Slant lines in histogram

[Pratap Pundir]

I have run a small experiment with 3 treatments (i.e. 4 groups total including control), and am trying to overlay demographic charts to show the groups are largely comparable.

The following command to generate overlaid histograms for the 4 groups for property income (catgeorical) generates the following chart.

twoway (hist income2 if control == 1, fcolor(blue%0) lcolor(blue) frequency xlabel(#6, valuelabel labsize(vsmall))  width(1) start(1) discrete xtitle("")) ///
(hist income2 if treat1 == 1, fcolor(red%0) lcolor(red) frequency xlabel(#6, valuelabel labsize(vsmall))  width(1) start(1) discrete xtitle("")) ///
(hist income2 if treat2 == 1, fcolor(green%0) lcolor(green) lpattern(dash) frequency xlabel(#6, valuelabel labsize(vsmall))  width(1) start(1) discrete xtitle(""))  ///
(hist income2 if treat3 == 1, fcolor(black%0) lcolor(black) lpattern(dash) frequency xlabel(#6, valuelabel labsize(vsmall))  width(1) start(1) discrete xtitle("")) ///
, legend(order(1 "Control" 2 "Treatment 1" 3 "Treatment 2" 4 "Treatment 3"))

What I don't get is what are the slanted lines "joining" certain histogram bars to other, and how to get rid of them.

I should mention that for certain categorical codings (12, 13, 15), there are no observations.

P.S. This does not seem to be a problem unique to `twoway`. I get the same output if I run `hist`

hist income2 if treat3 == 1, fcolor(black%0) lcolor(black) lpattern(dash) frequency xlabel(#6, valuelabel labsize(vsmall))

[Andrew Musau]

With 4 categories, it is difficult to make out the frequencies with your design. Consider tabplot from the Stata Journal. You will need to do the binning yourself.

findit tabplot

Otherwise, see https://www.statalist.org/forums/for...-diagonal-line which reports a similar issue.

[Nick Cox]

I agree with Andrew Musau that -- although the immediate problem is evidently a bug with lp(dash) -- histograms aren't going to work very well here. You have binned income data but I'd still prefer some kind of quantile or cumulative distribution plot for such data.

[Nick Cox]

Your data could be shown here as 4 x 16 frequencies,

[Pratap Pundir]

Thank you Andrew Musau and Nick Cox !

Yes, the lines are an artefact of a lpattern(dash) bug.

The data itself are categorical, unfortunately - I don't know actual incomes, just which bin a particular observation's income falls into.

Is there another, better way of plotting the comparison to show that the four groups are similarly distributed on income?

[Nick Cox]

I've already made suggestions in #3. See #4 for the suggestion that you show whatever data you're plotting in #1 i.e.

dataex income2 control treat? 

[Nick Cox]

#6 would be better as showing the results of

contract income2 treat? control 
list