LR test for multilevel estimation with bhhh

[Jessica Rayse]

I am adapted to use the xtmixed command for multilevel models. With this command, the LR test vs. linear model is as standard outuput. I had to run a model with bhhh maxization, but the result of the lr test did not come. How can I perform the lr comparison test with the OLS model after using the bhhh estimation?
Command used for Model HLM:meglm da tempo avgroa avgat avgend i.setor i.SL avgpib avghdi avgcpibot if ifrs == 2 || país:tempo || id: tempo, family(gaussian) link(identity) difficult technique(bhhh) nonrtolerance

Thanks,

[Andrew Musau]

You can use the lrtest command or calculate the test statistic by hand. Here is an example where I replicate the test at the foot of the meglm output table.

webuse nlswork, clear
regress ln_w grade age c.age#c.age ttl_exp tenure c.tenure#c.tenure
estimates store m1
meglm ln_w grade age c.age#c.age ttl_exp tenure c.tenure#c.tenure || id:, family(gaussian) link(identity)
estimates store m2
lrtest m1 m2, force

Res.:

. meglm ln_w grade age c.age#c.age ttl_exp tenure c.tenure#c.tenure || id:, family(gaussian) link(identity)

Fitting fixed-effects model:

Iteration 0:   log likelihood =     -13550  
Iteration 1:   log likelihood =     -13550  

Refining starting values:

Grid node 0:   log likelihood = -11021.153

Fitting full model:

Iteration 0:   log likelihood = -11021.153  
Iteration 1:   log likelihood = -9302.8602  
Iteration 2:   log likelihood = -9171.3842  
Iteration 3:   log likelihood = -9148.9175  
Iteration 4:   log likelihood = -9148.9073  
Iteration 5:   log likelihood = -9148.9073  

Mixed-effects GLM                               Number of obs     =     28,099
Family:                Gaussian
Link:                  identity
Group variable:          idcode                 Number of groups  =      4,697

                                                Obs per group:
                                                              min =          1
                                                              avg =        6.0
                                                              max =         15

Integration method: mvaghermite                 Integration pts.  =          7

                                                Wald chi2(6)      =    8365.72
Log likelihood = -9148.9073                     Prob > chi2       =     0.0000
-----------------------------------------------------------------------------------
          ln_wage |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
------------------+----------------------------------------------------------------
            grade |   .0707903   .0017957    39.42   0.000     .0672708    .0743099
              age |   .0318439   .0027203    11.71   0.000     .0265122    .0371755
                  |
      c.age#c.age |  -.0006513    .000045   -14.48   0.000    -.0007394   -.0005632
                  |
          ttl_exp |   .0352285   .0011392    30.92   0.000     .0329958    .0374612
           tenure |   .0371336   .0015723    23.62   0.000     .0340519    .0402153
                  |
c.tenure#c.tenure |  -.0018382   .0001013   -18.15   0.000    -.0020367   -.0016397
                  |
            _cons |   .1472124   .0447254     3.29   0.001     .0595523    .2348726
------------------+----------------------------------------------------------------
idcode            |
        var(_cons)|   .0715801   .0019392                      .0678785    .0754836
------------------+----------------------------------------------------------------
    var(e.ln_wage)|   .0853937     .00079                      .0838593    .0869561
-----------------------------------------------------------------------------------
LR test vs. linear model: chibar2(01) = 8802.19       Prob >= chibar2 = 0.0000

.
. estimates store m2

.
. lrtest m1 m2, force

Likelihood-ratio test                                 LR chi2(2)  =   8802.19
(Assumption: m1 nested in m2)                         Prob > chi2 =    0.0000

.

[Jessica Rayse]

Thank you, Andrew Musau !