Graph twoway line by group: adding a different vertical line for each group

[Erik Beecroft]

I'm using twoway line to create graphs of earnings by year. I want to create separate graphs for 20 individuals. For each graph, I want to create a vertical line for the year in which that individual left the program. The year will differ by individual, and is captured by the variable lastyear2.

Here is the syntax without the vertical line part:

twoway (line wage_year2 acadyear, sort) if sample2<=20 & award==1, ///
ytitle(Annual earnings) ytitle(, size(small)) ylabel(, angle(horizontal) format(%9.0fc)) ///
xtitle(Year) xtitle(, size(small)) ///
by(, title(Earnings for a Random Sample of Program Completers, size(medsmall)) ///
note("")) ///
by(sample2)

I think xline won't work because it wants a number, and doesn't seem to allow a separate value for each individual.

scatteri also doesn't allow an argument that varies by individual.

I tried using the function command, but I'm not sure function can graph a vertical line.

A subset of the data is below.

+------------------------------------------+
sample2 acadyear wage_y~2 lastye~2

1 2010 5000 2013
1 2011 7000 2013
1 2012 2000 2013
1 2013 33000 2013
1 2014 0 2013
1 2015 0 2013

2 2008 10000 2014
2 2009 13000 2014
2 2010 14000 2014
2 2011 13000 2014
2 2012 9000 2014
2 2013 15000 2014
2 2014 47000 2014
2 2015 52000 2014

3 2010 25000 2014
3 2011 28000 2014
3 2012 21000 2014
3 2013 20000 2014
3 2014 0 2014
3 2015 5000 2014

4 2009 4000 2012
4 2010 20000 2012
4 2011 17000 2012
4 2012 6000 2012
4 2013 0 2012
4 2014 0 2012
4 2015 0 2012

5 2008 2000 2014
5 2009 1000 2014
5 2010 0 2014
5 2011 7000 2014
5 2012 9000 2014
5 2013 0 2014
5 2014 12000 2014
5 2015 54000 2014

6 2009 0 2014
6 2010 1000 2014
6 2011 4000 2014
6 2012 3000 2014
6 2013 0 2014
6 2014 19000 2014
6 2015 60000 2014

7 2009 0 2014
7 2010 0 2014
7 2011 28000 2014
7 2012 29000 2014
7 2013 30000 2014
7 2014 41000 2014
7 2015 17000 2014

8 2009 11000 2014
8 2010 2000 2014
8 2011 0 2014
8 2012 4000 2014
8 2013 10000 2014
8 2014 0 2014
8 2015 0 2014

9 2008 0 2010
9 2009 0 2010
9 2010 30000 2010
9 2011 36000 2010
9 2012 35000 2010
9 2013 44000 2010
9 2014 45000 2010
9 2015 56000 2010

10 2010 5000 2012
10 2011 5000 2012
10 2012 12000 2012
10 2013 11000 2012
10 2014 22000 2012
10 2015 29000 2012

11 2009 3000 2013
11 2010 0 2013
11 2011 0 2013
11 2012 0 2013
11 2013 42000 2013
11 2014 61000 2013
11 2015 27000 2013

12 2008 4000 2016
12 2009 4000 2016
12 2010 11000 2016
12 2011 15000 2016
12 2012 22000 2016
12 2013 22000 2016
12 2014 14000 2016
12 2015 38000 2016

13 2008 19000 2010
13 2009 16000 2010
13 2010 40000 2010
13 2011 38000 2010
13 2012 42000 2010
13 2013 46000 2010
13 2014 47000 2010
13 2015 46000 2010

14 2009 1000 2011
14 2010 0 2011
14 2011 0 2011
14 2012 0 2011
14 2013 0 2011
14 2014 0 2011
14 2015 0 2011

15 2009 0 2012
15 2010 1000 2012
15 2011 5000 2012
15 2012 21000 2012
15 2013 11000 2012
15 2014 0 2012
15 2015 0 2012

