How to group categorical variables in marginsplot of an interaction

[Keith Bolland]

I am running margins after a logistic regression to generate predicted probabilities for the interaction of two independent categorical variables.

webuse nhanes2
quietly logit highbp ib1.race ib8.sizplace
margins sizplace#race

Now I'd like to use -marginsplot, recast(bar) horizontal- to create a plot of these probabilities.

If I was doing this for a summary statistic, I could use -graph hbar- with multiple -, over()- options to get the categories nicely grouped on the y-axis:

graph hbar (mean) bpsystol, over(sizplace) over(race)

Yields:

Is there a way to replicate this layout using -marginsplot-? The closest I can get is:

marginsplot, recast(bar) horizontal xdimension(race sizplace) yscale(reverse)

…which gives me 95% of the plot I want, but has repetitive labels and doesn't add spacing between the 'race' groups:

I have also tried:

marginsplot, recast(bar) horizontal xdimension(race) plotdimension(sizeplace) yscale(reverse)

…but obviously having the bars overplotted on one another like this is not terribly useful.

Thanks in advance!

[Andrew Musau]

ssc install coefplot, replace

webuse nhanes2, clear
quietly logit highbp ib1.race ib8.sizplace
margins sizplace#race, post
est sto m1
set scheme s1mono
coefplot (m1,  keep(*1.race) asequation(White) \m1, ///
keep(*2.race) asequation(Black) \ m1, keep(*3.race) ///
asequation(Other)), recast(bar) bcolor(black%40)

I leave you to work a bit on changing the labels!

[Keith Bolland]

Thank you very much!

[Todd Nobles]

Is it possible to have a different color for each level of Rural in the above chart? For instance, the first bars in White, Black, and Other groups would be the same color, the second bar in each group would be a different color, and so on.