Unordered multinomial logit

[Simonas Raipa]

Dear all.

I have unordered data and have estimated the multinomial logit model.
This is the example of my dataset

* Example generated by -dataex-. For more info, type help dataex
clear
input long(transp_mode age_cat female) float distance_km long weather int year
2 1 1  4.979599 2 2018
4 1 0 18.339369 2 2018
1 2 1 .08924986 3 2018
1 1 1  1.104875 3 2018
4 2 0  81.17729 3 2018
4 1 1  7.263874 3 2018
1 4 1 .51526684 1 2018
4 1 0  8.373368 2 2018
2 2 0  16.95495 1 2018
5 2 0  2.393716 2 2018
4 2 0  6.160923 2 2018
5 1 0  1.087017 3 2018
2 2 0 16.391123 2 2018
2 3 0  7.263874 3 2018
4 4 0  1.380034 2 2018
2 4 0 2.8397374 1 2018
1 1 1 .08924986 1 2018
3 2 0 22.721685 3 2018
4 3 0  .4328967 3 2018
5 1 1 1.0040315 3 2018
3 2 0 14.635754 1 2018
5 4 0  1.295433 1 2018
4 1 1  95.39267 3 2018
4 1 1  95.39267 3 2018
5 1 0  .8791119 2 2018
4 1 0  11.43512 2 2018
4 1 0 12.354066 2 2018
4 2 0  1.295433 3 2018
2 2 0  2.393716 3 2018
2 2 1  7.994223 1 2018
4 1 1  4.661379 2 2018
4 1 0 174.02007 2 2018
4 1 1  9.885872 2 2018
4 1 1 20.233477 2 2018
2 2 0    5.2509 1 2018
1 2 1  .4328967 2 2018
4 1 0  4.420173 2 2018
2 2 1 18.761187 2 2018
2 2 0 23.143587 2 2018
4 3 0 28.426075 2 2018
4 3 0 19.425905 1 2018
5 1 0 2.2534153 2 2018
4 1 1  9.797473 2 2018
4 1 1 17.886116 2 2018
5 1 0 4.3756337 1 2018
2 3 1  86.65344 2 2018
2 3 1 19.038084 2 2018
2 2 0  19.25578 2 2018
3 1 1  6.160923 1 2018
3 3 0  25.56547 1 2018
3 3 1  66.38572 1 2018
5 3 1  1.232677 1 2018
4 2 0  2.096273 1 2018
3 2 0   16.4633 1 2018
2 3 0 17.351875 1 2018
4 1 0 .08924986 1 2014
5 1 0         0 1 2014
5 2 1         0 1 2014
1 4 1 .08924986 1 2014
5 2 1         0 1 2014
5 1 0 1.5548694 1 2014
1 1 1 .08924986 1 2014
. 1 1 .08924986 1 2014
1 2 1         0 1 2014
1 1 0         0 1 2014
5 1 1 .51526684 1 2014
5 1 1         0 1 2014
1 2 1 .51526684 1 2014
4 3 1 .08924986 1 2014
1 3 1 .08924986 1 2014
5 2 0 4.3756337 1 2014
5 2 0 .08924986 1 2014
5 2 0         0 1 2014
5 1 0 .08924986 1 2014
4 1 1         0 1 2014
4 2 1 .08924986 1 2014
1 1 1 .08924986 1 2014
4 1 1         0 1 2014
1 1 0         0 1 2014
3 2 1 .08924986 1 2014
3 1 0 .08924986 1 2014
4 1 1         0 1 2014
1 1 0 .51526684 1 2014
3 2 0         0 1 2014
5 2 0 .08924986 1 2014
1 1 0 1.5548694 3 2014
4 2 0 .08924986 3 2014
5 1 0 .51526684 3 2014
3 1 1 1.5548694 3 2014
1 1 0 .08924986 3 2014
4 1 0         0 3 2014
1 1 0 .08924986 3 2014
5 2 1         0 3 2014
5 2 1 .08924986 3 2014
1 1 0 .08924986 3 2014
1 1 0         0 3 2014
4 2 1         0 3 2014
4 1 1 .08924986 1 2014
1 2 1 .08924986 1 2014
5 2 1 .08924986 1 2014
end
label values transp_mode transp_mode_lab
label def transp_mode_lab 1 "Bike or moped", modify
label def transp_mode_lab 2 "Car", modify
label def transp_mode_lab 3 "Car + Public transport", modify
label def transp_mode_lab 4 "Public transport", modify
label def transp_mode_lab 5 "Walking", modify
label values age_cat age_cat
label def age_cat 1 "18-30", modify
label def age_cat 2 "31-50", modify
label def age_cat 3 "51-65", modify
label def age_cat 4 "Over 65", modify
label values female female
label def female 0 "male", modify
label def female 1 "female", modify
label values weather weather
label def weather 1 "Cloudy", modify
label def weather 2 "Rainy", modify
label def weather 3 "Sunny", modify

