marginsplot using with pwcompare

[Marry Lee]

Dear all,
I used the following command

probit High_Q_S  i.TCZ##ib1997.year_birth Xs
margins year_birth, dydx(TCZ) noestimcheck post pwcompare(effects)

To get the following table

Pairwise comparisons of average marginal effects
Model VCE    : Robust

Expression   : Pr(High_Q_S), predict()
dy/dx w.r.t. : 1.TCZ

-------------------------------------------------------------------------------
              |   Contrast Delta-method    Unadjusted           Unadjusted
              |      dy/dx   Std. Err.      z    P>|z|     [95% Conf. Interval]
--------------+----------------------------------------------------------------
1.TCZ         |
   year_birth |
1993 vs 1992  |   .0338556   .0947729     0.36   0.721    -.1518958    .2196071
1994 vs 1992  |   .1447371   .0810518     1.79   0.074    -.0141216    .3035958
1995 vs 1992  |  -.0182139   .0897682    -0.20   0.839    -.1941563    .1577286
1996 vs 1992  |   .0640309   .0997219     0.64   0.521    -.1314205    .2594823
1997 vs 1992  |   .0549888   .0832132     0.66   0.509    -.1081061    .2180838
1998 vs 1992  |   .1352737   .0891437     1.52   0.129    -.0394448    .3099921
1999 vs 1992  |   .0828824   .0994943     0.83   0.405    -.1121228    .2778875
1994 vs 1993  |   .1108815   .0712267     1.56   0.120    -.0287203    .2504832
1995 vs 1993  |  -.0520695   .0666381    -0.78   0.435    -.1826778    .0785388
1996 vs 1993  |   .0301753   .0838794     0.36   0.719    -.1342254    .1945759
1997 vs 1993  |   .0211332   .0694659     0.30   0.761    -.1150174    .1572838
1998 vs 1993  |    .101418   .0703358     1.44   0.149    -.0364376    .2392737
1999 vs 1993  |   .0490267   .0700038     0.70   0.484    -.0881781    .1862316
1995 vs 1994  |   -.162951   .0699753    -2.33   0.020       -.3001   -.0258019
1996 vs 1994  |  -.0807062    .089244    -0.90   0.366    -.2556211    .0942088
1997 vs 1994  |  -.0897482   .0693053    -1.29   0.195    -.2255841    .0460876
1998 vs 1994  |  -.0094634   .0740236    -0.13   0.898    -.1545471    .1356202
1999 vs 1994  |  -.0618547   .0759032    -0.81   0.415    -.2106223    .0869128
1996 vs 1995  |   .0822448   .0857709     0.96   0.338    -.0858632    .2503527
1997 vs 1995  |   .0732027   .0665905     1.10   0.272    -.0573122    .2037177
1998 vs 1995  |   .1534875   .0668704     2.30   0.022      .022424     .284551
1999 vs 1995  |   .1010962   .0729121     1.39   0.166    -.0418088    .2440013
1997 vs 1996  |   -.009042   .0748278    -0.12   0.904    -.1557019    .1376178
1998 vs 1996  |   .0712428   .0732969     0.97   0.331    -.0724166    .2149021
1999 vs 1996  |   .0188515   .0841226     0.22   0.823    -.1460258    .1837287
1998 vs 1997  |   .0802848   .0728376     1.10   0.270    -.0624742    .2230438
1999 vs 1997  |   .0278935   .0695323     0.40   0.688    -.1083874    .1641744
1999 vs 1998  |  -.0523913   .0747191    -0.70   0.483     -.198838    .0940554
-------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.

I am only interested in the comparisons of each year with 1997.
So I saved the results, keep only the marginal effects I am interested in as follows:

* Example generated by -dataex-. To install: ssc install dataex
clear
input byte(_deriv _term _predict _at _atopt) float(_margin _se _statistic _pvalue _ci_lb _ci_ub) int _m1 byte _pw
2 1 . . .  -.05498885  .08321324   .6608185 .50872874  -.1081061  .2180838 1992  5
2 1 . . . -.021133216  .06946585   .3042245  .7609568 -.11501735  .1572838 1993 11
2 1 . . .   .08974823  .06930527   -1.29497  .1953306 -.22558407  .0460876 1994 16
2 1 . . . -.073202714  .06659049  1.0992969 .27163857 -.05731224 .20371766 1995 20
2 1 . . .  .009042049  .07482783 -.12083805  .9038193  -.1557019  .1376178 1996 23
2 1 . . .   .08028481 .072837554  1.1022447  .2703553 -.06247418  .2230438 1998 26
2 1 . . .   .02789351  .06953234   .4011588  .6883032 -.10838737  .1641744 1999 27
end
label values _deriv _deriv
label def _deriv 2 "1.TCZ", modify
label values _term _terml
label def _terml 1 "High quality high school", modify
label values _pw _pwl
label def _pwl 5 "1992 vs 1997", modify
label def _pwl 11 "1993 vs 1997", modify
label def _pwl 16 "1994 vs 1997", modify
label def _pwl 20 "1995 vs 1997", modify
label def _pwl 23 "1996 vs 1997", modify
label def _pwl 26 "1998 vs 1997", modify
label def _pwl 27 "1999 vs 1997", modify

Then used;

marginsplot using marginsHigh, yline(0) level(95)

I got the following

high.gph

While I want a graph where the reference is 1997 so the marginal effect for 1997 to be shown as 0 on the graph as in here :

high2.gph

I really appreciate any remarks. Thank you.

[Jeff Pitblado (StataCorp)]

Instead of pairwise comparisons, you can request contrasts of the marginal means and specify a custom base for the year variable.

probit High_Q_S  i.TCZ##ib1997.year_birth Xs
margins rb1997.year_birth#TCZ, noestimcheck
marginsplot

[Marry Lee]

Thank you for your answer Jeff Pitblado (StataCorp).
That's exactly what I was looking for!
But can I also add the point 1997 vs 1997 which should appear as 0 on the graph, please?

[Jeff Pitblado (StataCorp)]

You will have to modify the data that marginsplot uses to produce the plot.
It is possible, you just need to know about margins option saving() and how to modify the dataset.
Here is an example based on a dataset used in the probit documentation.

webuse union

probit union age grade not_smsa south##b73.year

* save the -margins- results to a dataset that can be used by -marginsplot-
margins rb73.year#south, saving(marg, replace)

* default margins plot
marginsplot, xlabel(, angle(90))

* use -frame- to modify a copy of the -margins- results dataset,
* use -preserve- and -restore- if you cannot use -frame-
frame create marg
frame marg {
    use marg
    label list
    label define _m1 73 "73 vs 73", modify
    expand 2 in 1
    replace _m1 = 73 in 1
    replace _margin = 0 in 1
    replace _se_margin = 0 in 1
    replace _statistic = 0 in 1
    replace _pvalue = . in 1
    replace _ci_lb = . in 1
    replace _ci_ub = . in 1
    save marg-mod, replace
}

* call marginsplot with your modified margins results (dataset)
marginsplot using marg-mod, xlabel(, angle(90))

The relevant documentation is available via help margins saving and help marginsplot using.

[Marry Lee]

Hi Jeff Pitblado (StataCorp), thank you so much, this is of great help for me!!