Saturday, 9 March 2013

Is the search for exogeneity pointless in complex systems?


This idea came up in a response to a reviewer.

One of my hypotheses is that a psychological trait, honesty, may affect certain social or economic outcomes. The reviewer raised the legitimate point that other unobserved personality traits, correlated with honesty, could be the real causal factor. This is undeniable. We don’t know enough about human personality to control for everything that can affect a person’s social relationships.

The normal economist’s response would be to seek a source of random variation that affected honesty but nothing else: for example, an exogenous instrumental variable; or a natural experiment; or a real experiment. For example, if you want to see if economic growth makes war less likely in poor countries, but you are worried about reverse causality or unobserved confounds, then you may argue that weather is random, affectseconomic growth and does not affect war except via economic growth.

For psychological variables, though, this just seems impossible. Even if you could run an experimental treatment to change someone’s level of honesty (maybe an extra hour of Sunday school?), how could you guarantee that this would not change other aspects of their personality too? In fact, that would be extremely unlikely. A person’s character is a complex and interconnected whole. There seems no way to rule out unobserved heterogeneity – short of major neurosurgery, perhaps.

So it seems that if we want to investigate human psychology, we are stuck with finding the major dimensions of personality variation (such as the “Big Five”) via various forms of dimensionality reduction, and then controlling for them.

It seems as if this argument should also apply to other areas of science which study complex systems not subject to precise manipulation – say, ecology or climate science. What can we do about that?