Irregular gaps between bars on graph bar

[Carolin Thol]

Dear Statalisters

I ran into a strange problem when making a bar graph. I have a lot of categories over which the graph is plotted, which seems to be causing problems with the spacing of graphs, so they don't line up with labels. Here is a reproducible example using the auto.dta dataset:

sysuse auto, clear
graph bar price, over(foreign) asyvars over(make, sort(price) des label(angle(90))) legend(pos(6) col(4)) ytitle("Price") scale(0.5)

This produces the following graph:
Graph.gph

As you see, the labels under the graphs do not line up. It is unclear which bar belongs to which label. Here's what I've tried to no avail:

• Use hbar instead of bar -- same issue
• Change the scale
• Played around with bargap

Is there anything I am doing wrong or is this just a bug? Any hints appreciated.

[Wouter Wakker]

The problem lies in the nature of the data. You are trying the graph two bars (domestic and foreign) for each type of car. But by construction, every type of car is either domestic or foreign, but not both. Therefore you will only get one bar per type of car. If it's domestic, you get a blue bar towards the left of the label, for foreign cars you get a red bar towards the right side of the label.

If you are interested in the difference in price between foreign and domestic cars, here are two alternatives:

sysuse auto, clear
* Alternative 1
graph bar price, over(make, sort(price) des label(angle(90))) over(foreign) legend(pos(6) col(4)) ytitle("Price") scale(0.5) nofill
* Alternative 2
separate price, by(foreign)
graph bar price0 price1, over(make, sort(price) des label(angle(90))) legend(pos(6) col(4)) ytitle("Price") scale(0.5) ///
nofill yvaroptions(relabel(1 "Domestic"  2"Foreign"))

[Nick Cox]

Wouter Wakker gives an excellent answer, which can be tweaked.

separate price, by(foreign) veryshortlabel

is undocumented but designed for problems like these. See also https://www.stata-journal.com/sjpdf....iclenum=gr0023

[Carolin Thol]

Thank you very much, both, that was exactly what I needed. Wouter, thanks for the excellent explanation for why my approach didn't work.