16 2010 0 2013
16 2011 0 2013
16 2012 0 2013
16 2013 37000 2013
16 2014 38000 2013
16 2015 29000 2013

17 2008 22000 2011
17 2009 6000 2011
17 2010 20000 2011
17 2011 47000 2011
17 2012 46000 2011
17 2013 44000 2011
17 2014 53000 2011
17 2015 64000 2011

18 2009 1000 2014
18 2010 0 2014
18 2011 6000 2014
18 2012 8000 2014
18 2013 11000 2014
18 2014 12000 2014
18 2015 6000 2014

19 2010 8000 2013
19 2011 10000 2013
19 2012 9000 2013
19 2013 35000 2013
19 2014 46000 2013
19 2015 48000 2013

20 2009 7000 2011
20 2010 5000 2011
20 2011 24000 2011
20 2012 36000 2011
20 2013 32000 2011
20 2014 6000 2011
20 2015 0 2011

[Dave Airey]

Maybe a third approach base on

sysuse auto
bysort foreign, egen max_weight = max(weight)
summarize max_weight if foreign, meanonly
graph twoway scatter length weight if foreign, xline(`r(mean)')
summarize max_weight if !foreign, meanonly
graph twoway scatter length weight if !foreign, xline(`r(mean)')

You could wrap something like this in a foreach loop to get the levels of each individual.

[Nick Cox]

With a little surgery your data and code allow a reproducible example. Please note the use of CODE delimiters.

clear 
input sample2 acadyear wage_year2 lastyear2 
1 2010 5000 2013 
1 2011 7000 2013 
1 2012 2000 2013 
1 2013 33000 2013 
1 2014 0 2013 
1 2015 0 2013 
2 2008 10000 2014 
2 2009 13000 2014 
2 2010 14000 2014 
2 2011 13000 2014 
2 2012 9000 2014 
2 2013 15000 2014 
2 2014 47000 2014 
2 2015 52000 2014 
3 2010 25000 2014 
3 2011 28000 2014 
3 2012 21000 2014 
3 2013 20000 2014 
3 2014 0 2014 
3 2015 5000 2014 
4 2009 4000 2012 
4 2010 20000 2012 
4 2011 17000 2012 
4 2012 6000 2012 
4 2013 0 2012 
4 2014 0 2012 
4 2015 0 2012 
5 2008 2000 2014 
5 2009 1000 2014 
5 2010 0 2014 
5 2011 7000 2014 
5 2012 9000 2014 
5 2013 0 2014 
5 2014 12000 2014 
5 2015 54000 2014 
6 2009 0 2014 
6 2010 1000 2014 
6 2011 4000 2014 
6 2012 3000 2014 
6 2013 0 2014 
6 2014 19000 2014 
6 2015 60000 2014 
7 2009 0 2014 
7 2010 0 2014 
7 2011 28000 2014 
7 2012 29000 2014 
7 2013 30000 2014 
7 2014 41000 2014 
7 2015 17000 2014 
8 2009 11000 2014 
8 2010 2000 2014 
8 2011 0 2014 
8 2012 4000 2014 
8 2013 10000 2014 
8 2014 0 2014 
8 2015 0 2014 
9 2008 0 2010 
9 2009 0 2010 
9 2010 30000 2010 
9 2011 36000 2010 
9 2012 35000 2010 
9 2013 44000 2010 
9 2014 45000 2010 
9 2015 56000 2010 
10 2010 5000 2012 
10 2011 5000 2012 
10 2012 12000 2012 
10 2013 11000 2012 
10 2014 22000 2012 
10 2015 29000 2012 
11 2009 3000 2013 
11 2010 0 2013 
11 2011 0 2013 
11 2012 0 2013 
11 2013 42000 2013 
11 2014 61000 2013 
11 2015 27000 2013 
12 2008 4000 2016 
12 2009 4000 2016 
12 2010 11000 2016 
12 2011 15000 2016 
12 2012 22000 2016 
12 2013 22000 2016 
12 2014 14000 2016 
12 2015 38000 2016 
13 2008 19000 2010 
13 2009 16000 2010 
13 2010 40000 2010 
13 2011 38000 2010 
13 2012 42000 2010 
13 2013 46000 2010 
13 2014 47000 2010 
13 2015 46000 2010 
14 2009 1000 2011 
14 2010 0 2011 
14 2011 0 2011 
14 2012 0 2011 
14 2013 0 2011 
14 2014 0 2011 
14 2015 0 2011 
15 2009 0 2012 
15 2010 1000 2012 
15 2011 5000 2012 
15 2012 21000 2012 
15 2013 11000 2012 
15 2014 0 2012 
15 2015 0 2012 
16 2010 0 2013 
16 2011 0 2013 
16 2012 0 2013 
16 2013 37000 2013 
16 2014 38000 2013 
16 2015 29000 2013 
17 2008 22000 2011 
17 2009 6000 2011 
17 2010 20000 2011 
17 2011 47000 2011 
17 2012 46000 2011 
17 2013 44000 2011 
17 2014 53000 2011 
17 2015 64000 2011 
18 2009 1000 2014 
18 2010 0 2014 
18 2011 6000 2014 
18 2012 8000 2014 
18 2013 11000 2014 
18 2014 12000 2014 
18 2015 6000 2014 
19 2010 8000 2013 
19 2011 10000 2013 
19 2012 9000 2013 
19 2013 35000 2013 
19 2014 46000 2013 
19 2015 48000 2013 
20 2009 7000 2011 
20 2010 5000 2011 
20 2011 24000 2011 
20 2012 36000 2011 
20 2013 32000 2011 
20 2014 6000 2011 
20 2015 0 2011 
end 

