4 ms·
Wonderful commentary, thank you! Can you describe some other techniques used to eliminate unobserved variable bias, or do you know of a resource that describes
by mediaman 17y ago
Wonderful commentary, thank you!
Can you describe some other techniques used to eliminate unobserved variable bias, or do you know of a resource that describes these?
I remember in one of the other studies on the impact of "elite" education versus state-provided university education, they used people who were accepted to an elite university, but chose not to attend. Is that using "acceptance" as an instrumental variable?
- stanleydrew 17y agoI've been out of the econ phd world for a little while now (see blog post here: http://hellosorld.com/personal/2009/07/16/why-i-quit-my-phd-in-economics.html http://hellosorld.com/personal/2009/07/16/why-i-quit-my-phd-...) so it's hard to remember what all the resources were, but i'll ask around for some good material and post it here. Natural random experiments are another good source of identification (i.e. elimination of unobserved variable bias). For the elite vs state university question, choosing to attend isn't an instrument since it's still an observed decision. If there was some source of randomness that was uncorrelated with all your other variables besides attending vs not, then that would be an instrumental variable. Say the government decided to take a pool of 10,000 elite college acceptees, gave them each a number from one to 10,000 and then said only the even numbers could actually attend. That would be your instrument. Trouble is that those kinds of natural random experiments don't occur often. That's why economists can get tenure simply by finding a good instrument for some study.