Random effect model with the independent variable a dummy variable

[Rafael Giorgio]

Dear all,

I have a problem regarding my research. When performing a GLS regression, the coefficient vallues of GDPpercapita are very strange.
My research is about whether imposed import sanction influence the export performance of the target country. Hereby, the IV is the import restriction (1 or 0 in a specific year) and the DV the export performance. I use inflation, total export, GDP, GDP per capita as control variables. The interaction variable is directimposedmod, which is 1 if the sanction is directly imposed and 0 if there was first a threat.

Using panel data, is this the right way to perform my research?

Kind regards

Rafael

[Clyde Schechter]

The magnitude of these coefficients is, for the most part, very large and somewhat difficult to wrap one's mind around. You will get more understandable results if you change the units of exporttosend from dollars (or Euros, Yen, whatever) to millions of dollars (resp. Euros, Yen, whatever). (i.e. divide the variable by 1,000,000.) When you do that, you should make a similar transformation to totalexport and GDP.

I am not an economist so I cannot comment on whether this model makes sense from a scientific perspective.

Since you want to run an interaction model, you would be better off using factor-variable notation (unless you are using a very old version of Stata). But I can't show you how to do that because it is not evident which variables directimposedmod is the interaction of. But instead of using a calculated interaction variable, you would be better off using the factor-variable equivalent. See -help fvvarlist-. Then after running the regression you can use the -margins- command to make interpretation of your results easier. See the excellent Richard Williams' https://www3.nd.edu/~rwilliam/stats/Margins01.pdf for a very clear introduction to the -margins- command, including worked examples with interaction models.