[Maria Ravio]

Sure,
this is the code I run on stata:

reg log_trim_earnhrppp_cont_dcl i.vet_yes c.centered_age i.vet_yes#c.centered_age vetcnew female $control8 $book west_europe east_europe north_europe south_europe usa canada japan korea [pw=weight_adjusted] , vce(cluster cntryid)

This is what I found:

Sorry for the figure (I did not quite understand how to better incorporate those results in statalist).

[Carlo Lazzaro]

Maria:
1) 19 clusters are definitely not enough to safely invoke -vce(cluster clusterid)-. Therefore, if you have detectec heteroskedasticity, apply the -robust- option (as you already did). Otherwise, stick with default standdar errors;
2) run -linktest- but only on the OLS model that gives the fairest and truest view of the data generating process you're studying (testing seven specifications in not the way to go).