How interpret a model with one # in the interation term

[Marry Lee]

Hello,


I run heterogeneity analysis by interacting my variable of interest which is continuous with an indicator for gender.

The problem is that my variable of interest is also endogenuous so I use an IV in the following way.

ivreg2 Y (c.X#i.male = c.IV#i.male) controls

I got the follwoing result

male#c.X|
                                    0  |   .0078223     .00254     3.08   0.002     .0028439    .0128007
                                    1  |   .0102124    .003351     3.05   0.002     .0036446    .0167802

What is bothering me is that I used only one # in this model.
How do I interpret these result please??


Also, ivreg2 does not allow for ##, so I used the following:

ivreg2 Y (X c.X#i.male = IV c.IV#i.male) controls

Is this correct? and which model is better suited in my case for my heterogenity analysis?

Result with the second model:

   X|   .0078223     .00254     3.08   0.002     .0028439    .0128007
                                       |
male#c.X|
                                    1  |     .00239   .0031224     0.77   0.444    -.0037298    .0085099

Thank you so much!

[Daniel Santos Torres]

Hi Marry, I'm not 100% sure, but here's my opinion in case it might be helpful:

I'd suggest to look at the first stage to better understand what's happening. You can do it with the option ffirst. Having said that, I'm not sure about including c.IV#i.male as an instrument per se. I'd try the following instead:

ivreg2 Y (c.X c.X#i.male = IV) i.male controls

Where X tells you the effect of X for male = 0 and c.X#i.male tells you if there is any difference in the effect of X between male = 1 and male = 0, if I'm not mistaken.

I'd also recommend to look at other posts in the forum related with instrumental variables and interactions with the endogenous variable that you might find helpful, such as the following:

https://www.statalist.org/forums/for...tion-violation

https://www.statalist.org/forums/for...nous-variables

Best regards,
Daniel