no observations error for regressing two IVs that are mutually exclusive

[Flora Yin]

Dear Statalist,

I have a dataset with European and non-European countries. I define "europe" as:

gen europe=1 if Countrycode=="AT"|Countrycode=="BG"|Countrycode=="CH"
replace europe=0 if europe==.

and "non-europe" as

gen noneurope=1 if Countrycode=="AU"|Countrycode=="BR"|Countrycode=="CN"
replace noneurope=0 if noneurope==.

The definition here europe and non-europe are mutually exclusive, there is no overlap or a third category.

Then I generate an investment score, "score", which is the aggregate score of all investors in the same country:

bysort id year :egen score_europe = total(score) if europe==1
bysort id year :egen score_noneurope = total(score) if noneurope==1

and I want to regress both score_europe and score_noneurope in the same regression with AEM as the DV and some control variables:

reg AEM score_europe score_noneurope ROA MB growth leverage i.industry i.year i.country_id ,cl(id)

And Stata tells me there is no observation. I kind of understand why there is no observation as the two IVs have no overlap. But if I want to regress score_europe score_noneurope in the same regression, what should I do to achieve this?

Thanks!

[Carlo Lazzaro]

Flora:
the issue seems to rest on missing values in both -score_europe- and -score_noneurope- predictors, the sum of which actually listwise deletes all the observations.
Just to give you an idea, look at the following toy-example:

use "C:\Program Files\Stata16\ado\base\a\auto.dta"
. g new_domestic=1 if foreign==0
(22 missing values generated)

. g new_foreign=1 if foreign==1
(52 missing values generated)

. regress price i.new_domestic i.new_foreign
no observations
r(2000);

To work this issue around, you have to create an unique -investment_score- (say; 0=European countries; 1=Non-european countries) and use -label- to detail them.
Frankly, I would have used the very same approach for classifying European and non-European countries.

[Fei Wang]

Flora, the code for generating score_europe makes non-European countries have missing values for this variable, while European countries have missing values in score_noneurope. As European and non-European countries are mutually exclusive, no observation would have non-missing values in both scores. The key is to define the values of score_europe and score_noneurope for non-European and European countries, respectively.

[Carlo Lazzaro]

Just followed Fei to enjoy the very same party!

[Flora Yin]

Brilliant! Thanks a lot Carlo and Fei!