How to add lines in parts of a twoway graph?

[Gunnar Ek]

Hi everyone,

I'm creating a sort of box plot using twoway graphs but I don't know how to add lines indicating if there's a significant difference between my groups. See picture below. The horizontal square brackets and asterisks (indicating differences between group 1 vs group 2-5) in the picture were added using powerpoint which isn't really optimal. Does anyone know how to do this within Stata?

The variables I'm looking at are on a continuous scale with 20-200 observations per group. I'm using the Stata command ranksum to estimate differences between groups. My code and a fake example of my data is below.

I'm using Stata v16.1 (on Mac).

Many thanks!

Gunnar
.

    sort Group
    by Group: egen med = median(Protein_1)
    by Group: egen lqt = pctile(Protein_1), p(25)
    by Group: egen uqt = pctile(Protein_1), p(75)

    twoway     (rbar lqt med Group if Group==1, barw(.5) fcolor(red) lcolor(black) ) /// 
            (rbar lqt med Group if Group==2, barw(.5) fcolor(midgreen) lcolor(black) ) ///
            (rbar lqt med Group if Group==3, barw(.5) fcolor(orange) lcolor(black) ) ///
            (rbar lqt med Group if Group==4, barw(.5) fcolor(olive_teal) lcolor(black) ) ///
            (rbar lqt med Group if Group==5, barw(.5) fcolor(midblue) lcolor(black) ) ||  ///
            (rbar med uqt Group if Group==1, barw(.5) fcolor(red) lcolor(black) ) /// 
            (rbar med uqt Group if Group==2, barw(.5) fcolor(midgreen) lcolor(black) ) ///
            (rbar med uqt Group if Group==3, barw(.5) fcolor(orange) lcolor(black) ) ///
            (rbar med uqt Group if Group==4, barw(.5) fcolor(olive_teal) lcolor(black) ) ///
            (rbar med uqt Group if Group==5, barw(.5) fcolor(midblue) lcolor(black) ) ||  ///
            (scatter Protein_1 Group if diagnosis3==1, graphregion(fcolor(gs15)) mfcolor(red) mlcolor(black) msymbol(o) legend(off)) ///
            (scatter Protein_1 Group if diagnosis3==2, graphregion(fcolor(gs15)) mfcolor(midgreen) mlcolor(black) msymbol(o) legend(off)) ///
            (scatter Protein_1 Group if diagnosis3==3, graphregion(fcolor(gs15)) mfcolor(orange) mlcolor(black) msymbol(o) legend(off)) ///
            (scatter Protein_1 Group if diagnosis3==4, graphregion(fcolor(gs15)) mfcolor(olive_teal) mlcolor(black) msymbol(o) legend(off)) ///
            (scatter Protein_1 Group if diagnosis3==5, graphregion(fcolor(gs15)) mfcolor(midblue) mlcolor(black) msymbol(o) legend(off)), ///
            xlabel(1 "Group 1" 2 "Group 2" 3 "Group 3" 4 "Group 4" 5 "Group 5", labsize(small)) ///
            xtitle("") ///
            ytitle("") ///
            yscale(log range(50 22000)) ///
            ylabel(1000 "100" 10000 "1,000" 100000 "10,000", labsize(vsmall) angle(horizontal) grid gmin gmax) ///
            graphregion(color(white)) ///
            ysize(7) xsize(5) ///
            title("A. Protein 1", position(11) size(medlarge))

Fake example of my data
clear
input float(Group Protein_1)
1 123
1 245
1 1044
1 334
1 535
1 355
1 455
2 355
2 354
2 432
2 132
2 82
end

[George Ford]

picture is not showing up. fake data does not permit a run of your program.

[Gunnar Ek]

I'm sorry. I don't know what happened. I'll try to repost the picture here.

I wish I could share my data but unfortunately I can't.

Gunnar

[Andrew Musau]

See https://www.statalist.org/forums/for...l-significance

[Gunnar Ek]

That's great! Thanks a lot, Andrew. I searched but didn't find that before. Your code from Jan 13 is indeed very helpful!

I'm using a logarithmic scale on the x axis. Would you know if there's a way to get the same visual distance between each horizontal line that I add? Now it seems I have to fiddle around to get an approximately similar distance between each line.

[Andrew Musau]

I do not understand the question. Can you present a data example illustrating the issue?

[Gunnar Ek]

Hi Andrew,

Maybe if I show you this in code and a graph.

My code is now as shown below.

