Binscatter: Ordering x axis according to size of dots.

[Calvin Wright]

Hello.
I have data similar to the following :

* Example generated by -dataex-. To install: ssc install dataex
clear
input int price float group
 4099 4
 4749 4
 3799 4
 4816 4
 7827 2
 5788 4
 4453 4
 5189 4
10372 3
 4082 4
11385 3
14500 1
15906 1
 3299 4
 5705 4
 4504 4
 5104 4
 3667 4
 3955 4
 3984 4
 4010 4
 5886 4
 6342 4
 4389 4
 4187 4
11497 3
13594 1
13466 1
 3829 4
 5379 4
 6165 4
 4516 4
 6303 4
 3291 4
 8814 2
 5172 4
 4733 4
 4890 4
 4181 4
 4195 4
10371 3
 4647 4
 4425 4
 4482 4
 6486 4
 4060 4
 5798 4
 4934 4
 5222 4
 4723 4
 4424 4
 4172 4
 9690 2
 6295 4
 9735 2
 6229 4
 4589 4
 5079 4
 8129 2
 4296 4
 5799 4
 4499 4
 3995 4
12990 1
 3895 4
 3798 4
 5899 4
 3748 4
 5719 4
 7140 2
 5397 4
 4697 4
 6850 4
11995 3
end
label values group group_label1
label def group_label1 1 "A", modify
label def group_label1 2 "B", modify
label def group_label1 3 "C", modify
label def group_label1 4 "D", modify

I am making a binned scatter plot of price and group and I want to order the xaxis according to the size/value of the bins. I want to sort the bins in descending order. So after running:

  binscatter price group, xlabel(1 "A" 2 "B" 3 "C" 4 "D",  noticks)

I want the axis to be ordered A C B D and not A B C D

.

[Nick Cox]

binscatter is a community-contributed command, so by FAQ Advice #12, you are asked to say where it comes from.

I know that many people find it useful, and it may well offer a solution, but the task here of calculating and plotting means is also possible directly. The twist of getting the groups in order of the means was discussed in detail earlier today. See https://www.statalist.org/forums/for...occupation-sum

Here is your graph -- and what might possibly be considered an even more helpful one, showing each distribution too.

clear
input int price float group
 4099 4
 4749 4
 3799 4
 4816 4
 7827 2
 5788 4
 4453 4
 5189 4
10372 3
 4082 4
11385 3
14500 1
15906 1
 3299 4
 5705 4
 4504 4
 5104 4
 3667 4
 3955 4
 3984 4
 4010 4
 5886 4
 6342 4
 4389 4
 4187 4
11497 3
13594 1
13466 1
 3829 4
 5379 4
 6165 4
 4516 4
 6303 4
 3291 4
 8814 2
 5172 4
 4733 4
 4890 4
 4181 4
 4195 4
10371 3
 4647 4
 4425 4
 4482 4
 6486 4
 4060 4
 5798 4
 4934 4
 5222 4
 4723 4
 4424 4
 4172 4
 9690 2
 6295 4
 9735 2
 6229 4
 4589 4
 5079 4
 8129 2
 4296 4
 5799 4
 4499 4
 3995 4
12990 1
 3895 4
 3798 4
 5899 4
 3748 4
 5719 4
 7140 2
 5397 4
 4697 4
 6850 4
11995 3
end
label values group group_label1
label def group_label1 1 "A", modify
label def group_label1 2 "B", modify
label def group_label1 3 "C", modify
label def group_label1 4 "D", modify

egen mean = mean(-price), by(group)
egen group2 = group(mean)

* labmask is from the Stata Journal
labmask group2, values(group) decode

label var group2 " "
set scheme s1color
replace mean = -mean
scatter mean group2, xla(1/4, valuelabel) name(G1, replace)

* stripplot is from SSC; for an overview see https://www.statalist.org/forums/forum/general-stata-discussion/general/209911-stripplot-updated-on-ssc
stripplot price, over(group2) cumul cumprob refline vertical name(G2, replace)

Note: the additional option centre (or even center) will place each point cluster display in the second graph directly over the axis label below.

[Calvin Wright]

Thank you, Nick! That works perfectly on my dataset. stripplot from SSC is also quite useful. A follow up query: Let's say I wanted to show the mean price by(foreign). So for A, there would be a dot for foreign and a dot for domestic and so on. The axis would still be ordered on the mean for only one of the groups, say foreign =0.

* Example generated by -dataex-. To install: ssc install dataex
clear
input byte foreign int price float group
0  4099 4
0  4749 4
0  3799 4
0  4816 4
0  7827 2
0  5788 4
0  4453 4
0  5189 4
0 10372 3
0  4082 4
0 11385 3
0 14500 1
0 15906 1
0  3299 4
0  5705 4
0  4504 4
0  5104 4
0  3667 4
0  3955 4
0  3984 4
0  4010 4
0  5886 4
0  6342 4
0  4389 4
0  4187 4
0 11497 3
0 13594 1
0 13466 1
0  3829 4
0  5379 4
0  6165 4
0  4516 4
0  6303 4
0  3291 4
0  8814 2
0  5172 4
0  4733 4
0  4890 4
0  4181 4
0  4195 4
0 10371 3
0  4647 4
0  4425 4
0  4482 4
0  6486 4
0  4060 4
0  5798 4
0  4934 4
0  5222 4
0  4723 4
0  4424 4
0  4172 4
1  9690 2
1  6295 4
1  9735 2
1  6229 4
1  4589 4
1  5079 4
1  8129 2
1  4296 4
1  5799 4
1  4499 4
1  3995 4
1 12990 1
1  3895 4
1  3798 4
1  5899 4
1  3748 4
1  5719 4
1  7140 2
1  5397 4
1  4697 4
1  6850 4
1 11995 3
end
label values foreign origin
label def origin 0 "Domestic", modify
label def origin 1 "Foreign", modify
label values group group_label1
label def group_label1 1 "A", modify
label def group_label1 2 "B", modify
label def group_label1 3 "C", modify
label def group_label1 4 "D", modify

The "by" option isn't allowed with scatter and I unsuccessfully tried to overlay two different graphs because of how we defined the groups.

[Calvin Wright]

HI Nick, I think your most recent response was deleted while the server was down. Do you have any insight on my follow up question above?

[Nick Cox]

Indeed. I don't have a record of what I said, but it was probably similar to this.

The "by" option isn't allowed with scatter

Not so. Perhaps what you mean is that you tried to use it but got an error for some other reason.

However, starting from your example data in #3 I did this. There are many other ways of showing the same or similar results using variations on options over() by() separate() in stripplot.

* for more on this see section 9 in https://www.stata-journal.com/article.html?article=dm0055
egen mean = mean(cond(foreign == 0, -price, .)), by(group)

egen group2 = group(mean)

* labmask is from the Stata Journal
labmask group2, values(group) decode

set scheme s1color 

* stripplot is from SSC; for an overview see https://www.statalist.org/forums/forum/general-stata-discussion/general/209911-stripplot-updated-on-ssc
stripplot price, over(foreign) by(group2, legend(off) row(1) note("") compact) xtitle("") centre cumul cumprob refline vertical separate(foreign)

My main point is that wanting to show the means -- in your examples so far 4 or 8 of them -- is fair enough, but you have enough space to show more detail without serious risk of confusion.