Fully Interacted Model with 2SLS

[Chinmay Sharma]

Hi All,

I am attempting to estimate a Mincer-type equation, which basically models wages for an individual i at time t as a function of various covariates. Given that education is generally thought of as being correlated with unobservables, such as ability, one can think of the coefficient on education being biased. Suppose I have an instrument z, which randomly assigns education to individuals. I have data by country and by year. The regression I wish to estimate resembles:

ivregress 2sls wage (sex)##(i.country i.year c.experience) (education=z)

I get an invalid syntax error in this regression. What I wish to do is to basically allow for differential effects of experience, country of origin, and year by the sex of an individual rendering this as a fully interacted model, whilst modelling education as being instrumented by z. I suspect that the 2sls command recognizes brackets as identifying the endogenous variable and instrument, but the presence of brackets in my case also indicates the full interaction effects, evidenced by the ##. This I suspect is the reason for the error.

Is there any way to combine this sort of interaction model with 2SLS?

Thanks,
CS

[Chinmay Sharma]

One solution is just to multiply the brackets out:

 
 ivregress 2sls wage sex##i.country sex##i.year sex##c.experience (education=z)