Announcement

Collapse
No announcement yet.
X
Collapse
  • Page of 1
  • Filter
  • Time
  • Show
Clear All
new posts
  • #1

    Combine two margins calculations in same output

    Hi,

    I wonder if it as possible to have different calculations of margins (one over a variable, one general) in order to be able to compare them after posted them.
    The marginal effects I'm calculated come from a probit model.

    Code:
    . probit Y  i.Year##(ib(0).Woman  ib(0).Less_18)  , vce(robust)  noomitted

Iteration 0:  Log pseudolikelihood = -16109.479  
Iteration 1:  Log pseudolikelihood = -15789.582  
Iteration 2:  Log pseudolikelihood = -15784.681  
Iteration 3:  Log pseudolikelihood = -15784.679  
Iteration 4:  Log pseudolikelihood = -15784.679  

Probit regression                                       Number of obs = 66,655
                                                        Wald chi2(32) = 649.99
                                                        Prob > chi2   = 0.0000
Log pseudolikelihood = -15784.679                       Pseudo R2     = 0.0202

------------------------------------------------------------------------------
             |               Robust
           Y | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
        Year |
       2010  |   .1281352   .0863191     1.48   0.138    -.0410472    .2973176
       2011  |   .0060481   .0854791     0.07   0.944    -.1614879    .1735841
       2012  |   .0287616   .0814254     0.35   0.724    -.1308293    .1883526
       2013  |  -.0023281   .0789545    -0.03   0.976     -.157076    .1524199
       2014  |   .0294893   .0793437     0.37   0.710    -.1260215    .1850002
       2015  |   .0738905    .081634     0.91   0.365    -.0861093    .2338903
       2016  |   .0573486   .0818858     0.70   0.484    -.1031446    .2178418
       2017  |   .1550162   .0858574     1.81   0.071    -.0132611    .3232936
       2018  |   .0300782   .0858373     0.35   0.726    -.1381597    .1983161
       2019  |   .0372958   .0881112     0.42   0.672     -.135399    .2099907
             |
       Woman |
     1. Yes  |   .5333064   .0702314     7.59   0.000     .3956554    .6709574
             |
     Less_18 |
     1. Yes  |   .1970167   .0673456     2.93   0.003     .0650217    .3290117
             |
  Year#Woman |
2010#1. Yes  |  -.1455256   .0992115    -1.47   0.142    -.3399766    .0489253
2011#1. Yes  |  -.2722308   .0972943    -2.80   0.005    -.4629241   -.0815374
2012#1. Yes  |  -.2519585   .0943133    -2.67   0.008    -.4368093   -.0671078
2013#1. Yes  |   -.231189   .0923654    -2.50   0.012    -.4122219   -.0501561
2014#1. Yes  |  -.2627628   .0926483    -2.84   0.005      -.44435   -.0811756
2015#1. Yes  |  -.2993844   .0949187    -3.15   0.002    -.4854215   -.1133472
2016#1. Yes  |  -.2547177   .0950445    -2.68   0.007    -.4410014    -.068434
2017#1. Yes  |  -.3310648   .0986541    -3.36   0.001    -.5244231   -.1377064
2018#1. Yes  |  -.0919079   .0998531    -0.92   0.357    -.2876165    .1038006
2019#1. Yes  |  -.1722308   .1010827    -1.70   0.088    -.3703493    .0258877
             |
Year#Less_18 |
2010#1. Yes  |  -.0021239   .0945272    -0.02   0.982    -.1873938    .1831461
2011#1. Yes  |   .0963348   .0888794     1.08   0.278    -.0778657    .2705352
2012#1. Yes  |   .0418818   .0897145     0.47   0.641    -.1339553    .2177189
2013#1. Yes  |   .0563873   .0905064     0.62   0.533    -.1210019    .2337765
2014#1. Yes  |    .061398   .0880866     0.70   0.486    -.1112485    .2340446
2015#1. Yes  |   .0667762   .0895451     0.75   0.456     -.108729    .2422815
2016#1. Yes  |  -.0124785   .0872498    -0.14   0.886    -.1834849    .1585279
2017#1. Yes  |   .0340821   .0895138     0.38   0.703    -.1413617     .209526
2018#1. Yes  |  -.0845595   .0910924    -0.93   0.353    -.2630972    .0939783
2019#1. Yes  |   -.003382    .090501    -0.04   0.970    -.1807607    .1739967
             |
       _cons |   1.134713   .0608663    18.64   0.000     1.015417    1.254009
------------------------------------------------------------------------------
    Then I calculate marginal effect (global one, not year specific).

    Code:
    . margins  , dydx(Woman Less_18)

Average marginal effects                                Number of obs = 66,655
Model VCE: Robust

Expression: Pr(Y), predict()
dy/dx wrt:  1.Woman 1.Less_18

------------------------------------------------------------------------------
             |            Delta-method
             |      dy/dx   std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
       Woman |
     1. Yes  |   .0472766    .003443    13.73   0.000     .0405284    .0540248
             |
     Less_18 |
     1. Yes  |   .0261693   .0019901    13.15   0.000     .0222688    .0300698
------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.
    And I'm interested in the effect on a specific year (to see if the is an evolution in the effect).

