Stacked bar graph (stacking by group, rather than all)

[Jad Tamimi]

Hello I have the following data,

* Example generated by -dataex-. For more info, type help dataex
clear
input int year float(emp1 emp2 emp3 emp4)
2000  8.287  349.506   62.41  456.367
2001  7.862  427.525  69.546  598.399
2002 16.165  611.199  75.528  744.512
2003 11.914  777.101  87.031  897.903
2004 18.107  944.163  83.388 1076.389
2005 22.732 1100.708 115.292 1210.571
2006 20.412 1316.654 134.195 1365.223
2007 31.985 1526.965 168.284 1547.406
2008 25.674   1658.5 154.205 1680.176
2009 21.161 1739.472 140.922   1790.8
2010 28.375 1957.499 217.189 1889.972
2011 43.873 2267.242 167.622 2083.245
2012 32.195 2429.847 146.445 2084.692
2013 39.729  2697.61 124.509 2163.135
2014 32.863 3020.331 147.221 2248.292
2015 27.374 3376.271 151.724 2393.775
2016 39.595 3708.233 108.538 2611.653
end

I'm creating the following bar graph:

* draw graph for first decomposition (continuers & exiters)
graph bar emp1 emp2 emp3 emp4, over(year,label(labsize(small) angle(45)) gap(20)) stack ///
    bar(1, color(navy*0.8)) bar(2, color(navy)) bar(3, color(green)) bar(4, color(dkgreen)) graphregion(color(white)) ///
    ytitle("Manufacturing Employment (k workers)") ///
    legend(on label(1 "Foreign exiters") label(2 "Foreign continuers") ///
    label(3 "Private exiters") label(4 "Private continuers") ///
    region(lstyle(foreground)) rows(1) position(6) span) outergap(10)

I want to stack some data, hence why I included the stack option, however I only want to stack 'emp1' and 'emp2' together and seperately 'emp3' and 'emp4' together. So two stacked bars in each year. Not all four of them stacked together as currently, which is producing only one stacked bar in each year.
Any ideas how to implement this?

Thanks,
Jad

[Jad Tamimi]

Also I want to keep the current dataset format, so I want 4 different series for 4 different legends, really I want my graph to look exactly like this version here:

graph bar emp1 emp3 , over(year,label(labsize(small) angle(45)) gap(20)) ///
bar(1, color(navy)) bar(2, color(dkgreen)) graphregion(color(white)) ///
ytitle("Manufacturing Employment (k workers)") ///
legend(on label(1 "Foreign exiters") ///
label(2 "Private exiters") ///
region(lstyle(foreground)) rows(1) position(6) span) outergap(10)

With the only difference being I now have an additional bar stacked onto each of those two and I also have 4 legend options instead of two.

Please let me know if its possible to do this on Stata.

Thanks,
Jad

[Andrew Musau]

I wouldn't insist on a stacked design here, especially since it appears that you are stacking emp1 and emp3 as well as emp2 and emp4, and these variables have very different magnitudes. In stacked designs, small quantities are difficult to see. See #3 by Nick Cox for drawbacks relating to a stacked design: https://www.statalist.org/forums/for...cked-bar-chart. For an alternative, see tabplot from the Stata Journal.

search tablot

In terms of how you create a stacked bar, you need to switch to twoway bar. Here is an example that stacks emp1 and emp2 and emp3 and emp4, unlike your example.

* Example generated by -dataex-. For more info, type help dataex
clear
input int year float(emp1 emp2 emp3 emp4)
2000  8.287  349.506   62.41  456.367
2001  7.862  427.525  69.546  598.399
2002 16.165  611.199  75.528  744.512
2003 11.914  777.101  87.031  897.903
2004 18.107  944.163  83.388 1076.389
2005 22.732 1100.708 115.292 1210.571
2006 20.412 1316.654 134.195 1365.223
2007 31.985 1526.965 168.284 1547.406
2008 25.674   1658.5 154.205 1680.176
2009 21.161 1739.472 140.922   1790.8
2010 28.375 1957.499 217.189 1889.972
2011 43.873 2267.242 167.622 2083.245
2012 32.195 2429.847 146.445 2084.692
2013 39.729  2697.61 124.509 2163.135
2014 32.863 3020.331 147.221 2248.292
2015 27.374 3376.271 151.724 2393.775
2016 39.595 3708.233 108.538 2611.653
end



