Graphing marginal effects for for only one category

[Matt Roberts]

Hi, I'm trying to produce some graphs of marginal effects but they are almost unreadable. This is my model:

ologit dep_var i.var1##c.var2

The dep_var has 5 outcomes and var1 has three categories. Ideally, I'd like to produce separate marginal effects graphs for each of the three categories of var1. I've been using this to estimate the marginal effects:

margins, dydx(var1) at(var2=(10(10)120))

This produces marginal effects for each category other than the base category, but using marginsplot results in a very messy graph. Is there any way to just plot the marginal effects only for one category of var1?

Thanks

[Clyde Schechter]

To graph, say, the marginal effect of var1 on outcome 2, at those values of var2, try:

margins, dydx(var1) predict(outcome(2)) at(var2=(10(10)120))
marginsplot

You can do the analogous thing for each outcome.

[Matt Roberts]

Thanks Clyde. I don't think I explained my problem very well. In my case, 'var1' refers to three treatment groups. Let's just call them Group 1 (control group), Group 2 (treatment 1), Group 3 (treatment 2). What I am trying to do is only recover marginal effects for 'treatment 1' so that I can produce a graph just for that group.

I can do this by creating separate variables for each of the treatment groups. Maybe that's my only solution.

[Jeff Pitblado (StataCorp)]

Here is an example we can use to discuss and maybe get the plot of interest.

* simulated data
set seed 17
set obs 1000
gen var1 = runiformint(1,3)
gen var2 = runiformint(10,120)
gen xb = var1*var2
sum xb
replace xb = 3*(xb - r(mean))/(r(max)-r(min)) + rnormal()
gen dep_var = 1 + (xb>-1) + (xb>0) + (xb>1) + (xb>2)

* model fit
ologit dep_var var1##c.var2

* marginal effects of var2 on Pr(dep_var==1), at each level of var1
margins var1, dydx(var2) at(var2=(10(10)120)) predict(outcome(1))
marginsplot, name(outcome1)
graph export outcome1.png, replace

* marginal effects of var2 on Pr(dep_var==2), at each level of var1
margins var1, dydx(var2) at(var2=(10(10)120)) predict(outcome(2))
marginsplot, name(outcome2)
graph export outcome2.png, replace

Here are the resulting graphs.

[Jeff Pitblado (StataCorp)]

I misread the details in #3.
Here is another attempt.

* simulated data
set seed 17
set obs 1000
gen var1 = runiformint(1,3)
gen var2 = runiformint(10,120)
gen xb = var1*var2
sum xb
replace xb = 3*(xb - r(mean))/(r(max)-r(min)) + rnormal()
gen dep_var = 1 + (xb>-1) + (xb>0) + (xb>1) + (xb>2)

* model fit
ologit dep_var var1##c.var2

* marginal effects of treatement1 on Pr(dep_var==1), at some levels of var2
margins r(1 2).var1, at(var2=(10(10)120)) predict(outcome(1))
marginsplot, name(outcome1)
graph export outcome1.png, replace

* marginal effects of treatement1 on Pr(dep_var==2), at some levels of var2
margins r(1 2).var1, at(var2=(10(10)120)) predict(outcome(2))
marginsplot, name(outcome2)
graph export outcome2.png, replace

And the corresponding plots: