Sunday, May 23, 2010

Individuals vs. Groups.


To generalize is to be an idiot. To particularize alone is a distinction of merit.
~William Blake

Over at the Yale lecture series, there's a bunch of neat Intro to Psych lectures by Paul Bloom. In one of them, titled Why are People Different?, he notes that people have various attributes that help explain what they do: intelligence, openness, conscientiousness, extraversion, agreeableness, and neuroticism. These attributes are partially genetic, and they are interesting because they 1) are rather stable over an individual's life and 2) predict important real-world behavior.

Bloom highlights these are important because they explain why individuals are different. He then notes that when applied to groups, they are not important. His explanation here is rather unconvincing. He note Lewontin's famous anecdote of a crop of wheat, where seeds have a genetic potential that is very important, but a field of wheat can have different characteristics because one field is fertilized differently. True enough. But Bloom then says, because this is true, one can't say anything about groups of individuals. This simply isn't true, as the same reasoning can be applied to individuals, even more so. Further, one can do studies that abstract from important omitted variables, by in effect controlling for things like fertilizer, just as one does when estimating the effect of a characteristic for an individual.

Think of a stock. It has characteristics like its industry group, beta, and P/E ratio. These characteristics are important because they have different implications for risk and, sometimes, expected returns. When analyzing a large portfolio, this way of slicing things is informative. However, when looking at a specific stock to buy, the more you analyze the particulars of a stock like GOOG, these broad characteristics become merely one datum among many, almost irrelevant. Basically, factor analysis explains a lot only when applied to groups, not single stocks, for which the error is sufficiently large as to not make it a very meaningful bit of information compared to a narrative of GOOG's business strategy and stock performance against its peers. In contrast, most econometric financial research involves grouping stocks into deciles, sorted by a variable of interest.

I find it interesting that top psychologists only want to apply human characteristics to its most irrelevant application, the individual. That is, the the characteristics he discusses seem to have greater relevance to groups than individuals, but they don't want to look at that based on a very weak objection, that members don't all behave like their group. As if a tendency has to be 100% or 0% to be interesting. A young woman who avoids a dark alley where a bunch of young men are standing around is making a generalization that may be unfair, but prudent. Individuals often behave counter to their type--being more agreeable, or intelligent, in different applications on different days--but people understand that does not invalidate any label like 'intelligence'. I'm not saying one should only look at groups, just that it's an interesting application, no one is doing research here, and it's important. There is a lot of discussion of group differences in socio-economic variables, and the achievement gap seems to be the top focus of any American urban school system. Race, gender, and ethnicity are somehow both very important when discussing social justice, but also totally unimportant--if not meaningless--biologically.

There is a logical error in supposing that all groups are fundamentally the same on every metric of human behavior. Many things are heritable, important, and vary systematically between groups. Individual differences don't automatically wash away at the group level, otherwise there would never be evolution outside of rare bottleneck events (the whole 'Preservation of Favoured Races' subtitles in Origin of Species). It's really a rather weak assertion--that individual characteristics are at least as important at the group level--yet one of the taboos of our time.

1 comment:

Michael F. Martin said...

So you're a fan of Duflo, then?