Hi,
I'm trying to construct an instrument of labor training to know the actual impact on firm productivity. It's widely known that researcher may compute averages from industry/size/region of each one of the variables he wants instrument but excluding the observation of each firm. That way the instrument would be completely exoneous, at least because of construction.
But when you have dummy variables, means are not categorical. You have to round the new variable to get the form of the original variable. I have a possible instrument that is 0 or 1. I took averages over the variable to get my instrument. However when I run a simple round command, the instrument lost all correlation with the variable I want to instrument because of that.
What it would be a criterion to round variables to construct dummy variables in this case?
Jenniffer Solorzano Mosquera
Labor Markets Division - Inter-American Development Bank
Washington, DC - USA
(202) 623-3659
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
