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).
Best Answer
I see no problem with this approach, if you only use one country-specific variable.
If you have more than one, maybe you could country-group ("rich countries" "poor country" or some other grouping). If you have further information, e.g. which region people come from, that may be used to create more variation.
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
coldiag2for analysing multicollinearity .