Clustering standard errors on Dummies

[Cla Tuor]

Hello all,

I am estimating a first differences model of the following form for 16 countries and 12 time periods: ΔY_it = ß₀ + ß₁ΔX_it + ø_t + λ_R + ɛ_it

ΔY_it is the change in the manufacturing share in country i between year t and t+1.
ΔX_it is the change in import exposure in country i between year t and t+1.
Vector λ_R contains five dummies for different parts of Europe (e.g. Dummy north is equal to 1 if the country is Sweden, Denmark, or Norway, and zero otherwise)

xtset year
xtreg diff_y diff_x north east south west central, r

Related literature uses standard errors clustered on these regions, i.e., clustered on these different dummies (north, east, south, west, central).

How do I incorporate this into my code?

Thank you!

[Carlo Lazzaro]

Cla:
first, you should create a single categorical variable called -region- with different level (one per each geographic zone).
Then you can go:

xtset countries year
xtreg diff_y diff_x north i.region, vce(cluster region)

[Cla Tuor]

Carlo Lazzaro Thank you!

Would you please elaborate on how I have to do the first step?

And why did you add country to the xtset command? Since I use FD I do not have country fixed-effects.

[Jared Greathouse]

I can't comment on much without an example dataset, but what I would do here is

* Just an example, adapt to your purposes
g region = .

replace regionname = "North" if inlist(country,1,2)

replace regionname = "South" if inlist(country,3,4)

egen region = group(regionname)

drop regionname
  
 xtset countries year xtreg diff_y diff_x north i.region, vce(cluster region)

If I understand you well, you simply want to cluster SEs on your regions. Make a unique ID for these regions, then cluster on them in xtreg.

[Jared Greathouse]

You add country to the xtset command because you have a panel of countries and want to tell this to Stata.

[Clyde Schechter]

I think Jared Greathouse means to say -drop regionname-, not -drop region- in #4.

[Jared Greathouse]

Good catch!! Indeed I did mean to say -drop regionnname- Clyde Schechter

[Carlo Lazzaro]

Cla:
Clyde and Jared reached the party sooner than me and replied brilliantly.
1) -xtset-ting your dataset with -timevar- is a bad idea, as Stata is forced to consider it as your -panelid- (which is -country-, in fact);
2) running -xtreg,re- instead of -xtreg,fe- is not a waiver for not to -xtset- your data properly as your first step.

[Cla Tuor]

Unfortunately, it is not clear to me yet how to proceed.

You can find an abstract of my dataset in the table below: 16 countries and six different regions.

Do I have to create dummies for these six different regions first? Or can I directly proceed with the code above?

Do you think you could update the code to my dataset, as I cannot follow it? E.g., what do the numbers stand for in the bracket after inlist?

Thank you, and please bear with me as I am new to STATA.

country	region
Austria	ger
Belgium	ben
Denmark	sca
Finland	sca
France	ben
Germany	ger
Greece	med
Ireland	bi
Italy	med
Netherlands	ben
Norway	sca
Portugal	ibe
Spain	ibe
Sweden	sca
Switzerland	ger
UK	bi

[Jared Greathouse]

Cla Tuor All you do here is

egen regionid = group(region)

I didn't know how your data look, so I couldn't do anything but give a toy example, but since you already have a region variable defined, you just use the syntax I give here in this post.

We'd be able to help you much easier if you used dataex to give an example of what your real data look like.

egen here gives a unique ID to each region that should apply to any nation that has the same regional name.

[Cla Tuor]

It worked! Thank you for your help!