constraints in joint model

[laura]

Hello,

We're trying to run a joint model with two response variables. For one response, we need to fix some of the covariances between level 2 random effects to zero. In mlwin, this model runs fine and everything looks nice.

In runmlwin, I can't get it to set the covariances to zero. I tried the same syntax as I would use in a single response model, i.e.:

matrix a = (1,1,1,0,0,1)
sort id num
runmlwin (kcal cons s1_07 s1_7_max source childcomp evalid2 evalid3 ///
male males1_07 males1_7_max, eq(1)) (bmi cons , eq(2)), ///
level2(id: (cons s1_07 s1_7_max, eq(1) reset(none)) (cons , reset(none) eq(2)) ) level1(seq: (cons agewk source, reset(none) elements(a) eq(1)))

It completely ignored this and estimated all terms of the covariance matrix.

So I tried using constraints:
constraint define 1 [RP1]cov(cons_1\source_1) = 0
constraint define 2 [RP1]cov(source_1\agewk_1) = 0
sort id num
runmlwin (kcal cons s1_07 s1_7_max source childcomp evalid2 evalid3 ///
male males1_07 males1_7_max, eq(1)) (bmi cons , eq(2)), ///
level2(id: (cons s1_07 s1_7_max, eq(1) reset(none)) (cons , reset(none) eq(2)) ) level1(seq: (cons agewk source, reset(none) eq(1))) constraints(1/2) maxiter(100)

In this model, the covariances were NEARLY zero, but not quite! And having them estimated causes problems, as the model cannot then estimate the variance of the level 1 random effect for one term (source).

Any ideas on how to fix these covariances to zero?

Many thanks,
Laura

[GeorgeLeckie]

Hi Laura,

Here is a somewhat artificial example of what you are after, Note that reset() and elements() are options that apply to the whole level. They are not equation specific options within each level. So you only need to specify them once.

Code: Select all

. use http://www.bristol.ac.uk/cmm/media/runmlwin/tutorial, clear

. matrix a = (1,1,1,0,0,1)

. runmlwin ///
>         (normexam cons girl vrband, eq(1)) ///
>         (standlrt cons girl vrband, eq(2)) ///
>         , ///
>         level2(school: ///
>                 (cons, eq(1)) ///
>                 (cons, eq(2)) ///
>         ) ///
>         level1(student: ///
>                 (cons vrband, eq(1)) ///
>                 (cons, eq(2)) ///
>                 , elements(a) ///
>         ) nopause
 
MLwiN 2.26 multilevel model                     Number of obs      =      4059
Multivariate response model
Estimation algorithm: IGLS

-----------------------------------------------------------
                |   No. of       Observations per Group
 Level Variable |   Groups    Minimum    Average    Maximum
----------------+------------------------------------------
         school |       65          2       62.4        198
-----------------------------------------------------------

Run time (seconds)   =       6.86
Number of iterations =          5
Log likelihood       = -9086.1338
Deviance             =  18172.268
------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
normexam     |
      cons_1 |   1.375687    .059028    23.31   0.000     1.259994     1.49138
      girl_1 |    .187321   .0336435     5.57   0.000     .1213811     .253261
    vrband_1 |  -.8016609   .0200792   -39.93   0.000    -.8410153   -.7623065
-------------+----------------------------------------------------------------
standlrt     |
      cons_2 |   1.949292   .0427322    45.62   0.000     1.865539    2.033046
      girl_2 |   .0559293   .0272636     2.05   0.040     .0024936     .109365
    vrband_2 |  -1.077162   .0175952   -61.22   0.000    -1.111648   -1.042677
------------------------------------------------------------------------------

------------------------------------------------------------------------------
   Random-effects Parameters |   Estimate   Std. Err.     [95% Conf. Interval]
-----------------------------+------------------------------------------------
Level 2: school              |
                 var(cons_1) |   .0951694   .0188035      .0583152    .1320236
          cov(cons_1,cons_2) |   .0250439   .0076361      .0100774    .0400103
                 var(cons_2) |   .0207908   .0051307      .0107347    .0308469
-----------------------------+------------------------------------------------
Level 1: student             |
                 var(cons_1) |   .5739033   .0991005      .3796699    .7681366
        cov(cons_1,vrband_1) |   .0054021   .0554558     -.1032893    .1140935
               var(vrband_1) |   .0024659   .0291387     -.0546448    .0595767
                 var(cons_2) |   .4741683   .0106074      .4533782    .4949585
------------------------------------------------------------------------------

Best wishes

George

[laura]

Thanks George, all working fine now!
L