Treatment variable is "omitted because of collinearity"

[James Sonela]

Dear Colleagues,

I am trying to conduct a diff-in-diff analysis with fixed effects in my panel. My entities are countries. I have created my diff-din-diff variables manually. I have only 1 treated country and that is the UK, where the remaining 7 countries are my control entities.

The problem is when I run my "xtreg treated post2014 didestimator controls, fe vce(cluster country_id)", I receive an error called "omitted because of collinearity". The issue is obvious. Since the treated variable is 1 for the UK and 0 for the all the other countries, and this does not change over time, the fixed effects model deletes the time-invariant variables. I have read a similar topic but could not see any recommendation other than running a random effects (which I have no issue but have to run a fixed-effects model too)

I will appreciate your help on this issue.

* didestimator is the interaction term between treated and post2014

	(1)
VARIABLES	Model 6
o.treated	-
post2014	0.915
	(0.898)
didestimator	-2.398*
	(1.092)
GDP_GROWTH	-0.0304
	(0.0297)
INFLATION	0.114
	(0.330)
UNEMP	-0.268
	(0.191)
Constant	-0.521
	(0.975)
Observations	259
R-squared	0.109
Number of country_id	8

[Carlo Lazzaro]

James:
welcome to this forum.
Shouldn't it be that a pre-treatement period is available for UK, too?

[James Sonela]

James:
welcome to this forum.
Shouldn't it be that a pre-treatement period is available for UK, too?

Hi Carlo,

Thanks for the reply. The post-2014 variable takes the value of 1 after the 2014 and 0 before that for all countries (including the UK). The treatment variable is 1 for the UK (for all the time periods). Am I doing something wrong?

[Carlo Lazzaro]

James:
your DID setup is correct.
I assume that the treatment variable is always 1 for UK after 2014 only

[James Sonela]

Carlo:

I am very sure that the treatment variable is 1 for both pre and post periods for the UK. I do not know how to figure out this. I have only 1 treated country and small number of entities. Do you think this creates a problem?

[Carlo Lazzaro]

James:
I'd recommend taking a look at https://www.princeton.edu/~otorres/DID101.pdf