»  Home »  Resources & support »  FAQs »  Marginal effects and the travelvl option
Note: This FAQ is for Stata 10 and older versions of Stata.

In Stata 11, the margins command replaced mfx.

## Running mfx on my dataset takes a long time, and I am worried it may have stopped. How can I tell if it is still running?

 Title Marginal effects and the travelvl option Author May Boggess, StataCorp

The tracelvl(1) option is very useful for watching mfx at work. Let’s look at an example:

 . sysuse auto, clear
(1978 Automobile Data)

. mlogit rep78 length displacement turn gear_ratio, nolog

Multinomial logistic regression                   Number of obs   =         69
LR chi2(16)     =      50.40
Prob > chi2     =     0.0000
Log likelihood = -68.492767                       Pseudo R2       =     0.2690

------------------------------------------------------------------------------
rep78 |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
1            |
length |  -.0077791   .0857701    -0.09   0.928    -.1758855    .1603272
displacement |  -.0203556   .0217033    -0.94   0.348    -.0628933    .0221822
turn |   .2606348   .4279448     0.61   0.542    -.5781215    1.099391
gear_ratio |   -1.84238   3.056085    -0.60   0.547    -7.832197    4.147436
_cons |  -2.307812   16.15227    -0.14   0.886    -33.96567    29.35005
-------------+----------------------------------------------------------------
2            |
length |  -.0257227   .0503671    -0.51   0.610    -.1244405    .0729951
displacement |  -.0227271   .0139469    -1.63   0.103    -.0500626    .0046083
turn |   .5413776   .2719147     1.99   0.046     .0084346    1.074321
gear_ratio |  -3.959613   2.582289    -1.53   0.125    -9.020805     1.10158
_cons |  -2.896173    9.00731    -0.32   0.748    -20.55018    14.75783
-------------+----------------------------------------------------------------
4            |
length |   .0485892   .0351597     1.38   0.167    -.0203226     .117501
displacement |  -.0006782    .009394    -0.07   0.942    -.0190901    .0177337
turn |  -.2454005   .1682703    -1.46   0.145    -.5752041    .0844032
gear_ratio |   1.615741   1.430829     1.13   0.259    -1.188633    4.420116
_cons |  -4.694191   7.392889    -0.63   0.525    -19.18399    9.795606
-------------+----------------------------------------------------------------
5            |
length |   .2154726   .0826013     2.61   0.009      .053577    .3773682
displacement |  -.1231205   .0456459    -2.70   0.007    -.2125848   -.0336562
turn |   -.561081   .2635563    -2.13   0.033    -1.077642   -.0445202
gear_ratio |  -4.795507   2.475835    -1.94   0.053    -9.648054    .0570406
_cons |    14.0879   11.59243     1.22   0.224    -8.632844    36.80864
------------------------------------------------------------------------------
(rep78==3 is the base outcome)

. mfx, predict(p outcome(2)) tracelvl(1)

calculating dydx (nonlinear method)

-------------------------
variable |      dy/dx
---------+---------------
length |      -.00184
displa~t |      -.00094
turn |       .02686
gear_r~o |      -.19349
-------------------------

calculating standard errors (nonlinear method)

length ... continuous
1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20
length: Std. Err. = .00272748

displacement ... continuous
1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20
displacement: Std. Err. = .00066346

turn ... continuous
1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20
turn: Std. Err. = .01852245

gear_ratio ... continuous
1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20
gear_ratio: Std. Err. = .11812354

Marginal effects after mlogit
y  = Pr(rep78==2) (predict, p outcome(2))
=  .04557427
------------------------------------------------------------------------------
variable |      dy/dx    Std. Err.     z    P>|z|  [    95% C.I.   ]      X
---------+--------------------------------------------------------------------
length |  -.0018412      .00273   -0.68   0.500  -.007187  .003505    188.29
displa~t |  -.0009443      .00066   -1.42   0.155  -.002245  .000356       198
turn |   .0268613      .01852    1.45   0.147  -.009442  .063165   39.7971
gear_r~o |  -.1934947      .11812   -1.64   0.101  -.425013  .038023   2.99928
------------------------------------------------------------------------------


I’ve chosen mlogit as an example because it is a multiple-equation estimation with many coefficients, so mfx can take quite some time to complete its calculations.

