-grc1leg- not omitting legend of component graphs

[Veridiana Nogueira]

Hello,

I'm having problems in generating a combined graph with a single common legend, as it is the purpose of -grc1leg-. I've tried -grc1leg2- without identical results. I've also tried producing subgraphs with the -by(, legend(off))- option, but it gave me not legends anywhere.

In another post here, it seems like Nick Cox suggests -grc1leg- does not handle subgraphs with the -by()- option. Is that the case? Any way around this issue?

I'm using Stata/IC 14.2 on Windows 10. Please find my code below.

* Example generated by -dataex-. To install: ssc install dataex
clear
input double(order_0th order_1st order_2nd order_3rd) str3 Country float E_int
.7713897514162157 .8983225631889338 .9502321908700299 .9751999295525718 "KOR" 1
.7728646528002182 .9383212578278355 .9819118726638241 .9942301335562106 "IRL" 1
.6520159877677851 .9121377064098004 .9747471517401809 .9917984486988558 "IRL" 0
.6120832068748473 .8152489201542453 .9045947423296314 .9509950438312654 "KOR" 1
.7631853372424123 .9394807242320112 .9829067108733993 .9945699072222791 "IRL" 1
 .903293987668906 .9781866263244458 .9936282476232506 .9978061162704196 "IRL" 1
 .865287920308868 .9622437133705122 .9871829259446699 .9952220573997735 "IRL" 0
.9999999999999248 .9999999999999863 .9999999999999963 .9999999999999989 "IRL" 1
.6792053150007263 .8063989295123989 .8937131164157297 .9441320938002318 "KOR" 1
.9197390025497783 .9779566782958434 .9927579596463956 .9973200689783972 "IRL" 1
.6365057955352291 .9023851889266424 .9734745091297393 .9922561223846332 "IRL" 1
.4225068227500901  .788998764115341 .9046037611433007 .9523722898556336 "KOR" 0
.9990615485362677 .9995811977144247 .9997887355023127 .9998918138706242 "KOR" 1
.9182802263367715 .9780545049436485 .9940366116334217  .998259042597569 "IRL" 1
.8725685337577995 .9672674068354333   .99025852065573 .9967693513075605 "IRL" 1
.9578451533226735 .9907174081433222 .9971845159005579 .9989958860915611 "IRL" 1
.8846973398371609 .9720648024371774 .9921325686225014 .9974391757004147 "IRL" 1
.7335101993068649 .8835785734219438 .9443981517357872 .9724995644900749 "KOR" 1
.8821281102440207 .9659512714005644  .988734623242631 .9959272584518145 "IRL" 1
.9466007799343315 .9868004469516986 .9957800546785351 .9984931323030795 "IRL" 1
end
label values E_int E_int_label
label def E_int_label 0 "Capital Intensive", modify
label def E_int_label 1 "Labour Intensive", modify

foreach v of varlist order_0th order_1st order_2nd order_3rd {
label variable `v' "mean"
statplot `v', over(E_int) by(Country, note("")) recast(bar) asyvars yla(, ang(h)) ///
    bar(2, bcolor(forest_green)) bar(1, bcolor(maroon)) ///
     ytitle(`v') xsize(15) ysize(8) ///
     by(, graphregion(fcolor(white) lcolor(white))) scheme(s2color) name(`v', replace)
}
*
grc1leg2 order_0th order_1st order_2nd order_3rd

Best regards,
Veridiana

P.S.: I've benefitted from the Statalist Forum for years, but this is my first post. I hope I made it right, but please let me know if I can improve something. I'm really grateful to you all, especially to Nick Cox, who put some much time in giving us valuable tools and tricks.

[Andrew Musau]

grc1leg2 is from http://digital.cgdev.org/doc/stata/MO/Misc/ and is authored by Mead Over (FAQ Advice #12).

net install grc1leg2.pkg, from (http://digital.cgdev.org/doc/stata/MO/Misc/)

In another post here, it seems like Nick Cox suggests -grc1leg- does not handle subgraphs with the -by()- option. Is that the case? Any way around this issue?

I don't think there has been a resolution to this, but see William Lisowski's suggested workaround from #10 of the linked thread below.

https://www.statalist.org/forums/for...pear-graphs-by

[Nick Cox]

Here's a different take. You want 2 x 4 x 2 = 16 means. One design avoids legends altogether with

countries as panels

capital and labour intensive as separate variables

different order as an axis variable.

I also use C and L as more nearly self-explanatory marker labels, avoid bars and starting at 0, and most eccentrically of all a folded root scale sqrt(p) - sqrt(1 - p), a weaker cousin of logit p = log p - log (1 - p), to stretch values near 1 suitably. Logit scale does even better, but I still prefer to show 1 as an axis label.

