Saturday, January 12, 2013

Genes as Instrumental Variables

So shopping trips can be intellectually stimulating if your partner is also a social scientist :) On a trip to Costco today I commented on what seemed to me a higher proportion of "nicely rounded" people than at other places in Canberra and we started "debating" about the causality. Then I wondered about what sort of instrumental variable could sort that out. Well obviously if we had a gene marker associated with being fat that would potentially be a perfect instrumental variable - it would be correlated with being fat but shouldn't affect going to Costco in any other way, you'd think... and would certainly not be caused by going to Costco. We then jokingly discussed some ways to get genetic information from people as well as their weight without their knowing. Unfortunately ethics rules would forbid any of these and you'd need to get people's agreement to participate :)

When we got home I did some research. A full genome sequence still costs close to $10k. There is some research along these lines out there. It seems that besides cost the biggest issue is the low correlation between any given gene and given phenotypes. Such instruments would be weak. So it seems that multiple indicators and or factor analysis is needed. But if these issues are overcome and the cost comes down further, I can see that this will be become a huge area in many areas of economics including health and education obviously.

No comments:

Post a Comment