Test command not working after svy:melogit

[Deepali godha]

Hi,
I am using svy:melogit to estimate a 2-level model with Stata 15. It has been suggested in previous threads that we can use the 'test' command to test svy models. But my test command is not working- I have tried using it in different ways. I was advised to use 'coeflegend' but that is not working and I am not sure how to use it for nested models. I will appreciate if someone can explain why this is happening and how to check my models.
My commands are

    svyset v021 , strata(v023) || _n, weight (pwt1) singleunit(missing)

    svy:melogit lbw ||v021:, cov(unstruct)
    est store b1
    svy:melogit lbw i.matgr i.educ i.anc ||v021:, cov(unstruct)
    est store b2    
    test [b1= b2]
    test [b1_lbw= b2_lbw]
    test [_est_b1 =_est_b2]

    drop b*
    svy:melogit lbw ||v021:, cov(unstruct)
    svy:melogit, coeflegend
    svy:melogit lbw i.matgr i.educ i.anc ||v021:, cov(unstruct) 
    svy:melogit, coeflegend

I show my output with the following example

      pweight: <none>
          VCE: linearized
  Single unit: missing
     Strata 1: v023
         SU 1: v021
        FPC 1: <zero>
     Weight 1: <none>
     Strata 2: <one>
         SU 2: <observations>
        FPC 2: <zero>
     Weight 2: pwt1

.
.         svy:melogit lbw ||v021:, cov(unstruct)
(running melogit on estimation sample)

Survey: Mixed-effects logistic regression

Number of strata   =        14                  Number of obs     =      1,641
Number of PSUs     =       366                  Population size   =      1,641
                                                Design df         =        352
                                                F(   0,    352)   =          .
                                                Prob > F          =          .

------------------------------------------------------------------------------
             |             Linearized
         lbw |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       _cons |  -2.001612   .1050336   -19.06   0.000    -2.208185    -1.79504
-------------+----------------------------------------------------------------
v021         |
   var(_cons)|   .2487629   .1988702                      .0516364    1.198437
------------------------------------------------------------------------------

.         est store b1

.         svy:melogit lbw i.matgr i.educ i.anc ||v021:, cov(unstruct)
(running melogit on estimation sample)

Survey: Mixed-effects logistic regression

Number of strata   =        14                  Number of obs     =      1,641
Number of PSUs     =       366                  Population size   =      1,641
                                                Design df         =        352
                                                F(   4,    349)   =       9.42
                                                Prob > F          =     0.0000

-----------------------------------------------------------------------------------
                  |             Linearized
              lbw |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
------------------+----------------------------------------------------------------
            matgr |
             <20  |   .3802068   .1511276     2.52   0.012     .0829802    .6774334
                  |
             educ |
               1  |  -.5807697   .2355042    -2.47   0.014    -1.043942   -.1175974
               2  |  -.6029564   .1936806    -3.11   0.002    -.9838731   -.2220397
                  |
              anc |
4 or more visits  |  -.7876988   .1817539    -4.33   0.000    -1.145159   -.4302387
            _cons |  -1.062004   .1935036    -5.49   0.000    -1.442573   -.6814354
------------------+----------------------------------------------------------------
v021              |
        var(_cons)|    .317312   .2316642                      .0754909    1.333763
-----------------------------------------------------------------------------------

.         est store b2    

.         test [b1= b2]

Adjusted Wald test
equation b1 not found

.         test [b1_lbw= b2_lbw]

Adjusted Wald test
equation b1_lbw not found

.         test [_est_b1 =_est_b2]

Adjusted Wald test
equation _est_b1 not found

.         quietly svy:melogit lbw ||v021:, cov(unstruct) 
.         svy:melogit, coeflegend
last estimates not found
r(301);

Thanks for your help

[Maarten Buis]

test only works for a single model. You can still compare your two nested models, but to do soy ou need to find the constraints that would turn your second model into the first model and test those constraints. In your case that would be the constraints that the effects of i.matgr, i.educ and i.anc are all simultaneously equal to zero. You can do that with test, but more convenient is in this case testparm (it is the exact same test, but a more convenient syntax for this specific test):

testparm i.matgr i.educ i.anc

[Deepali godha]

Many thanks Maarten- this makes sense. But, what should I do when I add random effects to the model? How can I decide if this one is better than the fixed effects only?

Thanks