Mediation Panel data

[OLMABA JALA]

Dear all,

I am currently conducting a fixed effects panel regression in Stata with the following variables:
CompanyID
Group (either 1 or 0)
Year (2005-2015)
Performance (profit in €)

The one group of companies went through a training (group 1) the other group of companies that is comparable in terms of their business model etc. did not attend the training (group 0)
The main question is what impact the training has on the dependent variables (among others profit).

In addition I want to include a mediator into the model: For instance I want to check if the company's leadership motivation is effected by the training and has an influence on the DVs (e.g. profit)
Do you know how to implement this into stata? Is it possible for me to run separate regressions? e.g. one regression xtreg leadership motivation Group, fe and then xtreg Profit leadership motivation, fe?

[Carlo Lazzaro]

Olmaba:
try:

xtreg profit i.group##i.training##i.leadership_motivation, fe vce(cluster companyid)

[OLMABA JALA]

Thank you so much Carlo
I am not sure what you were referring to with i.training since the variable group is defining if a company was in training (value 1) or not (value 2)

So I went with this
xtreg profit i.group##c.leadership_motivation, fe vce(cluster companyid)
I am getting a significant effect of group
I am getting a significant effect of leadership_motivation
But I am not getting a significant effect of group#leadership_motivation

When running separate regressions (- xtreg leadership motivation Group, fe - and - xtreg Profit leadership motivation, fe - I am getting a significant effect of group on profit directly but also a significant effect of group on leadership_motivation and from leadership_motivation on profit...

[Carlo Lazzaro]

Olmaba:
1) please report along with description the outcome table(s) returned by Stata (as per FAQ). Thanks;
2) -i.training- is a categorical variable that includes two levels: no training (better coded as 0) and training (better coded as 1);
3) you ran different regression models: no wonder that you obtained different results: which one is to be preferred, depends on the literature in your reserach field.

[OLMABA JALA]

Thanks Carlo! This is helping a lot.

But it is not generally wrong running separate regressions (one regression xtreg leadership motivation Group, fe and then xtreg Profit leadership motivation, fe) and interpreting the results for leadership motivation as a mediator?

[Carlo Lazzaro]

Olmaba:
yes, it is.
That's why I proposed one regression code only (with interactions) hoping to be on target.

[OLMABA JALA]

Thanks Carlo!

Sorry for questioning again. To avoid a misunderstanding: Is is wrong or not wrong to run separate regressions and interpreting the results for leadership motivation as a mediator?

[Carlo Lazzaro]

Olmaba:
yes, it's wrong.
Some researchers use -sem- for this kind of analysis (but I'm not familiar with that).