Graph sizing and ylabel(...) specification

[John Mullahy]

Colleagues:

I'm stumped on how/why my graph is resized in what strikes me as a counterintuitive way based on how ylabel is specified. I'm hoping there's a simple option I can use so that regardless of the ylabel specification the x-axis will have the same length. I'm generating a series of these graphs, each with a different ylabel, and hope to have them all align.

Compare the following graphs—

The twoway command is long and messy (displayed at the end of the posting), but key element is

ylab(`ll'(.1)`ul',...)

The only difference between the commands generating these two graphs is that for the first I specify

local ll=-.301
local ul=.099

while for the second I specify

local ll=-.3
local ul=.1

Any suggestions are welcome. (I suspect this is an easy fix, but it is eluding me.) Thanks in advance.

P.S. Here's the full command

twoway (bar y frqorder if frqorder==2|frqorder==8|frqorder==14|frqorder==20, yline(0, lcolor(black*.25)) base(-.3) barw(3) fcolor(%20) lw(0))  (scatter  penonpar1 penonpar2 frqorder in 1/24, mcolor(ebblue orange) msym(d s) yline(0, lcolor(black*.25)) msize(*1.25 *1.25) ylabel(`ll'(.1)`ul', angle(360)) xlabel(1(1)24 , val angle(45) labsize(*.85) tpos(o)) legend(rows(1) pos(12) order(2 3) region(lw(0) fcolor(bg)) label(2 "Age 18-44 {&rarr} Age 45-64") label(3 "Age 45-64 {&rarr} Age 65+") ) xti("{bf:q}") l2("PE{sub:{bf:q}}", orientation(horizontal))  xsc(r(.5 24.5)))

[Andrew Musau]

Does it help if you fix the graph width using the xsize() option? E.g.,

twoway (bar...., xsize(5))

[John Mullahy]

Thanks Andrew. That's a good idea, but unfortunately the same problem arises when the same xsize(.) is set for the different ylabels. I tried specifying xsize in both the bar and the scatter and it made no difference.

[Marc Kaulisch]

I guess (!) that it has to do with the relative sizes Stata applies to the graphs. I am not sure but the last articles by Ben Jann in Stata Journal (part 1 https://www.stata-journal.com/articl...article=gr0073 part 2 https://www.stata-journal.com/articl...ticle=gr0073_1 ) shed some light on it. From reading the articles I guess you may be able to fix the sizes by providing exact sizes in inch, points ...
But I do not know how to do this in your case or if it is possible...

[John Mullahy]

Thanks Marc. I'll take a look at those SJ articles. As a stopgap I tried "padding" the values in the ylabel with blanks. It's not exact (probably since it's not a fixed-width font) but it comes pretty close.

 ylabel(-.3 "   -.3" -.2 "   -.2" -.1 "   -.1" 0 "   0" .1 "   .1",...)

But the Jann articles may be what I need.

[Nick Cox]

I am not sure why you want to specify

local ll=-.301
local ul=.099

rather than -0.3(0.1)0.1. If you really do want that, then the rest of this may not be useful.

I see here a precision problem rising as a consequence of binary approximations failing to be exact for multiples of 0.1.

One way to insist on what you want is to use mylabels (SSC).

I get this

. mylabels -0.3(0.1)0.1, myscale(@) local(yla)
-.3 "-.3" -.2 "-.2" -.1 "-.1" 0 "0" .1 ".1"

The text within quotation marks will be shown literally, and the macro contents will can be used directly.

[John Mullahy]

Thanks Nick. That's most helpful.

For the record, the "-.301" etc. was just for my Statalist example. In the real work, however, some graphs use labels with just two characters (e.g. .1) while others use four (e.g. .025).