With this code I get the attached graph with the new horizontal lines in the upper part of the graph. The "visual" distance in the graph between the lines decrease when moving up the y axis because I'm using a logarithmic scale on the y axis. My question is if there's a way to make this visual distance between each line the same, or if I have to manually change each graph so that the distances between each line (as well as the small vertical hooks on each horizontal line) look approximately similar?

Best
Gunnar

sort Group
    by Group: egen med = median(Protein_1)
    by Group: egen lqt = pctile(Protein_1), p(25)
    by Group: egen uqt = pctile(Protein_1), p(75)

    twoway     (rbar lqt med Group if Group==1, barw(.5) fcolor(red) lcolor(black) ) ///
            (rbar lqt med Group if Group==2, barw(.5) fcolor(midgreen) lcolor(black) ) ///
            (rbar lqt med Group if Group==3, barw(.5) fcolor(orange) lcolor(black) ) ///
            (rbar lqt med Group if Group==4, barw(.5) fcolor(olive_teal) lcolor(black) ) ///
            (rbar lqt med Group if Group==5, barw(.5) fcolor(midblue) lcolor(black) ) ||  ///
            (rbar med uqt Group if Group==1, barw(.5) fcolor(red) lcolor(black) ) ///
            (rbar med uqt Group if Group==2, barw(.5) fcolor(midgreen) lcolor(black) ) ///
            (rbar med uqt Group if Group==3, barw(.5) fcolor(orange) lcolor(black) ) ///
            (rbar med uqt Group if Group==4, barw(.5) fcolor(olive_teal) lcolor(black) ) ///
            (rbar med uqt Group if Group==5, barw(.5) fcolor(midblue) lcolor(black) ) ||  ///
            (scatter Protein_1 Group if diagnosis3==1, graphregion(fcolor(gs15)) mfcolor(red) mlcolor(black) msymbol(o) legend(off)) ///
            (scatter Protein_1 Group if diagnosis3==2, graphregion(fcolor(gs15)) mfcolor(midgreen) mlcolor(black) msymbol(o) legend(off)) ///
            (scatter Protein_1 Group if diagnosis3==3, graphregion(fcolor(gs15)) mfcolor(orange) mlcolor(black) msymbol(o) legend(off)) ///
            (scatter Protein_1 Group if diagnosis3==4, graphregion(fcolor(gs15)) mfcolor(olive_teal) mlcolor(black) msymbol(o) legend(off)) ///
            (scatter Protein_1 Group if diagnosis3==5, graphregion(fcolor(gs15)) mfcolor(midblue) mlcolor(black) msymbol(o) legend(off)), ///
            (scatteri 13000 1 13000 2, recast(line) lw(medthick) mc(none) lc(black) lp(solid)) ///
            (scatteri 13000 1 13000 2, recast(dropline) base(12500) lw(medthick) mc(none) lc(black) lp(solid)) ///
            (scatteri 11000 1 11000 3, recast(line) lw(medthick) mc(none) lc(black) lp(solid)) ///
            (scatteri 11000 1 11000 3, recast(dropline) base(10500) lw(medthick) mc(none) lc(black) lp(solid)) ///
            (scatteri 9000 1 9000 4, recast(line) lw(medthick) mc(none) lc(black) lp(solid)) ///
            (scatteri 9000 1 9000 4, recast(dropline) base(8500) lw(medthick) mc(none) lc(black) lp(solid)) ///
            (scatteri 7000 1 7000 5, recast(line) lw(medthick) mc(none) lc(black) lp(solid)) ///
            (scatteri 7000 1 7000 5, recast(dropline) base(6500) lw(medthick) mc(none) lc(black) lp(solid)), ///
            xlabel(1 "Group 1" 2 "Group 2" 3 "Group 3" 4 "Group 4" 5 "Group 5", labsize(small)) ///
            xtitle("") ///
            ytitle("") ///
            yscale(log range(500 15000)) ///
            ylabel(100 "100" 1000 "1,000", labsize(vsmall) angle(horizontal) grid gmin gmax) ///
            graphregion(color(white)) ///
            ysize(7) xsize(5)

[Andrew Musau]

Take the antilog to estimate the point. Assuming base e, you have

di log(100)
di log(1000)

Res.:

. di log(100)
4.6051702

. di log(1000)
6.9077553

If you want the next point to be 2 units up, then the question becomes what number has the log value 8.9? Last I remember, you need the exponential function to find the inverse.

di exp(8.9)

Res.:

. di exp(8.9)
7331.9735

and that is your approximate value on the axis.

[Gunnar Ek]

Thanks for a quick reply. It seems it requires some manual work to make the graphs look good. At least I don't have to use ppt

Best
Ulf