Boxplot with whiskers go to actual min and max instead of 1.5 IQR

[Yapeng Wang]

Hello, I am using Stata 16.1 (MP). I would like to create a boxplot but with whiskers going to actual min and max instead of 1.5 IQR. I was wondering if I could do that and how. Thanks!

This is the graph I have for now. In the graph, you can see that all three variables' lower whiskers go to 1.5 IQR instead of the actual min: 2, 1.6, and 1.8. I was wondering if Stata could make the lower whiskers reach the actual min.

Here is the example data.

* Example generated by -dataex-. To install: ssc install dataex
clear
input byte ID double(var1 var2 var3)
 1  3.142857074737549                  3             3.0625
 2                  3  2.444444417953491             2.6875
 3  4.285714149475098                  4              4.125
 4 4.5714287757873535                  . 4.5714287757873535
 5  4.285714149475098  3.444444417953491             3.8125
 6 3.2857143878936768  3.444444417953491              3.375
 7 4.5714287757873535  4.666666507720947              4.625
 8                  2 1.5555555820465088               1.75
 9  4.285714149475098  3.222222328186035             3.6875
10 4.5714287757873535                  4               4.25
11  4.142857074737549  3.222222328186035              3.625
12 4.4285712242126465 3.6666667461395264                  4
13 4.4285712242126465  3.555555582046509             3.9375
14  4.285714149475098  3.444444417953491             3.8125
15  4.142857074737549 3.6666667461395264              3.875
16  4.285714149475098  4.111111164093018             4.1875
17  4.142857074737549                  4             4.0625
18 4.5714287757873535 3.8888888359069824             4.1875
19  4.285714149475098  4.111111164093018             4.1875
20 4.4285712242126465  3.555555582046509             3.9375
21  3.142857074737549 2.8888888359069824                  3
end

[Nick Cox]

One way to get this is through stripplot from SSC.

That said, while I tend to avoid the whiskers to at most 1.5 IQR rule, there is much scope here, and almost everywhere, to show informative and possibly important detail about the data as well as a minimal box plot. It is surprisingly common to find that researchers overlook the elementary but fundamental point that if half the values lie inside each box, then the other half lies outside and may need very detailed attention.

Here var1 is a case in point, as the short upper tail deserves a detailed scrutiny that no box plot will give it.


Further, the unnecessary legend can be removed in favour of more direct labelling.

* Example generated by -dataex-. To install: ssc install dataex
clear
input byte ID double(var1 var2 var3)
 1  3.142857074737549                  3             3.0625
 2                  3  2.444444417953491             2.6875
 3  4.285714149475098                  4              4.125
 4 4.5714287757873535                  . 4.5714287757873535
 5  4.285714149475098  3.444444417953491             3.8125
 6 3.2857143878936768  3.444444417953491              3.375
 7 4.5714287757873535  4.666666507720947              4.625
 8                  2 1.5555555820465088               1.75
 9  4.285714149475098  3.222222328186035             3.6875
10 4.5714287757873535                  4               4.25
11  4.142857074737549  3.222222328186035              3.625
12 4.4285712242126465 3.6666667461395264                  4
13 4.4285712242126465  3.555555582046509             3.9375
14  4.285714149475098  3.444444417953491             3.8125
15  4.142857074737549 3.6666667461395264              3.875
16  4.285714149475098  4.111111164093018             4.1875
17  4.142857074737549                  4             4.0625
18 4.5714287757873535 3.8888888359069824             4.1875
19  4.285714149475098  4.111111164093018             4.1875
20 4.4285712242126465  3.555555582046509             3.9375
21  3.142857074737549 2.8888888359069824                  3
end

set scheme s1color 

local what ytitle(some better text should go here -- or nothing)

stripplot var* , yla(, ang(h)) xla(, noticks) ms(none) box pctile(0) vertical `what' name(G1, replace)

stripplot var* , yla(, ang(h)) xla(, noticks) cumul height(0.6) box(barw(0.1)) boffset(-0.1) pctile(0) vertical `what' name(G2, replace)

[Nick Cox]

See also the concurrent thread https://www.statalist.org/forums/for...-plot-in-stata

[Yapeng Wang]

One way to get this is through stripplot from SSC.

That said, while I tend to avoid the whiskers to at most 1.5 IQR rule, there is much scope here, and almost everywhere, to show informative and possibly important detail about the data as well as a minimal box plot. It is surprisingly common to find that researchers overlook the elementary but fundamental point that if half the values lie inside each box, then the other half lies outside and may need very detailed attention.

Here var1 is a case in point, as the short upper tail deserves a detailed scrutiny that no box plot will give it.


Further, the unnecessary legend can be removed in favour of more direct labelling.

* Example generated by -dataex-. To install: ssc install dataex
clear
input byte ID double(var1 var2 var3)
1 3.142857074737549 3 3.0625
2 3 2.444444417953491 2.6875
3 4.285714149475098 4 4.125
4 4.5714287757873535 . 4.5714287757873535
5 4.285714149475098 3.444444417953491 3.8125
6 3.2857143878936768 3.444444417953491 3.375
7 4.5714287757873535 4.666666507720947 4.625
8 2 1.5555555820465088 1.75
9 4.285714149475098 3.222222328186035 3.6875
10 4.5714287757873535 4 4.25
11 4.142857074737549 3.222222328186035 3.625
12 4.4285712242126465 3.6666667461395264 4
13 4.4285712242126465 3.555555582046509 3.9375
14 4.285714149475098 3.444444417953491 3.8125
15 4.142857074737549 3.6666667461395264 3.875
16 4.285714149475098 4.111111164093018 4.1875
17 4.142857074737549 4 4.0625
18 4.5714287757873535 3.8888888359069824 4.1875
19 4.285714149475098 4.111111164093018 4.1875
20 4.4285712242126465 3.555555582046509 3.9375
21 3.142857074737549 2.8888888359069824 3
end

set scheme s1color

local what ytitle(some better text should go here -- or nothing)

stripplot var* , yla(, ang(h)) xla(, noticks) ms(none) box pctile(0) vertical `what' name(G1, replace)

stripplot var* , yla(, ang(h)) xla(, noticks) cumul height(0.6) box(barw(0.1)) boffset(-0.1) pctile(0) vertical `what' name(G2, replace)

[ATTACH=CONFIG]n1708954[/ATTACH] [ATTACH=CONFIG]n1708955[/ATTACH]

Dear Nick: thank you so much! Could I ask a follow-up question? Is it possible to add caps at the end of whiskers, like the graph box produced by Stata as follows:

[Nick Cox]

Add the option to my code

whiskers(recast(rcap))

and to change the cap length add

whiskers(recast(rcap) msize(*2))

or whatever.

[Yapeng Wang]

Add the option to my code

whiskers(recast(rcap))

and to change the cap length add

whiskers(recast(rcap) msize(*2))

or whatever.

This works! Thank you so much, Nick!