generating new variable based on changes in panel data

[Nevena Radoynovska]

Hi,

I've constructed a dataset that includes information about the political party in office for every town between 2002-2014. Elections were held in 2002, 2008 and 2014, so there are 3 election cycles: from 2002-end 2007, 2008-end 2013, and 2014-end 2014 (when my observation period ends). I want to create a variable that indicates whether there was a change in political party for each town between election cycles and how many changes there were total (0, 1 or 2) for the entire observation period.

The variables I have are:

polparty_4: categorical variable labeled as left, center, right, other
code_insee: unique identifier for each town
year: from 2002-2014

I've created the following variables based on the above:
generate party07 = polparty_4 if year==2007
generate party08 = polparty_4 if year==2008
generate party13 = polparty_4 if year==2013
generate party14 = polparty_4 if year==2014

My plan was to create a binary variable coded as 0 if party07=party08 (or party13=party14) for a particular commune, and 1 otherwise.

However, I'm struggling to generate the binary variable grouped by town (code_insee), as well as the count variable for the total number of changes in each town.

Thanks for your help!

[Andrew Musau]

Follow the FAQ advice to present data examples using dataex. For example, you may run the following and copy and paste the output here to increase your chances of getting a helpful reply

ssc install dataex
sort code_insee year
dataex polparty_4 code_insee year in 1/30

[Nevena Radoynovska]

Thank you for the tip. The output is below:

* Example generated by -dataex-. To install: ssc install dataex
clear
input float polparty_4 str5 code_insee int year
. ""         .
1 "01024" 2011
1 "01024" 2012
1 "01024" 2013
3 "01024" 2014
3 "01032" 2007
3 "01032" 2008
3 "01032" 2009
3 "01032" 2010
3 "01032" 2011
3 "01032" 2012
3 "01032" 2013
4 "01032" 2014
3 "01033" 2002
3 "01033" 2002
3 "01033" 2003
3 "01033" 2003
3 "01033" 2004
3 "01033" 2004
3 "01033" 2005
3 "01033" 2005
3 "01033" 2006
3 "01033" 2006
3 "01033" 2007
3 "01033" 2007
4 "01033" 2008
4 "01033" 2008
4 "01033" 2011
4 "01033" 2012
4 "01033" 2013
end
format %ty year

Not all years are available for all towns because the data are also organized according to organizations, so if an organization existed only between 2011-2014 (like in the first 4 lines of output), and it was the only organization in that town, only the years corresponding to the period from the organization's birth to its failure would appear.

[Andrew Musau]

Thanks for the data example. Here, I see that you have some duplicate observations. This does not matter for what you want to do but if you are going to analyze this data as a panel, you will have to address this.

encode code_insee, gen(town_id)
*GENERATE INDICATOR VARIABLE FOR CHANGE
bys town_id(year):gen change= cond(polparty_4[_n-1]!=polparty_4[_n], 1, 0 ) & _n!=1
*COUNT TOTAL NUMBER OF CHANGES
bys town_id:egen nchanges= total(change)

[Nevena Radoynovska]

Thanks so much! This works wonderfully!
The "duplicates" are due to the multiple organizations in a town and multiple observations per organization (for each year), but there is a unique organization-year identifier variable that I didn't include here for simplicity. I appreciate the heads up though.