I have run the command

mlogit transp_mode i.age_cat female i.weather distance_km i.year

And now want to calculate the conditional marginal effects for certain type of weather on the probability of using a certain transp_mode. What would be the appropriate command and perhaps someone could add in the math formula for it?
Thank you very much for your time

Kindest regards

[Kye Lippold]

margins , dydx(i.weather)

should do it. The math formulas appear in the PDF manual for the `margins` command (see `help margins`, view the PDF manual, and look at "Methods and Formulas").

[Simonas Raipa]

Thank you for the answer!

However, this command gives me the average marginal effects. Is there a way to get conditional marginal effects for observation with certain characteristics? E.g. male/certain year/distance_traveled etc

[Andrew Musau]

Have a look at Richard Williams' slides at https://www3.nd.edu/~rwilliam/stats/Margins01.pdf or his Stata Journal article for different ways to calculate marginal effects.

[Simonas Raipa]

Thank you for the answer!

However, both sources only look at the logit model. The model I have in an example is a multinomial logit of which if I understand correctly (please correct me if I am wrong), marginal effects are extracted in a different manner.

[Andrew Musau]

I do not see why it matters. For mlogit, see

h mlogit postestimation

There is an example:

sysuse auto, clear
. mlogit rep78 mpg displ

Compute the predicted probability at the regressors' means for each outcome
. margins, atmeans

and therefore instead of computing the predicted probability at the means, you could specify your own values, e.g.,

sysuse auto, clear
mlogit rep78 mpg displ
sum mpg disp
margins, at(mpg=12 displacement=425)

Res.:

. margins, at(mpg=12 displacement=425)

Adjusted predictions                            Number of obs     =         69
Model VCE    : OIM

1._predict   : Pr(rep78==1), predict(pr outcome(1))
2._predict   : Pr(rep78==2), predict(pr outcome(2))
3._predict   : Pr(rep78==3), predict(pr outcome(3))
4._predict   : Pr(rep78==4), predict(pr outcome(4))
5._predict   : Pr(rep78==5), predict(pr outcome(5))
at           : mpg             =          12
               displacement    =         425

------------------------------------------------------------------------------
             |            Delta-method
             |     Margin   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
    _predict |
          1  |   .0146779   .0315152     0.47   0.641    -.0470908    .0764467
          2  |   .2334596   .1568673     1.49   0.137    -.0739947     .540914
          3  |   .6665109   .1630082     4.09   0.000     .3470207    .9860011
          4  |   .0851745   .0698598     1.22   0.223    -.0517481    .2220972
          5  |    .000177   .0005446     0.32   0.745    -.0008904    .0012444
------------------------------------------------------------------------------

[Richard Williams]

This handout may also be helpful, as it talks specifically about multiple-outcome commands like mlogit and ologit:

https://www3.nd.edu/~rwilliam/xsoc73994/Margins05.pdf

Incidentally, the spost13 commands (described in the handout) produce nicer-looking output than margins does.

Also, if you are condemned to using an older version of Stata (e.g. Stata 13) the necessary margins commands are a bit more tedious but still doable. This is discussed in the appendix of the handout.

"Is there a way to get conditional marginal effects for observation with certain characteristics? E.g. male/certain year/distance_traveled etc"

The spost13 command mtable (discussed in the handout) provides an easy way to do this, although you could also just use the margins command.