Controlling for Year Dummies, Industry Dummies and Country Dummies

[Raed Yusuf]

I want to be guided on the stata commands for fixed and random effects when controlling for (1) year dummies, (2) industry dummies, (3) country dummies.

Fixed effect commands (with year dummies):
xtreg y x1 x2 x3 x4 x5 x6 x7 i.year, fe

areg y x1 x2 x3 x4 x5 x6 x7 i.year, absorb(country)
regress y x1 x2 x3 x4 x5 x6 x7 i.country i.year


Fixed effect commands (with year & industry dummies):

xtreg y x1 x2 x3 x4 x5 x6 x7 i.year i.industry, fe
areg y x1 x2 x3 x4 x5 x6 x7 i.year i.industry, absorb(country)
regress y x1 x2 x3 x4 x5 x6 x7 i.country i.year i.industry

However, if I want to control for year, industry, and country dummies, does it mean that I will have to use only the regress command (regress y x1 x2 x3 x4 x5 x6 x7 i.country i.year i.industry)?

Again, I discovered that when industry dummies and year dummies are included in the xtreg, fe command, the results for the industry dummies will be deleted. Does it mean that I can only use the regress command via the LSDV method?

[Carlo Lazzaro]

Raed:
welcome to this forum.
Too many dummies indeed.
Please note that, if you want to go -fe-, time-invariant variables will be wiped out. For instance, firms are usually embedded in the same industry; hence, due to demeaning (of a constant), industry predictor will be omitted.
This result does not depend on -i.year- inclusion (that you should keep in the right-hand side of your regression equation), but on time-invariance of industry.
The same comment holds for -country-, unless some firms delocalize abroad during the time-horizon your dataset stretches over.
That said:
- I would go:

xtset firm year
xtreg y x1 x2 x3 x4 x5 x6 x7 i.year, fe

Another option that you may want to consider is the community-contributed module -reghdfe- (just type -search reghdfe- from withn Stata to spot and install it).

[Raed Yusuf]

Many thanks for your support.

What if I want to use the Least Square Dummy Variable (LSDV) approach to estimate the fixed-effect model. Is it possible to control for year, industries and country dummies? Because we do assume that there is heterogeneity in different industries and countries when samples are collected from different industries across countries.

[Carlo Lazzaro]

Raed:
LSDV shows some drawbacks when you have >2 waves of data.
In addition, it will not solve the time-ivariance issue.
As an aside, the -fe-.estimator gets rid of both observed and unobserved time-invariant panel-specific heterogeneity.

[Raed Yusuf]

Thanks for the response. I want to be well guided. Are you saying that when using the -fe-. estimator, there is no need to control for industry and country dummies except for the year dummies? If yes, how do I control for industry and countries dummies as many top articles in my field have them in their models?

[Carlo Lazzaro]

Yusuf:
the issue rest on the -fe- machinery (as you may to read yourself in any decent panel data econometrics textbook): if -country- and -industry- are time-invariant predictors, there's no way -fe- estimator can give you back their coefficients.
The paper you refer to probaly used -re-.