gen x1= year-2000-.25
gen x2= year-2000+.25
gen emp2s = emp1+emp2
gen emp4s= emp3+emp4
twoway (bar emp2s emp1 x1, barwidth(0.4 ..)) ///
 (bar emp4s emp3 x2, barwidth(0.4 ..)), ///
 xtitle(2000s) xlab(0/16, format(%02.0f)noticks) ///
 leg(order(2 "emp1" 1 "emp2" 4 "emp3" 3 "emp4")) ///
 plotregion(margin(zero))

[Nick Cox]

A better graph might follow from knowing what emp1 emp2 emp3 emp4 are. (Surely any readers of this graph need to be told?)

For example emp1 + emp2 might be male employment and emp3 + emp4 might be female employment. In that case a more complicated display but one closer to what is wanted would compare the two totals

emp1 + emp2

emp3 + emp4

and then compare also

emp1 as a fraction of its total

emp3 as a fraction of its total.

That would be best done in two aligned line charts.

As Andrew Musau underlines a key limitation of stacked bar charts is that small amounts are hard to see. Also, it's often a toss-up on whether we want to see absolute or relative amounts.


#1 and #2 would have been clearer if the graphs had been shown as ..png attachments.

[Nick Cox]

Here is the idea of #4 developed.

#4 was unfair insofar as some explanation in terms exiters and continuers is included in the code.

* Example generated by -dataex-. For more info, type help dataex
clear
input int year float(emp1 emp2 emp3 emp4)
2000  8.287  349.506   62.41  456.367
2001  7.862  427.525  69.546  598.399
2002 16.165  611.199  75.528  744.512
2003 11.914  777.101  87.031  897.903
2004 18.107  944.163  83.388 1076.389
2005 22.732 1100.708 115.292 1210.571
2006 20.412 1316.654 134.195 1365.223
2007 31.985 1526.965 168.284 1547.406
2008 25.674   1658.5 154.205 1680.176
2009 21.161 1739.472 140.922   1790.8
2010 28.375 1957.499 217.189 1889.972
2011 43.873 2267.242 167.622 2083.245
2012 32.195 2429.847 146.445 2084.692
2013 39.729  2697.61 124.509 2163.135
2014 32.863 3020.331 147.221 2248.292
2015 27.374 3376.271 151.724 2393.775
2016 39.595 3708.233 108.538 2611.653
end

gen toshow1 = emp1 + emp2
gen toshow2 = emp3 + emp4
gen pc1 = 100 * emp1 / toshow1
gen pc2 = 100 * emp3 / toshow2

set scheme s1color

line toshow? year, lc(blue red) name(G1, replace) xsc(off) fysize(60) ///
|| scatteri 3750 2013 "emp1 + emp2", ms(none) mlabcolor(blue) mlabsize(medlarge) ///
|| scatteri 2200 2013 "emp3 + emp4", ms(none) mlabcolor(red) mlabsize(medlarge) ///
legend(off)

line pc? year, lc(blue red) name(G2, replace) fysize(40) ///
|| scatteri 7.5  2013 "emp1 as %", ms(none) mlabcolor(red) mlabsize(medlarge) ///
|| scatteri 2.5 2013 "emp3 as %", ms(none) mlabcolor(blue) mlabsize(medlarge) ///
legend(off) xla(2000(4)2016)

graph combine G1 G2, col(1) xcommon imargin(vsmall)

[Jad Tamimi]

Thanks! I decided to opt for this option:

