Stripplot labelling problem

[Lyle Scruggs]

country warming humans bad
AT 90 60 55
BE 96 53 56
CH 96 44 59
CZ 86 35 48
DE 95 54 63
EE 90 34 45
ES 94 57 75
FI 94 51 52
FR 96 47 62
GB 93 36 52
HU 88 45 61
IE 95 38 50
IS 97 55 68
IT 91 55 53
LT 87 30 49
NL 96 44 44
NO 93 38 55
PL 90 30 52
PT 96 48 70
RU 76 28 37
SE 97 51 64
SI 96 38 59
E~R 93 45 58

With the above data, I can produce a stripplot with label for 1 variable but not two: E.g.,

. stripplot humans , vert mlabel( country )


delivers :





But

.stripplot humans bad , vert mlabel(country country)


delivers :

variable country not found
r(111);

Any suggestions?

[Nick Cox]

stripplot is from SSC. You found a limitation, not to say a bug.

I will look more into that, but the graph would have got would not have been, in my view, nearly so good as something like this:

clear
input str2 country warming humans bad
AT 90 60 55
BE 96 53 56
CH 96 44 59
CZ 86 35 48
DE 95 54 63
EE 90 34 45
ES 94 57 75
FI 94 51 52
FR 96 47 62
GB 93 36 52
HU 88 45 61
IE 95 38 50
IS 97 55 68
IT 91 55 53
LT 87 30 49
NL 96 44 44
NO 93 38 55
PL 90 30 52
PT 96 48 70
RU 76 28 37
SE 97 51 64
SI 96 38 59
end

set scheme s1color 

graph dot (asis) humans bad, over(country, sort(1) descending) exclude0 marker(1, ms(Oh)) marker(2, ms(+)) linetype(line) lines(lw(vthin) lc(gs12)) ysc(alt r(25 .)) subtitle(some good title)

[Nick Cox]

First up. why was there a problem in #1? When there are two or more outcome variables, stripplot stacks them temporarily into a longer layout, but variables not being plotted directly don't get carried along. I can think of adding an extra option to insist that this happens, but I am not so convinced it's a good idea.

One reason: although marker label options are allowed, they aren't typically useful for most of the purposes of this command.


You seem to have been reaching for a graph like that produced by this code:

clear
input str2 country warming humans bad
AT 90 60 55
BE 96 53 56
CH 96 44 59
CZ 86 35 48
DE 95 54 63
EE 90 34 45
ES 94 57 75
FI 94 51 52
FR 96 47 62
GB 93 36 52
HU 88 45 61
IE 95 38 50
IS 97 55 68
IT 91 55 53
LT 87 30 49
NL 96 44 44
NO 93 38 55
PL 90 30 52
PT 96 48 70
RU 76 28 37
SE 97 51 64
SI 96 38 59
end

gen one = 1 
gen two = 2 

scatter humans one, ms(none) mla(country) mlabpos(0) || scatter bad two, ms(none) mla(country)  mlabpos(0) xlabel(1 "humans" 2 "bad", noticks) xsc(r(0.8 2.2)) aspect(1) legend(off)

So,

1. That is some technique.

2. I can't see that the graph is competitive with that in post #2 above. It can be improved, but there is a long way to go.

[Lyle Scruggs]

Thanks. The method for the first will get me there.. but as you say with more work required. The goal was to produce something like your first suggestion, i.e.. to show relative unit positions on several variables in the same graph with labels legible. The method for the second organizes around the units not the variables which is not my main goal.

Is there a method to assign the labels' positions at either 3 or 9 o'clock independently for each unit. (e.g., in order to be legible CZ might need to be on at 3 o'clock for humans but 9 o'clock for bad) .... maybe

gen h_left=1 if inlist(country, countries to plot at 9oclock)
gen h_right=1 if inlist (country, countries to plot at 3 oclock)
gen b_left=1 if inlist(country, countries to plot at 9oclock)
gen b_right=1 if inlist (country, countries to plot at 3 oclock)

and then something like scatter humans one if h_left==1, ms(oh) mla(country) mlabpos(9) || scatter humans one if h_right==1, ms(oh) mla(country) mlabpos(3) || scatter bad two b_left==1, ms(oh) mla(country) mlabpos(9) || scatter bad two b_right==1, ms(oh) mla(country) mlabpos(3) xlabel(1 "humans" 2 "bad", noticks) xsc(r(0.8 2.2)) aspect(1) legend(off) I'm going to need to fix the colors but I think this is getting close

[Nick Cox]

I would be curious to see the results. I tried a stem-and-leaf plot as below.

SJ-7-3  gr0028  . . . . . . . . . . .  Speaking Stata: Turning over a new leaf
        (help stemplot if installed)  . . . . . . . . . . . . . . .  N. J. Cox
        Q3/07   SJ 7(3):413--433
        discusses variants of stem-and-leaf plots using tools
        such as scatter, by() and a new command, stemplot

clear
input str2 country warming humans bad
AT 90 60 55
BE 96 53 56
CH 96 44 59
CZ 86 35 48
DE 95 54 63
EE 90 34 45
ES 94 57 75
FI 94 51 52
FR 96 47 62
GB 93 36 52
HU 88 45 61
IE 95 38 50
IS 97 55 68
IT 91 55 53
LT 87 30 49
NL 96 44 44
NO 93 38 55
PL 90 30 52
PT 96 48 70
RU 76 28 37
SE 97 51 64
SI 96 38 59
end

stack country humans country bad, into(country answer) clear 
rename _stack which 
label def which 1 humans 2 bad
label val which which 
set scheme s1color 
stemplot answer, yla(35(5)75) by(which, note("")) width(2) ms(none) mla(country) aspect(2.5)

[Lyle Scruggs]

gen one=1
gen two=2

gen left_h=1 if inlist(cntry,"BE","SE", "IE", "NO")
gen left_b=1 if inlist(cntry,"SE", "BE", "EE","GB","AT","CH")
scatter humans one if left_h==1, ms(oh) mla(cntry) mlabpos(9) mlabsiz(vsmall) || scatter humans one if left_h~=1, ms(oh) mla(cntry) mlabpos(3) mlabsiz(vsmall) ||scatter ccbad two if left_b==1, ms(oh) mla(cntry) mlabpos(9) mlabsiz(vsmall) || scatter ccbad two if left_b~=1, ms(oh) mla(cntry) mlabpos(3) xlabel(1 "humans" 2 "bad", noticks) xsc(r(0.8 2.2)) ysc(r(.20 .80)) aspect(1) legend(off) mlabsiz(vsmall)


Needs some tweaks still, but this is the idea.

[Nick Cox]

It's your graph and your decision, manifestly.

In my examples I left out the last country E~R but it doesn't appear in #6 either that I can see.