Tuesday, February 14, 2006

Nature-Nurture Entanglement

I finished Pinker’s The Blank Slate a couple of weeks ago, but I’ve been toting it around with me ever since with the intention of making a couple more blog posts.

Much of the book, and especially the chapter on children, attempts to categorize how much of human behavior is attributable to nature (genes) rather than nurture (environment); a leading theme of the book is that the former accounts for more, and the latter less, than many people would like to admit. In terms of explaining variation in human behavior, Pinker’s bottom line is like so: “Genes 50 percent, Shared Environment 0 percent, Unique Environment 50 percent (or if you want to be charitable, Genes 40-50 percent, Shared Environment 0-10 percent, Unique Environment 50 percent).” (Shared environment is that part of environment shared across siblings; unique environment is the rest.)

I have no beef with the argument, although I suspect Pinker may exaggerate his case slightly. But a point I find more interesting – which Pinker hints at but doesn’t seem to make directly – is that some portion of behavior (and not just the unexplained portion) cannot be attributed to either nature or nurture, but can only be understood as caused by both.

Consider an analogy (inspired by a problem my dad showed me a few years ago): Suppose your firm buys coal. In year 1, you buy 100 pounds for $1.50 a pound, for a total expenditure of $150. In year 2, you buy 120 pounds for $2 a pound, for a total expenditure of $240. How much of the $90 difference in expenditure is attributable to the increase in your coal usage, and how much to the rise in price? Holding quantity constant at 100, the price increase explains $50 of the expenditure increase. Holding price constant at $1.50, the quantity increase explains $30 of the expenditure increase. Add these together and you get only $80. The remaining $10 cannot be explained by either factor alone; it’s explained only by the fact that both price and quantity increased. It came from the price difference of $0.50 multiplied by the quantity difference of 20.

We could divvy up the $10 by finding the proportion of the total (already attributed) change that came from each cause. Of the $80 increase explained by price increase alone or quantity increase alone, $50 was attributable to price. So we might say that 5/8 of the $10, or $6.25, also came from price increase, and the other $3.75 from quantity increase. But this would just be an accounting device; it doesn’t alter the fact that some fraction of the expenditure jump depended crucially upon variation in both price and quantity.

Similarly, say there’s a gene (or gene complex) G1, which manifests as behavioral trait A. Gene G2 manifests as behavioral trait A in social environment A’ and behavioral trait B in social environment B’. Holding the gene pool constant (with both genes present), greater variation in the social environment will produce greater variation in behavior. Holding the social environment constant, greater variation in the gene pool will produce greater variation in behavior. If both genetics and social environment vary, some portion of variation will be attributable to each source of variation and some will be attributable to both.

Bonus points: Note the similarity with the point I made here.


Patri Friedman said...

For a good discussion of the interplay between nature and nurture, see Matt Ridley's "Nature Via Nurture", also published as "The Agile Gene".

Ben said...

I also thought Pinker was exagerating a bit. Reading the book made me feel like he was trying to convince me 90% of our behaviour is inate but then he gives us the 50% figure which I think most people would have found reasonable without further evidence.

His found his previous book "How the Mind Works" better. Don't read it though. It countain's mostly the same material but without the inateness agenda.

Ben said...

Also the issue with the nonlinearity is quite recurrent in multivariate analysis. Researchers almost always linearly approximate relations unless there is strong evidence that suggests linearization would not be appropriate. No one wants to do the extra mathematics and render their model hard to interpret with confusing relations.

Multiple hierarchical regression is one method of dealing with it.

PithLord said...

Another way Pinker cooks the books is by focusing on personality traits. Accepting that variance in optimism or grumpiness is almost all due to genes or individual environment, the same doesn't apply to beliefs or norms.