Re: st: Understanding how lincom gets results after margins

[jpitblado@stata.com (Jeff Pitblado, StataCorp LP)]

Ariel Linden <ariel.linden@gmail.com> is trying to use -lincom- after
-margins-, but is not getting the expected result:

> After running a poisson regression, I ran -margins- to get individual pre
> and post values for a treatment and control group. Next, I ran -lincom- in
> order to estimate the difference in differences, ie,. (treatment_post -
> treatment_pre) - (control_post - control_pre). 
> 
> The results from -lincom- do not produce the same values as when calculating
> them manually from the table produced by -margins-. Interestingly enough,
> the results from -lincom- are exactly as those produced in the poisson
> regression table for the interaction treatment#period. Can someone explain
> this discrepancy, and explain how lincom got these results? 
> 
> **** start code****
> 
> . margins postperiod , over(pilot ) coefleg
> 
> Predictive margins                                Number of obs   =
> 54794
> Model VCE    : Robust
> 
> Expression   : Predicted number of events, predict()
> over         : pilot
> ----------------------------------------------------------------------------
> ------
>                  |     Margin  Legend
> -----------------+----------------------------------------------------------
> ------
> pilot#postperiod |
>     control#pre  |   .0604865  _b[0bn.pilot#0bn.postperiod]
>    control#post  |   .0576589  _b[0bn.pilot#1.postperiod]
>     treated#pre  |   .0629165  _b[1.pilot#0bn.postperiod]
>    treated#post  |   .0602585  _b[1.pilot#1.postperiod]
> ----------------------------------------------------------------------------
> ------  
> 
> lincom (_b[1.pilot#1.postperiod] - _b[1.pilot#0bn.postperiod]) -
> (_b[0bn.pilot#1.postperiod] - _b[0bn.pilot#0bn.postperiod])
> 
> ( 1)  [imaging_post_index_tot]0b.pilot#0b.postperiod -
> [imaging_post_index_tot]0b.pilot#1o.postperiod -
>        [imaging_post_index_tot]1o.pilot#0b.postperiod +
> [imaging_post_index_tot]1.pilot#1.postperiod = 0
> 
> ----------------------------------------------------------------------------
> --
> imaging_po~t |      Coef.   Std. Err.      z    P>|z|     [95% Conf.
> Interval]
> -------------+--------------------------------------------------------------
> --
>          (1) |   .0047102   .0810194     0.06   0.954    -.1540849
> .1635053
> ----------------------------------------------------------------------------
> --
> 
> * calculating differences in differences manually from -margins- table:
> . di (.0602585 - .0629165) - (.0576589 - .0604865)
> .0001696
> 
> **** end code***

-lincom- operates on the current estimation results in -e()-.  Without the
-post- option in the call to -margins-, this means that -lincom- is using the
-e()- results from the most currently estimated model.  Ariel needs to specify
the -post- option with -margins- to get -lincom- to work with -margins- own
estimation results.

We notice that Ariel used the -coeflegend- with -margins- without also
specifying the -post- option.  -margins- should have produced an error
message, because it otherwise gives the impression that -e(b)- is from margins
instead of the originally fitted estimation.  We will fix this in a future
update to Stata 13.

--Jeff
jpitblado@stata.com
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/