# Cross-sectional Regression (individual level) with a few country-level variables

I have a small sample of 50 cross-sectional firms and 3 or 4 distinct explanatory variables -- all on the individual level. No time dimension. So far, I could employ OLS (I am using Stata: reg x y, robust).

However, my 50 individual observations come from 20 countries. I would like to include country-specific variables. Is it possible (in general) to just add these country-specific characteristics to the OLS regression on the individual level? Consequently, there would not be any variation in these small country-groups for the variable in question. Also I'd like to add dummy variables, that imply that the 50 firms belong to one of two distinct groups.

Do I need to use a multilevel model for nested data in this case? I don't want to see the effects of the countries explicitly, but only for the variable in question (imagine, e.g., GDP_GROWTH).

• Just to clarify: Are you thinking of something like "FrenchTaxesPaid" which only be non-zero for companies in France? Or are you thinking of something like "SovereignCreditRating" which would be the same for each firm in the same country? May 6 '16 at 21:49
• You might also want to cluster your errors on county if you include these, though you might not have enough data to include them. It might also be possible to do a panel fixed efdect model if you are not interested in the effect of the country-level variables. Sep 17 '17 at 18:28

You need to check multicollinearity issues, but you'll notice them as your betas and stderr will fly around . You could check out the Stata ado coldiag2 for analysing multicollinearity .