Truncating confidence intervals in marginsplot

[Gobinda Natak]

Hi everyone

I would like to create a bar chart with confidence intervals using -marginsplot-, kind of like this:

sysuse auto, clear

qui regress mpg i.rep78
qui margins, at(rep78 = (1/5))
marginsplot, recast(bar) name(panelA, replace)

This yields a beautiful figure:


But in actual reality, my data is more like this:

// Some random changes to the data
insobs 2
replace mpg  = 1000 if missing(mpg)
insobs 2
replace mpg  = -1000 if missing(mpg)
replace rep78 = 1 if missing(rep78)

and my Figure looks more like this:

qui regress mpg i.rep78
qui margins, at(rep78 = (1/5))
marginsplot, recast(bar) name(panelB, replace)

There's also nothing inherently wrong with this Figure, the fact that there is A LOT of uncertainty becomes very clear.

My problem arises when I combine the latter Figure with some other Figure, kind of like this:

graph combine panelA panelB, col(2) ycommon

This yields a Figure like this:



and here I do believe that the confidence intervals obfuscate the interpretation: Anyone would understand that the CIs include zero were they to extend only until say -1 or -10, showing their full extend however dwarfes the display of the variance in the left-hand panel.

I understand that one solution would be to remove the -ycommon- option in -graph combine-, but I still believe there are reasons to use the option. What i am looking for is, at the risk of being overly clear, a simple solution to truncate the CIs, as I did here in MS Paint:



Thanks so much!
Go