Margins and Ologit2

[Yawo Kokuvi]

Hello, I am fitting a ologit2 model for an ordered outcome - HIVrisk (none, low and high). I am trying to generate predicted probabilities and a plot to explain a significant interaction term in the model. Here is my full syntax:

gologit2 selfhiv2 i.Sex_female i.ego_nonwhite AGE Edu i.knowAIDSy i.condomlessy ///
 i.hivriskothers_highmed i.multi_nsy prop_race_homo  SEXUAL_CONTACTS c.Edu#i.knowAIDSy, auto lrforce
margins i.knowAIDSy, predict(outcome(1)) at(Edu=(-9.7(5)10.3)) atmeans
margins i.knowAIDSy, predict(outcome(2)) at(Edu=(-9.7(5)10.3)) atmeans
marginsplot, noci ytitle(Predicted Self-Perceived Risk) name(inteeduhowhiv)

The interaction is between education and aidsknowledge (c.Edu#i.knowAIDSy). The marginsplot command produces this: inteeduhowhiv.pdf

However, given that the DV is ordered, and there are multiple outcomes (none, low and high), with high being the excluded/reference category, which of the outcomes is the plot predicting ? How do we ensure the plot is indeed predicting none vs low & high, etc?

thanks in advance for your assistance.

best, cY

[Yawo Kokuvi]

An additional detail: or alternatlvely, I tried repeating the marginsplot command, changing the predicted outcome categories, but this didn’t have the desired effect, and mainly repeated the same graph, thus, I would appreciate some thoughts on whether one plot or two is needed in the case of ordered outcome variables. Thanks in advance

[Yawo Kokuvi]

Hello: I am providing additional detail - data extract - to help with this question.

Hopefully, that would help in getting some response to my question.

thanks - cY


----------------------- copy starting from the next line -----------------------

* Example generated by -dataex-. To install: ssc install dataex
clear
input double(selfhiv2 Sex_female ego_nonwhite AGE Edu knowAIDSy) float condomlessy double(hivriskothers_highmed multi_nsy prop_race_homo SEXUAL_CONTACTS)
1 1 0 24 12 0 1 1 1 .6666666666666666 601
2 1 0 23 12 1 1 1 0 .8333333333333334   1
0 0 1 72  5 0 1 0 0 .3333333333333333   1
1 1 1 29 12 . 1 1 1                 0   3
2 0 1 33 12 0 1 1 0              .125   3
1 0 1 59 12 0 1 0 0              .125   4
1 0 0 25 11 0 1 0 0                 1   0
0 0 0 42 13 1 0 1 0 .8333333333333334   2
1 1 0 26 14 1 1 1 1 .7692307692307693   4
1 1 0 28 12 0 1 1 0 .5833333333333334   1
end
label values selfhiv2 high_low_none
label def high_low_none 0 "none", modify
label def high_low_none 1 "low", modify
label def high_low_none 2 "high", modify

------------------ copy up to and including the previous line ------------------

[Richard Williams]

I was wondering what this ologit2 was (it is gologit2)!

For adjusted predictions and marginal effects with multiple-outcome models (including gologit2) take a look at

https://www3.nd.edu/~rwilliam/stats3/Margins05.pdf

For possible graphing ideas, check out Nick Winter's combomarginsplot, available from SSC. See especially the example in the help file for Combining plots for multiple outcomes from one model.

[Yawo Kokuvi]

Thanks so much - the combomarginsplot command worked beautifully. best wishes - cY