Margins for different percentiles

[Damian Tojeiro]

Dear Statalist,

I am trying to do a margin for an interaction depending on different values. I would like to do something like the next graph I will show. This is a toy model with the interactions I want to plot.

 mixed innprod L.offshoring L.c.Boffshoring##L.c.newlnBS_offshoring##L.c.newlnBS_offshoring  ||new_xactin: ||ident: L.offshoring , cov(unstructure)

Thereafter, I want the margins for the 25 and 75 percentiles of Boffshoring, that is, I need two different curves (25th and 75th) for several values of the other variable. But instead, it just give me for the 75th.

 margins,  at(L.newlnBS_offshoring = (0 (0.8) 4) (p25)(p75)L.Boffshoring)

Predictive margins                              Number of obs     =     93,036

Expression   : Linear prediction, fixed portion, predict()

1._at        : L.Boffshor~g    =           0 (p75)
               L.newlnBS_~g    =           0

2._at        : L.Boffshor~g    =           0 (p75)
               L.newlnBS_~g    =          .8

3._at        : L.Boffshor~g    =           0 (p75)
               L.newlnBS_~g    =         1.6

4._at        : L.Boffshor~g    =           0 (p75)
               L.newlnBS_~g    =         2.4

5._at        : L.Boffshor~g    =           0 (p75)
               L.newlnBS_~g    =         3.2

6._at        : L.Boffshor~g    =           0 (p75)
               L.newlnBS_~g    =           4

------------------------------------------------------------------------------
             |            Delta-method
             |     Margin   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         _at |
          1  |   .6513781   .1626375     4.01   0.000     .3326143    .9701418
          2  |   .6424268   .2395786     2.68   0.007     .1728614    1.111992
          3  |   .6149054   .3314346     1.86   0.064    -.0346945    1.264505
          4  |   .5688139   .4381152     1.30   0.194    -.2898761    1.427504
          5  |   .5041523   .5595803     0.90   0.368    -.5926049     1.60091
          6  |   .4209207   .6958102     0.60   0.545    -.9428422    1.784684
------------------------------------------------------------------------------

marginsplot, xdim(L.newlnBS_offshoring)

Is there any way to do something like the next graph?
Many thanks in advance.

[Phil Bromiley]

While margins will work with percentiles, I often just do a su on the variables and use actual values:

margins, at(x1=(1(1)17) x2 =(.207 .574 .853))

marginsplot ,recast(line) noci plot1opts(lpattern(line) lcolor(black) ) ///
plot2opts(lpattern("--")lcolor(blue) ) plot3opts(lpattern("-..-..") lcolor(red) ) /// ///
ytitle("Predicted OEE + Offset") ylabel(69(2)85) ///
legend(on order(1 "10th %tile" 2 "50th %tile" 3 "90th %tile" )) ///
title(Predicted OEE at Different Capacity Utilization Levels)

[Damian Tojeiro]

Thanks for the answer Phil, I will do a summary and get values from there as you suggest.