Hello:

I get different results using within and subpop specifications in the following case (YEAR is 2011-2017 but I show you the issue only for 2011):

Code:
. svy linearized, subpop(if $subpopulation) : logit $depvar YEAR##COHORT_10
(output ommited)
. margins gw.COHORT_10, within(2011.YEAR) contrast(nowald) subpop(if $subpopulation) vce(unconditional) mcompare(bonferroni adjustall)

Contrasts of adjusted predictions

Number of strata   =       628                  Subpop. no. obs   =    112,351
Number of PSUs     =   115,246                  Subpop. size      =          .
                                                Design df         =    114,618

Expression   : Pr(INT_3MESES), predict()

-----------------------------
               |    Number of
               |  Comparisons
---------------+-------------
All            |            7
-----------------------------

------------------------------------------------------------------------------
                             |             Linearized          Bonferroni
                             |   Contrast   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
              COHORT_10@YEAR |
(1935 o antes vs mean) 2011  |  -.5760896   .0066247      -.593911   -.5582683
   (1936-1945 vs mean) 2011  |  -.4712887   .0099504     -.4980569   -.4445205
   (1946-1955 vs mean) 2011  |  -.2590204   .0111165     -.2889254   -.2291153
   (1956-1965 vs mean) 2011  |   .0014224   .0103063     -.0263032    .0291479
   (1966-1975 vs mean) 2011  |   .1494979   .0089336      .1254651    .1735308
   (1976-1985 vs mean) 2011  |   .2427616   .0088764      .2188827    .2666405
   (1986-1995 vs mean) 2011  |   .3281684   .0078526      .3070436    .3492933
------------------------------------------------------------------------------

. margins gw.COHORT_10, contrast(nowald) subpop(if YEAR==2011 & $subpopulation) vce(unconditional) mcompare(bonferroni adjustall)

Contrasts of predictive margins

Number of strata   =       628                  Subpop. no. obs   =     18,801
Number of PSUs     =   115,246                  Subpop. size      =          .
                                                Design df         =    114,618

Expression   : Pr(INT_3MESES), predict()

---------------------------
             |    Number of
             |  Comparisons
-------------+-------------
All          |            7
---------------------------

-------------------------------------------------------------------------
                        |             Linearized          Bonferroni
                        |   Contrast   Std. Err.     [95% Conf. Interval]
------------------------+------------------------------------------------
              COHORT_10 |
(1935 o antes vs mean)  |  -.5640506    .006508     -.5815581   -.5465431
   (1936-1945 vs mean)  |  -.4592496   .0099134     -.4859184   -.4325809
   (1946-1955 vs mean)  |  -.2469813   .0111167     -.2768869   -.2170758
   (1956-1965 vs mean)  |   .0134614   .0103267      -.014319    .0412418
   (1966-1975 vs mean)  |   .1615369   .0089557      .1374446    .1856293
   (1976-1985 vs mean)  |   .2548007   .0088913      .2308817    .2787196
   (1986-1995 vs mean)  |   .3402075   .0078493      .3190917    .3613232
-------------------------------------------------------------------------
As you can see, contrast coefficient are not exactly the same in both examples, although marginal effects between COHORT_10 categories are always the same. I think this is because the background assumptions about the mean to compare with in each case; I am more interested in the second form derived from the "subpop option".

I would like to get the contrasts in the form of the "subpop version" for each year. Of course, I can do it with the "within option", writing
Code:
within(YEAR)
instead of
Code:
within(2011.YEAR)
, but not with the "subpop option". Again, the question: Is it possible to get the contrasts in the "subpop way" for all the years in just one command? [My goal is to easily plot the contrasts for all the cohorts across years, connecting the dots of the same cohort]

Thank you very much.