can ultimately be fitted to the characteristics of human intelligence would be difficult for me to judge, but that does not take away from the observation that the model is a massively entangled contraption.

I'll be blunt: Flynn's mistake is in shutting the door on a Factor X.

Flynn's reasoning is perfectly correct when he notes that not only is every iterated item on his Factor X list inadequate for explaining the Flynn effect, but that indeed no particular item or subset of items could conceivably possess the requisite power or ubiquitous reach to account for population-wide, generational increases in raw intelligence scores. But instead of giving up—instead of shutting the door—what is needed at this point is some thinking outside the box. Or as I like to put it, some thinking outside the human skull.

In my essay The Flynn Effect's Unseen Hand, I describe a model for human intelligence that defines the term environmental intelligence as the total amount of non-biological pattern, structure and form tangibly contained within the human environment. Environmental intelligence serves not only as the palpable location of human intelligence, but its consistent increase over the course of human history serves also as the direct driver of the Flynn effect. It might not be perfectly correct to label environmental intelligence as a Factor X, since it's not really a factor, not a thing—instead it captures the impact of the environment as a whole. But nomenclature aside, environmental intelligence carries all the necessary characteristics of a Factor X. Environmental intelligence is ubiquitous. Environmental intelligence is constantly increasing. Environmental intelligence remains independent of the genetic influences driving individual intelligence differences.

The common objection that might be raised against a model of environmental intelligence is that it places the material locus of human intelligence firmly outside the human skull, firmly outside the human brain. But in exchange for jettisoning the widespread assumption that intelligence is a product strictly of human neurons, in exchange for accepting the total amount of non-biological pattern, structure and form contained within the human environment as the ideal Factor X, what is regained thereby is a great deal of simplicity. Gone is the need for social multipliers. Gone is the need for triggers. Gone is the need for feedback loops. Indeed gone is the need for any type of genetic/environmental interaction at all, especially of the complex, intricate kind. What one regains by jettisoning the Dickens-Flynn model and accepting instead a model of environmental intelligence is all the simplicity and straightforwardness that Lewontin must have originally had in mind.

I may be alone in this opinion, but I believe simplicity and straightforwardness should still count for something. I like to catch my mice with a mousetrap.

Dickens, W. T. & Flynn, J. R. (2001). Heritability estimates versus large environmental effects: The IQ paradox resolved. Psychological Review, 108: 346–369.

Flynn, J. R. (2007). What is intelligence? Beyond the Flynn effect. New York: Cambridge University Press.

Saturday, April 21, 2012

Progress for the Masses

It's a funny thing about science these days. Brilliant science is as difficult to do as it always was, and bad science still requires a certain amount of (immoral) finesse. But good science is as easy as firing up the statistics package and networking with a few dozen co-authors on the iPhone.

Friday, April 20, 2012

Science Interrupted

At the heart of nearly every scientific achievement made over the past many centuries you will find two significant activities:

  1. The collection of an accurate set of data.
  2. The development of a cogent description consistent with that data.

Data mining doesn't fall under either of those headings.

Thursday, April 19, 2012

The Queen of Data Mining

Many years ago a weekly lottery was started in my home state, and for awhile the newspapers would print a front-page article announcing that Mr. So-and-So from Anytown had won that week's million-dollar prize. My grandmother would read those articles with her eyes wide as saucers. “Can you just imagine? Mr. So-and-So won a million dollars last night.” She said it as if some kind of miracle had taken place. I never could quite explain to her that in publishing the results after the lottery was held, the news was not quite as significant as it seemed, that only if it had been specifically predicted ahead of time that Mr. So-and-So from Anytown was going to win the lottery would the result be more akin to a miracle, and not just a meaningless certainty.

That anecdote might not seem relevant at first to autism, but it's the key to understanding Irva Hertz-Picciotto's approach to science. In fact these days, it's the key to understanding just about everyone's approach to science.

Tuesday, April 17, 2012

Diminished Capacity

I'm surprised no one has pointed out yet the obvious weakness in the latest Nature-published data mining attempts. Clocking in at a mere 59, 23 and 30 authors respectively, these studies clearly must be trivial compared to earlier findings. Heck, Geraldine Dawson didn't even bother to get herself placed on any of the authorship lists, the chances of which must have been something like 0.0001%. (I had Matthew State confirm the statistics on that last figure.)

Saturday, April 14, 2012

More Modern Phrenology

Here is the latest instance of the type of logic being applied in cognitive science—the feeling around for bumps inside the human skull.

To be honest, I'm not sure it's all that fair to compare phrenology to neuroscience. The insights are the same, but the phrenologists accomplish them at a far cheaper cost.