Ordered probit regression analysis in STATA

[Faradj Koliev]

Dear STATA users,

I have to run some oprobit models

oprobit y x1 x1 x2 x3 ...

in STATA and as a R-user I wonder if there are any ways of plotting the coefficients for each level? I have a trichotomous dependent variable (0,1,2) and it would be useful to she the coefficients for each level.

I appreciate all suggestions!

[Dimitriy V. Masterov]

I am not sure if this is particularly useful exercise, but there are a couple of ways of doing that:

webuse fullauto, clear
xtile mpg_cat = mpg, nq(3)

oprobit rep77 i.mpg_cat
coefplot

oprobit rep77 i.mpg_cat
margins, dydx(*) predict(xb)
marginsplot, recast(scatter)

-coefplot- is a user-written program that you can install from SSC.

[Faradj Koliev]

Hi Dimitriy - and thanks for this.

I was thinking something that would shoud coefficients for each level of the dependent variable. This command - as I see it - shows the "raw" coefficients. This is great for binary logistic models but makes little sense for ordered. I'm thinking that perhaps the margins command could fix this. For example by generating average marginal effects for all variables at each level of the DV...

[Maarten Buis]

. sysuse auto, clear
(1978 Automobile Data)

. recode rep78 1/2=3
(rep78: 10 changes made)

. oprobit rep78 mpg i.foreign, nolog

Ordered probit regression                       Number of obs     =         69
                                                LR chi2(2)        =      31.79
                                                Prob > chi2       =     0.0000
Log likelihood = -50.299137                     Pseudo R2         =     0.2401

------------------------------------------------------------------------------
       rep78 |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         mpg |   .0587834   .0299238     1.96   0.049     .0001337     .117433
             |
     foreign |
    Foreign  |    1.44881   .3612085     4.01   0.000     .7408545    2.156766
-------------+----------------------------------------------------------------
       /cut1 |    1.88242   .6278802                      .6517976    3.113043
       /cut2 |   3.074515   .6987686                      1.704953    4.444076
------------------------------------------------------------------------------

. margins, dydx(*)

Average marginal effects                        Number of obs     =         69
Model VCE    : OIM

dy/dx w.r.t. : mpg 1.foreign
1._predict   : Pr(rep78==3), predict(pr outcome(3))
2._predict   : Pr(rep78==4), predict(pr outcome(4))
3._predict   : Pr(rep78==5), predict(pr outcome(5))

------------------------------------------------------------------------------
             |            Delta-method
             |      dy/dx   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
mpg          |
    _predict |
          1  |  -.0163376   .0080357    -2.03   0.042    -.0320873   -.0005879
          2  |   .0068833   .0037369     1.84   0.065    -.0004409    .0142075
          3  |   .0094543   .0046804     2.02   0.043     .0002809    .0186277
-------------+----------------------------------------------------------------
0.foreign    |  (base outcome)
-------------+----------------------------------------------------------------
1.foreign    |
    _predict |
          1  |   -.506235   .1141719    -4.43   0.000    -.7300078   -.2824621
          2  |   .1899206   .0686743     2.77   0.006     .0553214    .3245198
          3  |   .3163143   .1013057     3.12   0.002     .1177588    .5148699
------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.

.
. qui margins, at(foreign=0) over(mpg)

. marginsplot, noci xlab(15(5)40)

  Variables that uniquely identify margins: mpg _outcome

[Dimitriy V. Masterov]

You asked for coefficients and got coefficients.

In any case, the code is easily modified for AMEs:

margins, dydx(*)
marginsplot, xdim(_predict) xlab(, novalue) xtitle("Outcome")

[Faradj Koliev]

Is it possible to generate marginal effects for all independent variables when (first) the DV(or rep78) is ==0, then (second) ==1, then ==3 and so on? The "mchange" can do that but its not easy to plot the results..

. sysuse auto, clear
(1978 Automobile Data)

. recode rep78 1/2=3
(rep78: 10 changes made)

. oprobit rep78 mpg i.foreign, nolog

Ordered probit regression Number of obs = 69
LR chi2(2) = 31.79
Prob > chi2 = 0.0000
Log likelihood = -50.299137 Pseudo R2 = 0.2401

------------------------------------------------------------------------------
rep78 | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
mpg | .0587834 .0299238 1.96 0.049 .0001337 .117433
|
foreign |
Foreign | 1.44881 .3612085 4.01 0.000 .7408545 2.156766
-------------+----------------------------------------------------------------
/cut1 | 1.88242 .6278802 .6517976 3.113043
/cut2 | 3.074515 .6987686 1.704953 4.444076
------------------------------------------------------------------------------

. margins, dydx(*)

Average marginal effects Number of obs = 69
Model VCE : OIM

dy/dx w.r.t. : mpg 1.foreign
1._predict : Pr(rep78==3), predict(pr outcome(3))
2._predict : Pr(rep78==4), predict(pr outcome(4))
3._predict : Pr(rep78==5), predict(pr outcome(5))

------------------------------------------------------------------------------
| Delta-method
| dy/dx Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
mpg |
_predict |
1 | -.0163376 .0080357 -2.03 0.042 -.0320873 -.0005879
2 | .0068833 .0037369 1.84 0.065 -.0004409 .0142075
3 | .0094543 .0046804 2.02 0.043 .0002809 .0186277
-------------+----------------------------------------------------------------
0.foreign | (base outcome)
-------------+----------------------------------------------------------------
1.foreign |
_predict |
1 | -.506235 .1141719 -4.43 0.000 -.7300078 -.2824621
2 | .1899206 .0686743 2.77 0.006 .0553214 .3245198
3 | .3163143 .1013057 3.12 0.002 .1177588 .5148699
------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.

.
. qui margins, at(foreign=0) over(mpg)

. marginsplot, noci xlab(15(5)40)

Variables that uniquely identify margins: mpg _outcome

[ATTACH=CONFIG]n1432143[/ATTACH]

[Mambwe Shiyenge]

I have a similar problem, i want my ordered probit model to show all the coefficients of the dependent variable under each independent variable is this possible?