* Example generated by -dataex-. For more info, type help dataex
clear
input int year float(emp1 emp2 emp3 emp4)
2000  8.287  349.506   62.41  456.367
2001  7.862  427.525  69.546  598.399
2002 16.165  611.199  75.528  744.512
2003 11.914  777.101  87.031  897.903
2004 18.107  944.163  83.388 1076.389
2005 22.732 1100.708 115.292 1210.571
2006 20.412 1316.654 134.195 1365.223
2007 31.985 1526.965 168.284 1547.406
2008 25.674   1658.5 154.205 1680.176
2009 21.161 1739.472 140.922   1790.8
2010 28.375 1957.499 217.189 1889.972
2011 43.873 2267.242 167.622 2083.245
2012 32.195 2429.847 146.445 2084.692
2013 39.729  2697.61 124.509 2163.135
2014 32.863 3020.331 147.221 2248.292
2015 27.374 3376.271 151.724 2393.775
2016 39.595 3708.233 108.538 2611.653
end

gen x1 = year-0.225
gen x2 = year+0.225
gen emp2_ = emp1+emp2 
gen emp4_= emp3+emp4 

twoway (bar emp2_ emp1 x1, barwidth(0.45 ..) color(dknavy navy) lwidth(none none)) ///
 (bar emp4_ emp3 x2, barwidth(0.45 ..) bcolor(dkgreen green) lwidth(none none)), plotregion(margin(zero)) ///
 xlabel(2000(1)2016, labsize(small) angle(45) noticks labgap(tiny) nogrid) ylabel(, angle(vertical) format(%9.0gc) gstyle(solid)) ///
 ysc(r(0 4175)) xsc(r(1999 2017)) ytitle("Manufacturing Employment (k workers)") ///
 legend(on order(2 "Foreign exiters" 1 "Foreign continuers" 4 "Private exiters" 3 "Private continuers") ///
region(lstyle(foreground)) rows(1) position(6) span)

[Jad Tamimi]

I feel as though there should be easier ways to make this bar graph.

[Nick Cox]

it’s a common feeling: all I want is a relatively simple bar chart!

It is easy enough to look at the graph and understand the principle. What’s biting is that Stata still needs to be told exactly what to put where.

You could call this anything you like — say a twinned double stacked bar chart — and write a dedicated command but it’s hard for me to imagine that the syntax would be much simpler than the direct syntax.

i don’t see that the stacked design can be used to read anything but gross trends.

[Nick Cox]

Here's a tabplot solution as mentioned by Andrew Musau

* Example generated by -dataex-. For more info, type help dataex
clear
input int year float(emp1 emp2 emp3 emp4)
2000  8.287  349.506   62.41  456.367
2001  7.862  427.525  69.546  598.399
2002 16.165  611.199  75.528  744.512
2003 11.914  777.101  87.031  897.903
2004 18.107  944.163  83.388 1076.389
2005 22.732 1100.708 115.292 1210.571
2006 20.412 1316.654 134.195 1365.223
2007 31.985 1526.965 168.284 1547.406
2008 25.674   1658.5 154.205 1680.176
2009 21.161 1739.472 140.922   1790.8
2010 28.375 1957.499 217.189 1889.972
2011 43.873 2267.242 167.622 2083.245
2012 32.195 2429.847 146.445 2084.692
2013 39.729  2697.61 124.509 2163.135
2014 32.863 3020.331 147.221 2248.292
2015 27.374 3376.271 151.724 2393.775
2016 39.595 3708.233 108.538 2611.653
end


reshape long emp, i(year) j(which)
label def which 1 "Foreign exiters" 2 "Foreign continuers" 3 "Private exiters" 4 "Private continuers"
label val which which

forval y = 2000/2016 {
    local this : di %tyYY `y'
    label def year `y' "`this'", add
}
label val year year
label var year "year 20.."

tabplot which year [iw=emp], showval(format(%1.0f)) barw(0.8) separate(which) subtitle("000") ytitle("")

[Nick Cox]

Anyone bemused by the minor trickery to get year 2000 to 2016 showing as 00 to 16 -- using value labels defined for the purpose -- may be reassured to learn that there is another way to do it.

label var year "year 20.."

tabplot which year [iw=emp], xasis xla(, format(%tyYY)) showval(format(%1.0f)) barw(0.8) separate(which) subtitle("000") ytitle("")