Hello all,
I use a first differences model to estimate the impact of globalization on the manufacturing share in Europe, using a panel dataset with 16 countries and yearly observations between 1995-2007.
The following first-differences model has been estimated, instrumenting ΔXit with ΔZit:
Does it make sense to use country fixed-effects in a first-differences regression? I have been told that adding country-fixed effects to the first-differences model captures different time trends across countries. I find it very confusing as I thought that one is already controlling for country fixed effects by taking first-differences.
In what way could adding country fixed-effects to the FD-model bias my results?
Furthermore, the explanatory variable of interest is only statistically significant when including country fixed-effects (see attached screenshots).
Isn't it bizarre that adding country fixed-effects to the FD model makes my coefficient more statistically significant?
Finally, I use robust standard errors in the model without country-fixed effects, as clustering at the country-level yields an error message. Is that OK?
I appreciate your help!
I use a first differences model to estimate the impact of globalization on the manufacturing share in Europe, using a panel dataset with 16 countries and yearly observations between 1995-2007.
The following first-differences model has been estimated, instrumenting ΔXit with ΔZit:
ΔYit = ß0 + ß1ΔXit + øt + λi + ɛit, where ΔYit denotes the change in the manufacturing share in country i from year t to t+1, and ΔXit denotes the change in the import exposure in country i from year t to t+1.
Code:
xtset country year xi: xtivreg diff_y (diff_x=diff_z) i.year, fe vce(cluster country)
In what way could adding country fixed-effects to the FD-model bias my results?
Furthermore, the explanatory variable of interest is only statistically significant when including country fixed-effects (see attached screenshots).
Isn't it bizarre that adding country fixed-effects to the FD model makes my coefficient more statistically significant?
Finally, I use robust standard errors in the model without country-fixed effects, as clustering at the country-level yields an error message. Is that OK?
I appreciate your help!
Attached Files
