Collinearity between two country variables, one nominal (country name) and the other metric (globalisation level)

[Bjorn Moller Ochsner]

Hi

I experience heavy (perfect) multicollinearity between a control variable and my main independent (which is also used as interaction term). But the control variable is vital and can't be excluded, so don't know what to do.

The data is Eurobarometer, i.e. individual level. I want to examine whether globalisation at the country level (X1) influences the effect of the Trump election (X2) on democratic satisfaction (Y). So, an interaction with an objective, country level measure for globalisation.

However, I also need to have the individual's country as a control variable, to leave out any influence from cultural, historic etc. factors. Given that I must include this control, is there anything I can do to avoid (perfect) collinearity?

I have already tried 1) to make the metric globalisation variable categorical, grouping countries in everything between 2 and 6 globalisation levels. But even when I make a dummy globalisation variable, the collinearity persists. 2) Running a - bysort reg - with these categorical globalisation variables. This obviously removes the collinearity problem, but doesn't feel methodologically justified(?)

Thanks

[Phil Bromiley]

Welcome to Statalist. You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex. Being able to replicate your problem can be essential to helping you.

If you're including dummies for country, then in a fe panel estimate, you lose all variables that don't vary within a panel. i.country in an ols regression is essentially the same as xtreg, fe with country as the panel. Take a look at xthybrid and the mundlak estimators.