Dear everyone,

I am using Stata 14 on Windows 10. I am using data from a longitudinal cohort study where subjects were measured three times: at ages 15, 18 and 25 years. I need to adjust my model for smoking status (smokers = 1 and nonsmokers = 0). But I am concerned that smoking status can change in time (some quit smoking, some start smoking etc). Is it OK if i only add smoking in the fixed part of the model like this:
HTML Code:
 mixed ZvMAO c.time ib2.lifetime_drug_use i.smoking if sex ==1|| ID: time, reml cov(unstructured)

Performing EM optimization:

Performing gradient-based optimization:

Iteration 0:   log restricted-likelihood = -1389.7426  
Iteration 1:   log restricted-likelihood = -1389.5391  
Iteration 2:   log restricted-likelihood = -1389.5365  
Iteration 3:   log restricted-likelihood = -1389.5365  

Computing standard errors:

Mixed-effects REML regression                   Number of obs     =      1,099
Group variable: ID                              Number of groups  =        457

                                                Obs per group:
                                                              min =          1
                                                              avg =        2.4
                                                              max =          3

                                                Wald chi2(3)      =      44.67
Log restricted-likelihood = -1389.5365          Prob > chi2       =     0.0000

-------------------------------------------------------------------------------------
              ZvMAO |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
--------------------+----------------------------------------------------------------
               time |   .0054066   .0049086     1.10   0.271    -.0042141    .0150273
1.lifetime_drug_use |   -.363068   .0939173    -3.87   0.000    -.5471425   -.1789935
          1.smoking |  -.2684042    .056397    -4.76   0.000    -.3789402   -.1578682
              _cons |    .279152   .1242205     2.25   0.025     .0356843    .5226196
-------------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects Parameters  |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
ID: Unstructured             |
                   var(time) |   .0009494   .0009033      .0001471    .0061272
                  var(_cons) |   .9532168   .3814805      .4350489    2.088552
             cov(time,_cons) |  -.0200538   .0179578     -.0552505    .0151429
-----------------------------+------------------------------------------------
               var(Residual) |   .3805867   .0320362      .3227031     .448853
------------------------------------------------------------------------------
LR test vs. linear model: chi2(3) = 290.78                Prob > chi2 = 0.0000

Note: LR test is conservative and provided only for reference.
Or should I add it to the random effects part as well? And if yes, then how should the code look like?

Kind regards,
Urmeli