Instrument as control variable for main hypothesis in earlier section of a paper, when 2SLS only used in later section

[Franz Hopp]

Dear Statalist Community,

I was wondering about good practice in economcetric research, in particular in the field of Strategic Management. I am working on a project where we could potentially have an issue of reverse causality. We look at a relationship from the following sort:

firm_strategy + controls --> firm_performance

And in our specific context, there is for example potential reason to assume that firm_performance might also influence which firm_strategy a company chooses. Regarding our paper submission, we plan to go about this reverse causality issue in the following way (not perfect, but the best we can do with our data, I fear):

• (1) For the main part of the data/findings section of the paper, we plan to have a simple RE regression. We control for past firm_peformance to address reverse causality
• (2) And, in order to look at the issue of reverse causality in more detail, we plan to report results from a 2SLS in the robustness section of the paper. But we plan to report 2SLS results only later on in the paper, in a later 'robustness check' section. The instrument we are thinking of is "CEO_change". (There is good reason to assume it drives our firm_strategy, but that it does not have a direct causal link on firm_performance)

So in the above context, my question relates to point (1) -- to the "normal" RE regression in the main data/findings section of the paper:

• Here, is it seen as good practice to include our instrument already in the regression in this main regression? Should we include CEO_change as a control variable already in this regression specification? Or is it not necessarily required to include the variable that we later on in the paper use as an instrument as a control already?

Thanks so much in advance.

Franz

[Carlo Lazzaro]

Franz:
if you have (as it seems) good reasons to suspect endogeneity via reverse causality, I would probably skip the step (1) altogether, Obviously, most depends on the literature in your research field and the customary rules adopted by the target journal you're planning to submit your paper to.