[Chris Boulis]

Thank you Clyde Schechter. To test my understanding, I ran -margins- after using ##, and I see (based on the output in #12) that it gave me the same HRs and standard errors (but slightly different Z, significance and confidence intervals).

. qui stcox i.hormon##i.x4
. margins x4, at(hormon=(0 1))

Adjusted predictions                            Number of obs     =        686
Model VCE    : OIM

Expression   : Predicted hazard ratio, predict()

1._at        : hormon          =           0

2._at        : hormon          =           1

------------------------------------------------------------------------------
             |            Delta-method
             |     Margin   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      _at#x4 |
        1 1  |          1          .        .       .            .           .
        1 2  |   2.430562   .7320229     3.32   0.001     .9958238    3.865301
        1 3  |   2.886813   .9192041     3.14   0.002     1.085206     4.68842
        2 1  |   .6954293   .3479263     2.00   0.046     .0135062    1.377352
        2 2  |    1.61053   .5062316     3.18   0.001     .6183339    2.602725
        2 3  |   2.533564    .902651     2.81   0.005        .7644    4.302727
------------------------------------------------------------------------------

A key benefit of this approach is that you get both the main and interaction effects in the output using ## and the total effects from the -margins- output. Is that what you meant?

[Clyde Schechter]

Yes.