Dear Statalisters,

I am writing for help because I need the following issue:

1) I am studying how the decision of a CEO influence the level of R&D intensity on firms (R&D expenses/sales). So I have firms from 50 different countries and for a time spam of 20 years.
2) My dependent variable is a ratio between 0 and 1, including several 0 (zeros). So I decided to use tobit for the regression, actually I am thinking to use xttobit. Since there are firms nested in countries, I tried with metobit, but I didn't have convergence, I think for reasons that I explain later.
3) My independent variable is a dummy, at a country level, related with political stability. Then, there are countries with variation in this dummy, but I also have countries with 0 all the years and 1s (ones) all the years.
4) I want to test culture (masculinity) as a moderator (a country level variable), a continuous variable (from 1 to 100), this variable is different between country, but is time invariant within countries. The same value of masculinity is expected for each year in a determined country.

That said, I am confused about how to test the model:

1) First, I was thinking to use a
Code:
xtset countryid
because I was thinking to create a
Code:
vce(cluster countryid)
, but I can't use vce using xttobit. I tried using bootstraping, but since I am planning around 40 different models including robust and post hoc analysis, I am going to finish in 2 months without turn it off the computer.
2) I was thinking in fixed effect, but, since my moderator is time invariant and the IV have also some countries with time invariant data, I am concerned that i.year or i.country is going to take all the variance. Actually, I tried
Code:
xtset firm year
with the metobit and didn't converge, reading other posts make me think that my (mis)specification with maximum likelyhood (with time-country invariant variables) is the reason.

Then, my question is, are there any specification (advice) that you could recommend me? How do you think I should proceed in order to avoid the problem of considering all the observations as from one country? or avoid the problem of ommited variables without including a country fixed effect?

Thank you very much for any guide about it.

Alejandro