Let’s go through the extra output that tracelvl(1) has created and see what it means. The first part of the output is

 calculating dydx (nonlinear method)

-------------------------
variable |      dy/dx
---------+---------------
length |      -.00184
displa~t |      -.00094
turn |       .02686
gear_r~o |      -.19349
-------------------------


First mfx tells us that it is calculating the marginal effects and that it is using the nonlinear method (see the FAQ stata.com/support/faqs/statistics/marginal-effects-methods for an explanation of what nonlinear means). We see this output appear very quickly indeed. Let’s look at the next piece of output:

 calculating standard errors (nonlinear method)

length ... continuous
1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20
length: Std. Err. = .00272748


mfx begins by letting us know it is doing the standard errors, again using the nonlinear method. Then, for each independent variable in the estimation, we have numbers counting across the screen in front of us. It is comforting to see them because at least we know mfx is still at work, but what do they mean?

To compute the standard error of each marginal effect, we need to compute a number corresponding to each coefficient in the estimation (see the FAQ stata.com/support/faqs/statistics/marginal-effects-methods for a full description of exactly what they are). So, the list of numbers tells you which variable it is up to in the computation. We can use tracelvl(2) if we want to see each of those numbers as mfx computes it:

 . webuse fish, clear

. zip count persons, inf(child camper) vuong nolog

Zero-inflated Poisson regression                  Number of obs   =        250
Nonzero obs     =        108
Zero obs        =        142

Inflation model = logit                           LR chi2(1)      =     417.56
Log likelihood  = -895.1643                       Prob > chi2     =     0.0000

------------------------------------------------------------------------------
count |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
count        |
persons |   .8061845   .0460454    17.51   0.000     .7159371    .8964318
_cons |  -.5040989   .1658804    -3.04   0.002    -.8292184   -.1789793
-------------+----------------------------------------------------------------
inflate      |
child |   1.513283   .2652983     5.70   0.000     .9933076    2.033258
camper |  -1.007936   .3499951    -2.88   0.004    -1.693914   -.3219581
_cons |  -.3299887   .2913634    -1.13   0.257    -.9010505    .2410732
------------------------------------------------------------------------------
Vuong test of zip vs. standard Poisson:            z =     3.94  Pr>z = 0.0000

. mfx, predict(n) nonlinear tracelvl(1)

calculating dydx (nonlinear method)

-------------------------
variable |      dy/dx
---------+---------------
persons |       1.7639
child |      -1.7485
camper |       1.1334
-------------------------

calculating standard errors (nonlinear method)

persons ... continuous
1  2  3  4  5
persons: Std. Err. = .17607926

child ... continuous
1  2  3  4  5
child: Std. Err. = .2973726

camper ... discrete
1  2  3  4  5
camper: Std. Err. = .37428072

Marginal effects after zip
y  = predicted number of events (predict, n)
=  2.1880131
------------------------------------------------------------------------------
variable |      dy/dx    Std. Err.     z    P>|z|  [    95% C.I.   ]      X
---------+--------------------------------------------------------------------
persons |   1.763942      .17608   10.02   0.000   1.41883  2.10905     2.528
child |   -1.74851      .29737   -5.88   0.000  -2.33135 -1.16567      .684
camper*|    1.13338      .37428    3.03   0.002   .399803  1.86696      .588
------------------------------------------------------------------------------
(*) dy/dx is for discrete change of dummy variable from 0 to 1

. mfx, predict(n) nonlinear tracelvl(2)

calculating dydx (nonlinear method)

-------------------------
variable |      dy/dx
---------+---------------
persons |       1.7639
child |      -1.7485
camper |       1.1334
-------------------------

calculating standard errors (nonlinear method)

persons ... continuous
1  d^2f/dxdb = 6.6472574
2  d^2f/dxdb = 1.7639419
3  d^2f/dxdb = -.63714517
4  d^2f/dxdb = -.54772145
5  d^2f/dxdb = -.93149901
persons: Std. Err. = .17607926

child ... continuous
1  d^2f/dxdb = -4.4202341
2  d^2f/dxdb = -1.7485094
3  d^2f/dxdb = -1.0882794
4  d^2f/dxdb = .05773436
5  d^2f/dxdb = .09819149
child: Std. Err. = .2973726

camper ... discrete
1  d^2f/dxdb = 2.8651844
2  d^2f/dxdb = 1.1333799
3  d^2f/dxdb = -.0730116
4  d^2f/dxdb = -1.1329221
5  d^2f/dxdb = -.10674212
camper: Std. Err. = .37428072

Marginal effects after zip
y  = predicted number of events (predict, n)
=  2.1880131
------------------------------------------------------------------------------
variable |      dy/dx    Std. Err.     z    P>|z|  [    95% C.I.   ]      X
---------+--------------------------------------------------------------------
persons |   1.763942      .17608   10.02   0.000   1.41883  2.10905     2.528
child |   -1.74851      .29737   -5.88   0.000  -2.33135 -1.16567      .684
camper*|    1.13338      .37428    3.03   0.002   .399803  1.86696      .588
------------------------------------------------------------------------------
(*) dy/dx is for discrete change of dummy variable from 0 to 1