gen top = 60000
gen zero = 0 

twoway line wage_year2 acadyear, sort  ///
ytitle(Annual earnings, size(small)) ylabel(, ang(h) format(%9.0fc)) ///
xtitle(Year, size(small)) ///
by(sample2, title(Earnings for a Random Sample of Program Completers, size(medsmall)) ///
note("") legend(off)) ///
|| rspike zero top lastyear2 , lw(thin)

[Dimitriy V. Masterov]

Please use -dataex- when posting data. It makes things so much easier and improves the odds that someone will help you.

I would probably loop over each observation and combine the graphs at the end like this:

cls
clear
input sample acadyear wage_year2 lastyear2 
1 2010 5000 2013 
1 2011 7000 2013 
1 2012 2000 2013 
1 2013 33000 2013 
1 2014 0 2013 
1 2015 0 2013 
2 2008 10000 2014 
2 2009 13000 2014 
2 2010 14000 2014 
2 2011 13000 2014 
2 2012 9000 2014 
2 2013 15000 2014 
2 2014 47000 2014 
2 2015 52000 2014 
3 2010 25000 2014 
3 2011 28000 2014 
3 2012 21000 2014 
3 2013 20000 2014 
3 2014 0 2014 
3 2015 5000 2014 
4 2009 4000 2012 
4 2010 20000 2012 
4 2011 17000 2012 
4 2012 6000 2012 
4 2013 0 2012 
4 2014 0 2012 
4 2015 0 2012 
end

levelsof sample, clean local(my_obs)

local graph_list ""

foreach i of local my_obs {
    sum lastyear2 if sample ==`i', meanonly
    tw line wage_year2 acadyear if sample==`i', xline(`r(mean)') name(g`i', replace)
    local graph_list "`graph_list' g`i'"
}

graph combine `graph_list'

[Erik Beecroft]

Thanks very much to Nick for a simple solution, to Dmitriy for his alternative and for letting me know about -dataex-. And thanks also to Dave for his suggestion.