Linear regression mixed effects graph

[Marco Lazzeretti]

Hi,
I'm using Stata 12 and I've a problem with a graphics of a linear regression with mixed effects. Tempomesi is a continuous variable while trattato is binary.

. xtmixed ild c.tempomesi##i.trattato || pid:, mle nolog variance 

Mixed-effects ML regression                     Number of obs      =       143
Group variable: pid                             Number of groups   =        61

                                                Obs per group: min =         2
                                                               avg =       2.3
                                                               max =         6


                                                Wald chi2(3)       =     28.76
Log likelihood =  -550.1897                     Prob > chi2        =    0.0000

--------------------------------------------------------------------------------------
                 ild |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
---------------------+----------------------------------------------------------------
           tempomesi |   .3300396   .0657651     5.02   0.000     .2011423    .4589369
          1.trattato |  -3.826831   4.163937    -0.92   0.358      -11.988    4.334336
                     |
trattato#c.tempomesi |
                  1  |  -.2629648   .1021319    -2.57   0.010    -.4631397   -.0627899
                     |
               _cons |   20.78178   3.101511     6.70   0.000     14.70293    26.86063
--------------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
pid: Identity                |
                  var(_cons) |   226.6245    44.9081      153.6852     334.181
-----------------------------+------------------------------------------------
               var(Residual) |   43.09052   6.757185       31.6885    58.59517
------------------------------------------------------------------------------
LR test vs. linear regression: chibar2(01) =    96.58 Prob >= chibar2 = 0.0000

When I graph this regression with this command:

twoway (lfit ild tempomesi if trattato==0) (lfit ild tempomesi if trattato==1)

but red line is wrong. Why?

[Maarten Buis]

That is because lfit just looks at a linear model and does not take into account panel nature of your data. To display your model you need to use margins and marginsplot, e.g.:

version 12
webuse nlswork, clear
xtmixed ln_w i.union##c.grade || id:
margins, over(grade union)
marginsplot

[Marco Lazzeretti]

Ok, but can i obtain only the predicted line?
Thank you anyway

[Marco Lazzeretti]

Mission accomplished, I do it