Comparison of margins after stcox

[Eddy Simms]

I have run the following:

stset et
. stcox i.x1 i.x2, nolog

        Failure _d: 1 (meaning all fail)
  Analysis time _t: et

Cox regression with Breslow method for ties

No. of subjects =     96                                Number of obs =     96
No. of failures =     96
Time at risk    = 46,172
                                                        LR chi2(4)    =  41.62
Log likelihood = -324.79515                             Prob > chi2   = 0.0000

------------------------------------------------------------------------------
          _t | Haz. ratio   Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
          x1 |
         4w  |      1.000  (base)
         6w  |      0.387      0.118  -3.1094   0.002        0.213       0.704
         8w  |      0.264      0.086  -4.0927   0.000        0.140       0.500
        10w  |      0.126      0.043  -6.0467   0.000        0.064       0.246
             |
          x2 |
          N  |      1.000  (base)
          Y  |      0.508      0.118  -2.9054   0.004        0.322       0.802
------------------------------------------------------------------------------

. margins x1, over(x2)

Predictive margins                                          Number of obs = 96
Model VCE: OIM

Expression: Predicted hazard ratio, predict()
Over:       x2

------------------------------------------------------------------------------
             |            Delta-method
             |     Margin   std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
       x2#x1 |
       N#4w  |      1.000          .        .       .            .           .
       N#6w  |      0.387      0.118   3.2760   0.001        0.155       0.619
       N#8w  |      0.264      0.086   3.0758   0.002        0.096       0.433
      N#10w  |      0.126      0.043   2.9180   0.004        0.041       0.210
       Y#4w  |      0.508      0.118   4.2896   0.000        0.276       0.740
       Y#6w  |      0.197      0.083   2.3670   0.018        0.034       0.359
       Y#8w  |      0.134      0.061   2.1952   0.028        0.014       0.254
      Y#10w  |      0.064      0.029   2.2087   0.027        0.007       0.121
------------------------------------------------------------------------------

and I would like to compare the margin Y#10w with N#10w but I cannot figure out how to do it. I tried

. lincom _b[N#4w] - _b[Y#4w]
[N#4w] not found

and various combinations with no success.

I would be grateful for any help.

Eddy

* Example generated by -dataex-. For more info, type help dataex
clear
input byte x1 long x2 double et
1 1  185
1 1  285
1 2  487
1 2  453
1 1  395
1 1  483
1 1  356
1 1  168
1 1  312
1 2  283
1 2  343
1 1  303
1 2  483
1 1  182
1 1  168
1 2  640
1 1  334
1 2  437
1 1  364
1 1  262
1 1  343
1 1  325
1 2  371
2 1  465
2 2  439
2 2  608
2 2  335
2 1  355
2 1  412
2 1  316
2 1  455
2 1  435
2 1  438
2 1  592
2 1  437
2 1  208
2 2  291
2 1  578
2 1  393
2 2  565
2 1  503
2 1  175
2 2  574
2 2  524
2 1  636
2 2  545
2 2  336
2 2  391
3 1  353
3 1  543
3 1  343
3 2  593
3 1  381
3 1  484
3 1  477
3 1  393
3 1  559
3 1  479
3 1  431
3 1  411
3 2  761
3 1  398
3 2  492
3 2  298
3 1  693
3 1  536
3 1  923
3 1  529
3 2  395
3 1  397
3 1  364
4 1  290
4 2  954
4 1  473
4 1  586
4 2  951
4 2 1060
4 1  776
4 2  693
4 1 1027
4 1  375
4 1  416
4 1  302
4 1  896
4 1  498
4 1  580
4 2  877
4 2  573
4 1  424
4 1  566
4 2  511
4 1  873
4 2  513
4 2  540
4 1  549
4 1  668
end
label values x1 grp
label def grp 1 "4w", modify
label def grp 2 "6w", modify
label def grp 3 "8w", modify
label def grp 4 "10w", modify
label values x2 ext
label def ext 1 "N", modify
label def ext 2 "Y", modify

[Clyde Schechter]

margins x1, over(x2) post
matrix list e(b)
lincom _b[1.x2#1.x1] - _b[2.x2#1.x1]

Note: You don't need to actually run the -matrix list e(b)- command. That's just there so you can see what the actual names -margins- uses are.
Though, in general, the principal is that the value labels you assigned to the variables are not used in the naming of the coefficients.

[Martyn Sherriff]

I have found that mlincom, spost13_ado from https://jslsoc.sitehost.iu.edu/stata, simplifies identifying the coefficients:

. lincom _b[1.x2#1.x1] - _b[2.x2#1.x1]

 ( 1)  1bn.x2#1bn.x1 - 2.x2#1bn.x1 = 0

------------------------------------------------------------------------------
             | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
         (1) |      0.492      0.118   4.1548   0.000        0.260       0.724
------------------------------------------------------------------------------

. mlincom (1-5)

             |   lincom    pvalue        ll        ul 
-------------+---------------------------------------
           1 |    0.492     0.000     0.260     0.724

[Eddy Simms]

Thank you both for your help.