* Example generated by -dataex-. To install: ssc install dataex
clear
input double(order_0th order_1st order_2nd order_3rd) str3 Country float E_int
.7713897514162157 .8983225631889338 .9502321908700299 .9751999295525718 "KOR" 1
.7728646528002182 .9383212578278355 .9819118726638241 .9942301335562106 "IRL" 1
.6520159877677851 .9121377064098004 .9747471517401809 .9917984486988558 "IRL" 0
.6120832068748473 .8152489201542453 .9045947423296314 .9509950438312654 "KOR" 1
.7631853372424123 .9394807242320112 .9829067108733993 .9945699072222791 "IRL" 1
 .903293987668906 .9781866263244458 .9936282476232506 .9978061162704196 "IRL" 1
 .865287920308868 .9622437133705122 .9871829259446699 .9952220573997735 "IRL" 0
.9999999999999248 .9999999999999863 .9999999999999963 .9999999999999989 "IRL" 1
.6792053150007263 .8063989295123989 .8937131164157297 .9441320938002318 "KOR" 1
.9197390025497783 .9779566782958434 .9927579596463956 .9973200689783972 "IRL" 1
.6365057955352291 .9023851889266424 .9734745091297393 .9922561223846332 "IRL" 1
.4225068227500901  .788998764115341 .9046037611433007 .9523722898556336 "KOR" 0
.9990615485362677 .9995811977144247 .9997887355023127 .9998918138706242 "KOR" 1
.9182802263367715 .9780545049436485 .9940366116334217  .998259042597569 "IRL" 1
.8725685337577995 .9672674068354333   .99025852065573 .9967693513075605 "IRL" 1
.9578451533226735 .9907174081433222 .9971845159005579 .9989958860915611 "IRL" 1
.8846973398371609 .9720648024371774 .9921325686225014 .9974391757004147 "IRL" 1
.7335101993068649 .8835785734219438 .9443981517357872 .9724995644900749 "KOR" 1
.8821281102440207 .9659512714005644  .988734623242631 .9959272584518145 "IRL" 1
.9466007799343315 .9868004469516986 .9957800546785351 .9984931323030795 "IRL" 1
end
label values E_int E_int_label
label def E_int_label 0 "Capital Intensive", modify
label def E_int_label 1 "Labour Intensive", modify

collapse (mean) order_*, by(Country E_int)

reshape long order, i(Country E_int) j(which) string
rename (order which) (whatever order)
gen x = real(substr(order, 2, 1)) 
gen toshow = sqrt(whatever) - sqrt(1 - whatever)
separate toshow, by(E_int) veryshortlabel 

* mylabels is from SSC 
mylabels 0.4(0.1)0.9, myscale(sqrt(@) - sqrt(1 - @)) prefix("0") local(yla)

gen text = cond(E_int == 0, "C", "L")  
set scheme s1color 
scatter toshow? x, by(Country, note("") legend(off)) xsc(r(-0.3 3.3)) xla(, grid glcolor(gs12) glw(vthin)) ms(none ..) mlabc(blue red) mlabsize(medium ..) mlabpos(0 0) mla(text text) yla(`yla' 1 "1", ang(h)) ytitle(whatever) xtitle(order)

[Veridiana Nogueira]

grc1leg2 is from http://digital.cgdev.org/doc/stata/MO/Misc/ and is authored by Mead Over (FAQ Advice #12).

I see your point. The link is broken, but thank you.

[Veridiana Nogueira]

Here's a different take. You want 2 x 4 x 2 = 16 means. One design avoids legends altogether with

countries as panels

capital and labour intensive as separate variables

different order as an axis variable.

I was just thinking about something on these lines while waiting for replies, but not sure how to put that in practice. Thank you for the code!

[William Lisowski]

Regarding the link in post #2 quoted in post #4: although it is not possible to open the link

http://digital.cgdev.org/doc/stata/MO/Misc

in a web browser, it is possible to issue the commands

net from http://digital.cgdev.org/doc/stata/MO/Misc
net describe grc1leg2, from(http://digital.cgdev.org/doc/stata/MO/Misc)
net install grc1leg2.pkg, from(http://digital.cgdev.org/doc/stata/MO/Misc/)

to access the directory for that user site, to describe grc1leg2, and to install grc1leg2.

