Panel data, Stata omits i.Year dummies due to collinearity

[Sander Wyckmans]

Hi all,

I am fairly new to Stata and I ran into a problem with my panel data set. My data set consists of multiple countries with each including the same 7 industries over a time period of 2010-2017. My supervisor advised me to use time dummies so I did using the i.Year function. Now my problem, when I run the code [by Country, sort : xtreg BERD FDIinflow GDPPCG Enterprises Turnover Educationlvl ULC Inflation i.Year, fe] I get the note that Stata omits variable i.2015 or other i.Years varying across countries due to collinearity. I don't get why Stata does this and if this affects the validity of my database. If i run xtreg and don't do it by country, so the whole dataset, I get an r² that is almost zero.

Thanks in advance and my apologies if I didn't make things clear or let me know if i should include any attachments!

(Using Stata 16.0)

[Andrew Musau]

Some of your variables, e.g., inflation will be collinear with the year indicators. If you have a panel of firms, such variables do not vary across firms in a given year. If these variables are just control variables, then there is nothing to worry about. However, if they are your independent variables of interest, you have a problem as you cannot identify their coefficients in the presence of year indicators. I would switch to reghdfe from SSC to easily identify the collinear variables.

ssc install reghdfe, replace
ssc install ftools, replace
reghdfe BERD FDIinflow GDPPCG Enterprises Turnover Educationlvl ULC Inflation, absorb(countryind Year)

Now my problem, when I run the code [by Country, sort : xtreg BERD FDIinflow GDPPCG Enterprises Turnover Educationlvl ULC Inflation i.Year, fe] I get the note that Stata omits variable i.2015 or other i.Years varying across countries due to collinearity. I don't get why Stata does this and if this affects the validity of my database.

Reading #1 again, you are creating the problem by running your regressions by country. You have cross-country panel data and your advisor probably wants a cross-country analysis. Do not bother with the R-squared (which is computed differently in xtreg compared to OLS [LSDV]), just specify your model, run it and interpret the results.

[Sander Wyckmans]

Andrew Musau
You are indeed correct that inflation is used as a control variable. Thanks a lot for your helpful insight! I really appreciate your help and the solution was very usefull!
In attachment, you can see the results that I have now thanks to your solution! Just to be clear, all these coefficients are meant to be interpreted as across countries? If I want to run xtreg just for one country, what would be the best way to approach that?

[Andrew Musau]

For a particular country, use the -if- qualifier

reghdfe BERD FDIinflow GDPPCG Enterprises Turnover Educationlvl ULC Inflation if country="UK", absorb(countryind Year)

If you are doing this for all countries:

bys country: reghdfe BERD FDIinflow GDPPCG Enterprises Turnover Educationlvl ULC Inflation, absorb(countryind Year)

or

levelsof country, local(countries)
foreach country of local countries{
    di "Regression for `country'"
    reghdfe BERD FDIinflow GDPPCG Enterprises Turnover Educationlvl ULC Inflation if country="`country'", absorb(countryind Year)
}

where I assume that "country" is a string variable above. If country is a numerical variable

levelsof country, local(countries)
foreach country of local countries{
    di "Regression for country code `country'"
    reghdfe BERD FDIinflow GDPPCG Enterprises Turnover Educationlvl ULC Inflation if country=`country', absorb(countryind Year)
}

The issue with running regressions by country is that variables such as inflation which do not vary across industries within a country at a point in time will drop out in the presence of year indicators. Otherwise, you should cluster your standard errors in the cross-country regression. You do not appear to have enough clusters when running the regressions by country.

reghdfe BERD FDIinflow GDPPCG Enterprises Turnover Educationlvl ULC Inflation, absorb(countryind Year) cluster(countryind)