    Code:
    . margins  , dydx(Woman Less_18)  over(Year)

Average marginal effects                                Number of obs = 66,655
Model VCE: Robust

Expression: Pr(Y), predict()
dy/dx wrt:  1.Woman 1.Less_18
Over:       Year

------------------------------------------------------------------------------
             |            Delta-method
             |      dy/dx   std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
0.Woman      |  (base outcome)
-------------+----------------------------------------------------------------
1.Woman      |
        Year |
       2009  |   .0746144   .0127245     5.86   0.000     .0496749    .0995539
       2010  |   .0499041   .0109911     4.54   0.000      .028362    .0714463
       2011  |   .0414397    .012102     3.42   0.001     .0177202    .0651591
       2012  |   .0437596    .011085     3.95   0.000     .0220335    .0654858
       2013  |   .0480971   .0107824     4.46   0.000      .026964    .0692303
       2014  |   .0415898   .0104321     3.99   0.000     .0211432    .0620364
       2015  |   .0345284   .0104533     3.30   0.001     .0140402    .0550166
       2016  |    .042104   .0109987     3.83   0.000      .020547     .063661
       2017  |   .0275068   .0104202     2.64   0.008     .0070836      .04793
       2018  |   .0642928   .0127358     5.05   0.000     .0393311    .0892545
       2019  |   .0528174   .0127015     4.16   0.000      .027923    .0777118
-------------+----------------------------------------------------------------
0.Less_18    |  (base outcome)
-------------+----------------------------------------------------------------
1.Less_18    |
        Year |
       2009  |   .0188117   .0059635     3.15   0.002     .0071235       .0305
       2010  |   .0186067   .0058635     3.17   0.002     .0071144     .030099
       2011  |    .037299   .0067525     5.52   0.000     .0240644    .0505336
       2012  |   .0301738   .0069441     4.35   0.000     .0165638    .0437839
       2013  |   .0327143      .0072     4.54   0.000     .0186025     .046826
       2014  |   .0327338   .0067081     4.88   0.000     .0195861    .0458815
       2015  |   .0325373   .0067895     4.79   0.000     .0192301    .0458445
       2016  |   .0232434   .0066888     3.47   0.001     .0101335    .0363533
       2017  |   .0266264   .0064072     4.16   0.000     .0140685    .0391843
       2018  |   .0125332   .0066481     1.89   0.059    -.0004968    .0255631
       2019  |   .0221292   .0065912     3.36   0.001     .0092106    .0350478
------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.
    I want to compare these last marginal effects (over Years) with the global one (previous margins output). Do you have an idea on how can I do to post this two kinds of marginal effects calculations into a single posted results?
    Last edited by valérie orozco; 06 May 2025, 01:49.
    Tags: None
  • #2
    The default margins calculation is somewhat complex, as it considers hypothetical scenarios. For example, in your case, it assumes that everyone in the sample is female (regardless of their actual gender), calculates their predicted probabilities, and does the same assuming they are all male. In contrast, using the -over()- option does the same thing but calculates the marginal effects over defined groups.

    Thus, the relationship between these two sets of marginal effects is straightforward: the former is simply a weighted average of the latter. Therefore, we can combine them using, for instance, xlincom from SSC. Here is an illustration:

    Code:
    webuse lbw, clear
probit low i.race##i.smoke i.ht
margins, dydx(ht)
margins, dydx(ht) over(race) post
tab race
xlincom (1.ht:1bn.race) (1.ht:2.race) (1.ht:3.race) ///
 ((.5079*_b[1.ht:1bn.race]) + (.1376*_b[1.ht:2.race]) + (.3545*_b[1.ht:3.race]) ), post

    Res.:

    Code:
    . margins, dydx(ht)

Average marginal effects                                   Number of obs = 189
Model VCE: OIM

Expression: Pr(low), predict()
dy/dx wrt:  1.ht

------------------------------------------------------------------------------
             |            Delta-method
             |      dy/dx   std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
        1.ht |   .2377479   .1416949     1.68   0.093    -.0399689    .5154647
------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.

. margins, dydx(ht) over(race) post

Average marginal effects                                   Number of obs = 189
Model VCE: OIM

Expression: Pr(low), predict()
dy/dx wrt:  1.ht
Over:       race

------------------------------------------------------------------------------
             |            Delta-method
             |      dy/dx   std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
0.ht         |  (base outcome)
-------------+----------------------------------------------------------------
1.ht         |
        race |
      White  |   .2170212   .1389204     1.56   0.118    -.0552579    .4893002
      Black  |   .2469114   .1380679     1.79   0.074    -.0236966    .5175195
      Other  |   .2638899   .1486651     1.78   0.076    -.0274884    .5552682
------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.

. tab race

       Race |      Freq.     Percent        Cum.
------------+-----------------------------------
      White |         96       50.79       50.79
      Black |         26       13.76       64.55
      Other |         67       35.45      100.00
------------+-----------------------------------
      Total |        189      100.00

. xlincom (1.ht:1bn.race) (1.ht:2.race) (1.ht:3.race) ///
>  ((.5079*_b[1.ht:1bn.race]) + (.1376*_b[1.ht:2.race]) + (.3545*_b[1.ht:3.race]) ), post

        lc_1:  1.ht:1bn.race = 0
        lc_2:  1.ht:2.race = 0
        lc_3:  1.ht:3.race = 0
        lc_4:  (.5079*_b[1.ht:1bn.race]) + (.1376*_b[1.ht:2.race]) + (.3545*_b[1.ht:3.race]) = 0

------------------------------------------------------------------------------
             | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
        lc_1 |   .2170212   .1389204     1.56   0.118    -.0552579    .4893002
        lc_2 |   .2469114   .1380679     1.79   0.074    -.0236966    .5175195
        lc_3 |   .2638899   .1486651     1.78   0.076    -.0274884    .5552682
        lc_4 |    .237749   .1416948     1.68   0.093    -.0399677    .5154658
------------------------------------------------------------------------------
    Last edited by Andrew Musau; 06 May 2025, 03:33.

    Comment

    • #3
      Dear Andrew,
      thank you for you clear explanations and ideas.
      it is perfect!

      Comment

      Previous Next
      Working...
      X