Interpreting marginal effects after ologit (1 control, 2 treatment groups)

[Ghanimah Almunaifi]

Hello, I have run

ologit playercontorder i.treatment i.below30 i.gender i.gov i.nationality i.school i.av_h_income i.child i.level i.student_level i.married i.order

margins, dydx(i.treatment)

and want to get the marginal effect of treatment. The control group is the base, and the outcome variable playercontorder is the level of contribution (in $).
I am a bit unclear as to how to interpret the marginal effects (table below).
Is this for e.g., correct: being in treatment 1, relative to the control group, decreases the prob of not contributing (level 1 of playercontorder) by 0.98%, on average?

[Erik Ruzek]

Yes. The marginal effect (dy/dx) puts the estimates into a probability increase/decrease metric. Another handy tool for looking at these results is the mtable command, which is in the user-written program spost13 (search spost13). It will give you a table of the predictions for all groups. Here's an example:

webuse nhanes2f, clear
ologit health i.region i.sex i.race weight
mtable, at(region = (1(1)4))

Resulting in the following table:

Expression: Pr(health), predict(outcome())

           |   region      poor      fair   average      good  excellent
 ----------+------------------------------------------------------------
         1 |        1     0.056     0.136     0.268     0.266      0.274
         2 |        2     0.060     0.144     0.274     0.262      0.260
         3 |        3     0.087     0.189     0.301     0.233      0.191
         4 |        4     0.075     0.171     0.292     0.246      0.216

You could reconstruct this table from the margins, dydx output, but this is much easier. You can further specify whether you want the predictions at various covariate values for other predictors in the model, just as you can with margins.

mtable, at(region = (1(1)4) weight = (61 71 81))

Expression: Pr(health), predict(outcome())

           |   region    weight      poor      fair   average      good  excellent
 ----------+----------------------------------------------------------------------
         1 |        1        61     0.052     0.129     0.261     0.269      0.288
         2 |        1        71     0.055     0.136     0.267     0.267      0.275
         3 |        1        81     0.059     0.143     0.273     0.263      0.262
         4 |        2        61     0.056     0.137     0.268     0.266      0.273
         5 |        2        71     0.059     0.143     0.274     0.263      0.261
         6 |        2        81     0.063     0.150     0.279     0.259      0.248
         7 |        3        61     0.081     0.180     0.298     0.239      0.202
         8 |        3        71     0.086     0.188     0.301     0.234      0.192
         9 |        3        81     0.091     0.196     0.303     0.228      0.182
        10 |        4        61     0.070     0.163     0.288     0.251      0.228
        11 |        4        71     0.075     0.170     0.292     0.246      0.217
        12 |        4        81     0.079     0.178     0.296     0.241      0.206

As usual, Richard Williams primer on margins with multicategory outcome models is highly recommended.

[Richard Williams]

One correction: spost13 is on Long's personal site, not on ssc. So, to find and install, give the command

search spost13

The spost commands are mostly shells for margins, but they can produce easier to read output.

[Ghanimah Almunaifi]

Thank you for the tips Erik Ruzek Richard Williams !