[Joao Santos Silva]

Dear David,

Thanks for this. I had a look at the original paper but I am afraid that the terminology they use is very different from the one I am familiar with and therefore I really am not in a position to advise you on this. You need to talk to someone with experience in that particular area.

Sorry I cannot be more helpful and best of luck with this.

Joao

[David Lu]

Dear David,

Thanks for this. I had a look at the original paper but I am afraid that the terminology they use is very different from the one I am familiar with and therefore I really am not in a position to advise you on this. You need to talk to someone with experience in that particular area.

Sorry I cannot be more helpful and best of luck with this.

Joao

Dear Joao,

Thanks for your reading. To the best of my knowledge, they use -regression- and obtain the residual from the first stage and use another -regression- with these residual to finalize the second-stage regression. So, in the context of -poisson-, does it make sense to you if using these commands as following:

glm endogenv1 ins1 cv1 cv2 cv3 endogenv2,family(poisson) link(log) vce(robust)
predict endogenv1_r, residual
glm endogenv2 ins2 cv1 cv2 cv3 endogenv1,family(poisson) link(log) vce(robust)
predict endogenv2_r, residual
glm dv1 c.endogenv1_r##c.endogenv2_r##c.iv3 cv1 cv2 cv3 ,family(poisson) link(log) vce(robust)

NB: endogenv1 and endgoenv2 refer to 2 endogenous variables, cv1-3 refer to 3 control variables, ins1 and ins2 refer to instrument variables while iv3 refers to the third independent variable.

Does it make sense to you in terms of the issue of data skewness and potential endogeneity?

Thanks again,
David