Hi,
I have a couple of questions about treatreg. I don't know much about
econometrics in general. In particular, I don't know much about this
treatreg procedure.
Here are the two questions:
1) Does it make sense to include an interaction term of the endogenous
treatment variable with a predictor variable in treatreg?
For example, in the model below, my hypothesis is that the effect of lfp
depends on the status of wc.
webuse labbor, clear
gen wc=0
replace wc=1 if we>12
xi: treatreg ww i.wc*lfp , treat(wc= wmed wfed) twostep
I think conceptually, this model should be doing the following:
ww = beta0 + beta1*lfp + beta2*lfp*wc + beta3*wc + e (1)
wc = g0 + g1*wmed + g2*wfed + u (2)
and e and u are bivariate normal with variance sigma and 1 and
covariance rho.
For this example, I do have trouble running it using ml. But I have seen
models that actually run with ml estimator. But the fact that it runs
mechanically does not really tell much if the model makes sense or if
the estimates are biased or not.
It seems that if any substitution of equation (2) happens in (1), it can
only happen to beta3, not to beta2 since this will make lfp a random
effect and that is not what has been hypothesized.
2) Does fixed-effect model make sense for treatreg? Let's say we have
data over multiple years and we want to control the year effect. Would
it make sense to just include the dummy variables for year in the model
to account for the fixed effect of year?
I hope these questions make sense and I really appreciate any hints on
them from the experts.
Best,
Xiao Chen
Statistical Consulting Group
UCLA Academic Technology Services
http://www.ats.ucla.edu/stat/
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/