IV-estimation and conditional effects

[Alexander Jepsen]

Hi everyone,

I am wondering what the correct way is to use an interaction term in Stata using xtivreg. As the ## is not allowed I have made a manually interaction variable with the gen var = var1 * var2.

When using the xtivreg I have created an instrument for my main independent variable. I need to be sure how I tell Stata that the variation my instrument causes to my inpendent variable is interacted in the xtivreg commando.

In other words what is the correct way to use an interaction term using IV-estimation in Stata, where the instrument then needs to be conditioned on a Z-variable?

Thanks in advance!

[Roman Mostazir]

Fitting the model with Generalized structural equation model (GSEM) will allow you to use interaction term using factor variable notation.

[Alexander Jepsen]

Fitting the model with Generalized structural equation model (GSEM) will allow you to use interaction term using factor variable notation.

I really am not an advanced user of stata, and I do not know about GSEM. Is there a way you could shortly elaborate as to what I need to perform in Stata?

[Roman Mostazir]

I am sorry, there is no short cut to this and I do not currently have the time to followup this. Allison discusses fitting a basic instrumental variable model using sem here and compares with the 2sls command but it is with single level data. The correct model in his discussion is the 2nd sem model where he added covariance between the error terms. I suggest you invest some time reading the sem manual and get familiar with the language and concepts. SEM itself is a language; the paths, notations, arrows, boxes all carry important concepts (see Allison's discussion). It is a broader subject area but not difficult to learn if you read through the manual. Once you are familiar with the concept, read the section on how to fit multi-level models using gsem as you have repeated observations (I am only assuming as you have used -xtivreg-) and you will eventually know how to fit the instrumental variable model using factor variable notation for interaction term.

Please note it is important that you read the Statalist forum FAQ section and get familiar with how to make useful post by providing data examples using -dataex-. That way you are likely to get more responses by the contributing members. Without data example, people really can't help much on analysis.