[Mead Over]

I agree with Nick Cox that embedding a multi-panel graph (such as those created by graph ..., by() or by graph combine) into another multi-panel graph is likely to produce a result that is hard for the reader to interpret. But since there may be some circumstances where such recursive multiplicity is desireable, I have revised grc1leg2 so that it now accommodates component panels which have been produced by graph ..., by() or by graph combine. Using version 1.4 of grc1leg2, the following code:

* Example generated by -dataex-. To install: ssc install dataex
cap program drop grc1leg2
graph drop _all
clear
input double(order_0th order_1st order_2nd order_3rd) str3 Country float E_int
.7713897514162157 .8983225631889338 .9502321908700299 .9751999295525718 "KOR" 1
.7728646528002182 .9383212578278355 .9819118726638241 .9942301335562106 "IRL" 1
.6520159877677851 .9121377064098004 .9747471517401809 .9917984486988558 "IRL" 0
.6120832068748473 .8152489201542453 .9045947423296314 .9509950438312654 "KOR" 1
.7631853372424123 .9394807242320112 .9829067108733993 .9945699072222791 "IRL" 1
 .903293987668906 .9781866263244458 .9936282476232506 .9978061162704196 "IRL" 1
 .865287920308868 .9622437133705122 .9871829259446699 .9952220573997735 "IRL" 0
.9999999999999248 .9999999999999863 .9999999999999963 .9999999999999989 "IRL" 1
.6792053150007263 .8063989295123989 .8937131164157297 .9441320938002318 "KOR" 1
.9197390025497783 .9779566782958434 .9927579596463956 .9973200689783972 "IRL" 1
.6365057955352291 .9023851889266424 .9734745091297393 .9922561223846332 "IRL" 1
.4225068227500901  .788998764115341 .9046037611433007 .9523722898556336 "KOR" 0
.9990615485362677 .9995811977144247 .9997887355023127 .9998918138706242 "KOR" 1
.9182802263367715 .9780545049436485 .9940366116334217  .998259042597569 "IRL" 1
.8725685337577995 .9672674068354333   .99025852065573 .9967693513075605 "IRL" 1
.9578451533226735 .9907174081433222 .9971845159005579 .9989958860915611 "IRL" 1
.8846973398371609 .9720648024371774 .9921325686225014 .9974391757004147 "IRL" 1
.7335101993068649 .8835785734219438 .9443981517357872 .9724995644900749 "KOR" 1
.8821281102440207 .9659512714005644  .988734623242631 .9959272584518145 "IRL" 1
.9466007799343315 .9868004469516986 .9957800546785351 .9984931323030795 "IRL" 1
end
label values E_int E_int_label
label def E_int_label 0 "Capital Intensive", modify
label def E_int_label 1 "Labour Intensive", modify

set graph off
foreach v of varlist order_0th order_1st order_2nd order_3rd {
label variable `v' "mean"
statplot `v', over(E_int) by(Country, subtitle(`v') note("")) recast(bar) asyvars yla(, ang(h)) ///
    bar(2, bcolor(forest_green)) bar(1, bcolor(maroon)) ///
     ytitle(mean proportion) xsize(15) ysize(8) ///
     ylabel(0(.2)1)  ///
     by(, graphregion(fcolor(white) lcolor(white))) scheme(s2color) name(`v', replace)
}
set graph on
*

grc1leg2 order_0th order_1st order_2nd order_3rd, name(combined, replace) l1tol1title

produces the following result.

I have taken the liberty of moving the panel labels from the y-axis to the subtitles and adding the y-title "mean proportion" which I infer might be the measure captured by the bars. This change, allows me to demonstrate how grc1leg2 can reduce clutter with the option l1tol1title, which suppresses the redundant y-axis labels and moves one of them to the common "l1" axis label so that it applies to all panels. (The option ytol1title, used when component graphs are single panels, does not work here because the y-axis title produced by the command statplot, and indeed by many multipanel graph commands, is located in l1title, not in ytitle.)

Instead of using the option ycommon to line up the axes of the four panels, I specified the option ylabel(0(.2)1) for the statplot command.

As mentioned by @William Lisowsky, the program grc1leg2 can be found and installed from inside Stata with

net describe grc1leg2, from(http://digital.cgdev.org/doc/stata/MO/Misc)

Some users might prefer to type

view net describe grc1leg2, from(http://digital.cgdev.org/doc/stata/MO/Misc)

in order to see the package description and clickable links in their viewer instead of in their results window.

[Veridiana Nogueira]

I agree with Nick Cox that embedding a multi-panel graph (such as those created by graph ..., by() or by graph combine) into another multi-panel graph is likely to produce a result that is hard for the reader to interpret. But since there may be some circumstances where such recursive multiplicity is desireable, I have revised grc1leg2 so that it now accommodates component panels which have been produced by graph ..., by() or by graph combine.

I've also decided to change the plot in this case. I initially had all industries. When I decided to go for group means, the problem with the legend stuck me. Nevertheless, your fixture certainly might be useful for me and others in the future.

I believe this concludes my enquiry. Thank you for all who contributed.