Help with linear mixed mdoel

[Al Bothwell]

I run a linear mixed model. How do I determine whether the outcome variable changed over time? Do I simply interpret the coefficient for time (called visitnumber in my dataset)?

mixed mmse ageatvisit i.decline first_age i.sex visitnumber || id:

Performing EM optimization ...

Performing gradient-based optimization:
Iteration 0:   log likelihood =  -5753.401  
Iteration 1:   log likelihood = -5753.4008  

Computing standard errors ...

Mixed-effects ML regression                     Number of obs     =      2,794
Group variable: id                              Number of groups  =        608
                                                Obs per group:
                                                              min =          2
                                                              avg =        4.6
                                                              max =         17
                                                Wald chi2(5)      =     199.16
Log likelihood = -5753.4008                     Prob > chi2       =     0.0000

------------------------------------------------------------------------------
        mmse | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
  ageatvisit |  -.0916696    .010122    -9.06   0.000    -.1115084   -.0718308
             |
     decline |
   Declined  |  -1.086876    .193037    -5.63   0.000    -1.465222   -.7085304
   first_age |   .0400132   .0109605     3.65   0.000     .0185311    .0614953
             |
         sex |
       Male  |   .0587166   .1235942     0.48   0.635    -.1835236    .3009567
 visitnumber |   .0449721   .0188307     2.39   0.017     .0080647    .0818795
       _cons |   37.02418   .4801567    77.11   0.000     36.08309    37.96527
------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects parameters  |   Estimate   Std. err.     [95% conf. interval]
-----------------------------+------------------------------------------------
id: Identity                 |
                  var(_cons) |   1.052458   .1108727      .8561182    1.293826
-----------------------------+------------------------------------------------
               var(Residual) |   2.960638    .089904      2.789571    3.142196
------------------------------------------------------------------------------
LR test vs. linear model: chibar2(01) = 258.86        Prob >= chibar2 = 0.0000

.

[Leonardo Guizzetti]

Presuming you don’t have any other variables that denote time, then yes you would interpret the coefficient for time. I would caution that you may need to change time (as visit number) into a more meaningful measure because you are modeling time as a linear, continuous effect, so a unit of time (like weeks) is more sensible. Unless visits are equally spaced, then you will not be correctly modeling time.

[Al Bothwell]

Thank you, Leonardo.