[Katerina Petrou]

Jeff Wooldridge Sorry for writing again but I took your comment in #14 into consideration and tested my y variable for unit roots and I found out that it is non-stationary; therefore I changed my base model reg from fixed effects to first difference method.
But I have a question..
Is it possible to do the instrumental variable estimation above with first differenced variable? or is this not allowed?
I would really appreciate your feedback.

[Daipayan Dhar]

Jeff Wooldridge in your previous reply you used the command xtivreg y (w = z1 ... zM) x1 ... xK i.year, vce(cluster id)
for the 2 stage model. Later, in your second reply, you used the command xtivreg y (w = z1 ... zM) x1 ... xK i.year, fe As I can understand, in the first one, you are clustering the ids and in the second one you are not. So, which is more appropriate? Also, do I need to keep i.year when I am running a fixed effect model with ",fe" in the end of the command? Will it be okay to just take the "Year" instead of the dummy? Also, wherever I am not specifying the fixed effect in the command, my regression is taking the Random Effect GLS. So is it okay?

[Daipayan Dhar]

Jeff Wooldridge Also, I wanted to know if xtivreg2 is a better command. Because there I can specify both my model, that is fixed effect, and then also cluster my IDs. Along with that it is also coming up with post estimation tests to check for endogeneity and overidentification. I haven't found any blog where they specify the difference between xtivreg2 and xtivreg. Everywhere it just says that xtivreg2 is a user developed command.