Counting the number of times a dummy variable changes value

[Zsolt Marai]

Hello,

I have an unbalanced dataset from 1999 to 2019, but with gaps (not sure if it's important to mention).

I have a variable called public_dum which is equal to one if the individual works in the public sector, and 0 if he/she works in the private sector.

tab public_dum

 public_dum |      Freq.     Percent        Cum.
------------+-----------------------------------
          0 |      8,816       82.65       82.65
          1 |      1,851       17.35      100.00
------------+-----------------------------------
      Total |     10,667      100.00

My goal is to count the number of individuals that goes from the private to the public sector over years. For simplicity, I would stop counting at the first occurrence (even if some individuals might switch several times).

So ideally, I would create a variable called switch that takes 1 if the individual switched (at least once) from private to public, and 0 if she/he has never switched.

I am not very experienced in Stata, and I have difficulty in starting it, so I would appreciate if someone could shed some lights on it.

Here is a subset of the three variables concerned:

* Example generated by -dataex-. To install: ssc install dataex
clear
input long idpers int year float public_dum
 11101 1999 .
 21101 2001 0
 26102 1999 .
 42101 2000 .
 42101 2001 0
 42101 2002 0
 42101 2003 0
 42101 2004 0
 42101 2005 0
 42101 2006 0
 42101 2007 .
 42101 2008 0
 42101 2009 0
 42101 2010 0
 42101 2011 0
 42101 2012 0
 42101 2013 0
 42101 2014 0
 42101 2015 0
 42101 2016 .
 42101 2017 0
 42102 2000 0
 42102 2001 0
 42102 2002 1
 42102 2003 1
 42102 2004 1
 42102 2005 1
 42102 2016 1
 45101 1999 0
 45101 2000 0
 45101 2001 0
 45101 2007 .
 45101 2008 .
 45101 2009 .
 45101 2010 0
 45101 2011 .
 45101 2012 .
 45101 2013 0
 45101 2014 .
 45101 2015 .
 45101 2016 0
 45101 2017 0
 45101 2018 0
 45102 1999 0
 45102 2000 0
 45102 2001 0
 45102 2007 0
 45102 2008 0
 45102 2009 0
 45102 2010 0
 45102 2011 0
 45102 2012 0
 45102 2013 0
 45102 2014 0
 45102 2015 0
 45102 2016 0
 45102 2017 0
 45102 2018 0
 45102 2019 0
 45103 2015 .
 45103 2016 .
 45103 2017 .
 45103 2018 .
 45103 2019 .
103101 1999 0
103101 2000 0
103101 2001 0
103101 2009 0
103101 2010 0
103104 2001 .
103104 2009 .
103104 2010 .
103104 2011 .
103104 2012 .
103104 2013 .
103104 2014 .
103105 2000 0
117101 1999 0
117101 2000 0
117101 2001 0
117101 2002 0
117101 2003 0
117101 2004 0
117101 2005 0
117101 2006 0
117101 2007 0
117101 2008 0
117101 2009 0
117101 2010 0
117101 2011 0
117101 2012 0
117101 2013 0
117101 2014 0
117101 2015 0
117101 2016 0
117101 2017 .
117101 2018 .
117102 1999 .
117102 2000 .
117102 2001 .
end
label values idpers IDPERS

[Carlo Lazzaro]

Zsolt:
you may want to try:

. bysort idpers: gen wanted=sum(public_dum)

. g switch=1 if wanted==1

[Asjad Naqvi]

What are missing values for?
And in your data example the switch is happening only once for one person. So your data description is not reflected in the example data.

A rudimentary code below:

xtset idpers year
gen switch = D.public_dum

first line declares your data a panel dataset
The second line just takes the first difference of values [t - t-1] and respects the ID variable (that is it won't subtract the starting value of a new idper from the end value of the previous idper.

Where

 switch == 1

is where the switch happens. You can make your conditions more specific on the actual data, especially if you are only looking for the first occurrence of a switch.

HTH!

[Zsolt Marai]

Thank you for your answer, Carlo.

Actually, I was thinking about doing it in this spirit:

gen switch=0
bys idpers: replace switch=1 if public_dum[_n]=!public_dum[_n-1]

So that I can count individuals that switch from public to private as well, but Stata warns be "invalid syntax".

Do you eventually see what is wrong?

Thank you

[Zsolt Marai]

Thank you, Asjad. I will try your code.

Missing values refer to a particular kind of public employees that I am not interested in.

[Nick Cox]

bysort idpers (year) : gen switch = public == 1 & public[_n-1] == 0 

by idpers : egen ever_switch = max(switch) 

egen tag = tag(idpers) 

count if tag & ever_switch

is other technique. See the help on egen and count for more.

[Zsolt Marai]

Thank you, Nick. I understand the first line. Can you please briefly explain what the others do, if you don't mind? Thank you.

[Nick Cox]

The use of egen, max() here matches https://www.stata.com/support/faqs/d...ble-recording/

tag() is explained tersely by the help for egen.The original 1999 publication has a bit more at dm70 in https://www.stata.com/products/stb/journals/stb50.pdf

90% of the understanding of count comes from understanding the command name. The manual entry has links to other papers.