counting without dropping

[Paris Rira]

Hi dear Profs and colleagues,

I have 3 variables, firms ID: NPC_FIC , year & Firmsage: firm_age. I am going to obtain firmsage i,t. i for each firm and year 2010-2011 while keeping all obs.
(The period is for 12 years I mentioned an example for the first 2 years)

* Example generated by -dataex-. For more info, type help dataex
clear
input long NPC_FIC byte firm_age int year
500000002 25 2010
500000002 25 2010
500000002 25 2010
500000002 25 2010
500000002 25 2010
500000119 22 2010
500000119 22 2010
500000119 22 2010
500000119 22 2010
500000119 22 2010
500000119 22 2010
500000119 22 2010
500000119 22 2010
500000346 18 2010
500000346 18 2010
500000346 18 2010
500000346 18 2010
500000346 18 2010
500000376 64 2010
500000376 64 2010
500000376 64 2010
500000376 64 2010
500000376 64 2010
500000376 64 2010
500000376 64 2010
500000376 64 2010
500000376 64 2010
500000395 12 2010
500000395 12 2010
500000395 12 2010
500000395 12 2010
500000395 12 2010
500000395 12 2010
500000856 25 2010
500000856 25 2010
500000856 25 2010
500000856 25 2010
500000856 25 2010
500000856 25 2010
500000856 25 2010
500000856 25 2010
500000856 25 2010
500000856 25 2010
500000002 26 2011
500000002 26 2011
500000002 26 2011
500000002 26 2011
500000002 26 2011
500000119 23 2011
500000119 23 2011
500000119 23 2011
500000119 23 2011
500000119 23 2011
500000119 23 2011
500000119 23 2011
500000119 23 2011
500000346 19 2011
500000346 19 2011
500000346 19 2011
500000346 19 2011
500000346 19 2011
500000376 65 2011
500000376 65 2011
500000376 65 2011
500000376 65 2011
500000376 65 2011
500000376 65 2011
500000376 65 2011
500000376 65 2011
500000376 65 2011
500000395 13 2011
500000395 13 2011
500000395 13 2011
500000395 13 2011
500000395 13 2011
500000395 13 2011
500000856 26 2011
500000856 26 2011
500000856 26 2011
500000856 26 2011
500000856 26 2011
500000856 26 2011
500000856 26 2011
500000856 26 2011
500000856 26 2011
500000856 26 2011
end

if I do this:

bysort NPC_FIC : keep if _n == _N

only one firm will be left, but I don't look for that. Is there any way to compute firms' age for i & t without dropping?

All my thanks.

[Joro Kolev]

In the example you provide the firm age is constant by firm and by year:

. bysort NPC_FIC year: assert firm_age[1]==firm_age[_N]

.

so the assertion is correct. So it is not clear what you want to compute, in your example firms age is already computed.

[Paris Rira]

Thank you for getting back to me.
you are right. Firms' age has already been computed. What I need is to obtain each firm's age. Now there are repeated firms ID and naturally repeated firms age. So I need to keep/count one without dropping (firms age I,t).

[Joro Kolev]

I still do not understand...

Do you want to tag only one observation per firm/year group? If so

egen mytag = tag(NPC_FIC year)

and then whatever computations that you want to do that include only one observation per firm and year, you add the qualifier if mytag. Say

. tabstat firm_age if mytag, by(year) stats(mean count)

Summary for variables: firm_age
Group variable: year 

    year |      Mean         N
---------+--------------------
    2010 |  27.66667         6
    2011 |  28.66667         6
---------+--------------------
   Total |  28.16667        12
------------------------------

[Paris Rira]

Let me put it this way. Please ignore #1.
I start from the beginning.
I need to add a control variable namely: firms' age.
Since some firms have several establishments, therefore they appear more than once in the data.
Question: if I enter firms' age as a control variable in this condition ( duplicated numbers for firms' age and firms' IDs) is that correct?
The explanatory variable has 84 obs. while the firms age is around some thousands (number of establishments exist in the data)
Thank you.Enable GingerCannot connect to Ginger Check your internet connection
or reload the browserDisable in this text fieldRephraseRephrase current sentence6Edit in Ginger×

[Joro Kolev]

Now I think I understand :P.

1. If your dependent variable, and your main explanatory variable vary at the establishment level, it is correct to include the repeated age. You do not need to do anything, you just run the regression on your raw data. In this case here you fundamentally are conducting the analysis at establishment level, and not on firm level.

2. If your dependent variable and your explanatory variable vary only at the firm level, that is, your regression is fundamentally a firm level regression and not and establishment level regression, then you need to do what I showed you in #4. You tag one establishment per firm, and you run the regression only on the tagged observations.

[Paris Rira]

Dear Kolev, Thank you so much for the comprehensive explanation. The econometric model is a bit complex though

i: firms
s: sectors
d: districts (7 districts)
t: 2010-2021

the explanatory variable of interest, Sdt has 7 districts * 12 years=84
Xit: firms age (control variable)

The dimension of the explanatory variable and the control variable is not the same. Is still the second option in #6 valid?