Formatting graph legend and markers

[Scott Rick]

Hi. I am trying to format the legend size of the graph, but when I add "legend(size(small))" to the code below, I get the strange legend in the graph appear. However, if I take off the legend bit, it's shows as the dots as "Point estimate" and the shaded region as "Confidence interval", which is what I want.

I'm also unsure why the two points at either end are different in size and color.

Any help in sorting out the legend and the end points will be much appreciated.

eventdd rate_con_avg7day, hdfe absorb(state) vce(cluster state) timevar(time_to_event) ///
 graph_op(mcolor(gs1) msize(vsmall)  ylabel(, labsize(2)) legend(size(small)) ///
 xtitle("Time (weeks)", size(small)) xlabel(-6(5)35,  labsize(2))) ci(rarea , color(gs14%33))  leads(6) lags(35) accum

[Nick Cox]

eventdd is a community-contributed command, as you're asked to explain (FAQ Advice #12). The most recent version appears to be that on SSC.

There isn't a data example here that can be used to experiment (same place). In this particular case, I doubt that would help much.

Evidently your variation on graph defaults isn't being handled cleanly by the code.

A simpler alternative is to try making your changes in the Graph Editor.

If that doesn't work and you don't get a better answer I would contact the program authors directly.

All that said, I am not clear why a legend is really needed here at all.

[Nick Cox]

Another alternative is to

set trace on 

set tracedepth 1

and try to follow what is happening with your extra graph options. The handling of graphics inside the program is a little complicated. That's not a criticism, but it means that I was unable to guess quickly how your syntax would work out.

[Maarten Buis]

As an alternative to set trace on that give a bit less output you can use graph describe to see the twoway command that was used to create the graph.

[Nick Cox]

Maarten Buis makes an excellent point.

It would be good to have expository papers on debugging. I have a vague recollection of starting to write one once, but either I got bored after a few hundred words or the topic seemed too big and formless.

[Scott Rick]

Thanks Nick Cox and Maarten Buis. The legend I can do without and that's okay. However, I'm unable to figure how to edit the two end points in graph editor. I'd greatly appreciate any help here. I've also included the output from graph describe below, which I don't know how to make sense of.

graph describe
---------------------------------------------------------------------------------------------------------------------------- begin graph ---
- if d(`=c(born_date)') < d(23Jul2004) {
= if d(13 Jul 2023) < d(23Jul2004) {
  di as err "your Stata executable is out of date"
  di as err "    type -update executable- at the Stata prompt"
  exit 498
  }
- local ver = string(_caller())
- if (_caller() < 8.2) version 8
- else if (_caller() < 10 ) version 8.2
- else version 10
- gdi record = yes
- gdi maybedraw = yes
- if "`._Gr_Global.isa'" == "" {
= if "class" == "" {
  ._Gr_Global = .global_g.new
  }
- ._Gr_Global.callerver = "`ver'"
= ._Gr_Global.callerver = "18"
- capture noisily Graph `0'
= capture noisily Graph describe

Graph stored in memory

        name:  Graph
      format:  live
     created:  11 Jan 2025 13:36:52
      scheme:  s1color
        size:  4 x 5.5
    dta file:  C:\Users\dsouz\Dropbox\UNC\Dissertation\Dissertation\Aim 1\Data\analysis\input\matching.dta dated 2 Sep 2024 18:12
     command:  twoway rarea __000003 __000002 __000000, color(gs14%33) yline(0, lcolor(red)) || scatter __000001 __000000, mcolor(gs1)
               msize(vsmall) ylabel(, labsize(2)) legend(size(small)) xtitle("Time (weeks)", size(small)) xlabel(-6(5)35, labsize(2))
               xline(-1, lcolor(black) lpattern(solid)) || scatter __000001 __000000 if __000000==-6 | __000000==35,
- local rc = _rc
- gdi record = yes
- gdi maybedraw = yes
- gdi end
- exit `rc'
= exit 0

[Nick Cox]

Scott Rick / dsouz

Thanks for the detail. I am as puzzled as before.

[Chen Samulsion]

Scott Rick hi, use the example dataset of -eventdd- command, I can replicate your problem. I guess reasons of it are: 1) you use tempfile to graph as the example guides you to do, so (when something get wrong, however, I don't know what's going wrong) there are tempvar in your legend; 2) as far as I know, you should specify explicitly the legend keys by using order() option, for example, legend(size(small) order(1 "Confidence Interval" 2 "Point Estimates")); 3) the two points at either end are different in size and color because you set leads and lags option, if you take them off, the problem will disappear; or 4) if you want to set leads and lags, you should add endpoints_op to modify the plot, for example, endpoints_op(mcolor(gs1) msize(vsmall)). I will post my illustrative code and graph below.

webuse set www.damianclarke.net/stata/
webuse bacon_example.dta, clear
gen timeToTreat = year - _nfd
tempfile example
save "`example'", replace
eventdd asmrs pcinc asmrh cases, hdfe absorb(year) timevar(timeToTreat) ///
 graph_op(mcolor(gs1) msize(vsmall)  ylabel(, labsize(2)) legend(size(small)) ///
 xtitle("Time (weeks)", size(small)) xlabel(-6(5)35,  labsize(2))) ///
 ci(rarea , color(gs14%33))  leads(21) lags(27) accum

     command:  twoway rarea __000003 __000002 __000000, color(gs14%33) yline(0, lcolor(red)) || scatter __000001

               __000000, mcolor(gs1) msize(vsmall) ylabel(, labsize(2)) legend(size(small)) xtitle("Time
               (weeks)", size(small)) xlabel(-6(5)35, labsize(2)) xline(-1, lcolor(black) lpattern(solid)) ||
               scatter __000001 __000000 if __000000==-21 | __000000==27,

eventdd asmrs pcinc asmrh cases, hdfe absorb(year) timevar(timeToTreat) ///
 graph_op(mcolor(gs1) msize(vsmall)  ylabel(, labsize(2)) ///
 legend(size(small) order(1 "Confidence Interval" 2 "Point Estimates")) ///
 xtitle("Time (weeks)", size(small)) xlabel(-6(5)35,  labsize(2))) ///
 ci(rarea , color(gs14%33)) leads(21) lags(27) accum endpoints_op(mcolor(gs1) msize(vsmall))

[Scott Rick]

Chen Samulsion Thanks a lot! That solved the problem and the explanation was very helpful!