Likelihood ratio test between glm and gllamm models

[David Torres]

Hello, all,

I'm trying to do a likelihood ratio test to compare glm and gllamm models, but get an error message. I've seen others report comparing these models (mostly in textbooks), but they never show code.

Any help is appreciated.

Thanks,
DDT

[Andrew Musau]

Can you present a data example resulting in the error?

[David Torres]

Sure:

. eststo: glm collcareer, f(b) eform

Iteration 0:   log likelihood = -4933.2867  
Iteration 1:   log likelihood =  -4932.925  
Iteration 2:   log likelihood =  -4932.925  

Generalized linear models                         No. of obs      =      7,128
Optimization     : ML                             Residual df     =      7,127
                                                  Scale parameter =          1
Deviance         =  9865.850081                   (1/df) Deviance =   1.384292
Pearson          =         7128                   (1/df) Pearson  =    1.00014

Variance function: V(u) = u*(1-u)                 [Bernoulli]
Link function    : g(u) = ln(u/(1-u))             [Logit]

                                                  AIC             =   1.384379
Log likelihood   = -4932.925041                   BIC             =  -53363.37

------------------------------------------------------------------------------
             |                 OIM
  collcareer |       Odds   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       _cons |   .9104798    .021592    -3.95   0.000     .8691286    .9537983
------------------------------------------------------------------------------
(est1 stored)

. eststo: gllamm collcareer, i(school_id) f(binomial) eform

Iteration 0:   log likelihood = -4405.6689  
Iteration 1:   log likelihood = -4390.8763  (not concave)
Iteration 2:   log likelihood = -4384.6925  
Iteration 3:   log likelihood = -4384.4694  
Iteration 4:   log likelihood = -4384.4693  
 
number of level 1 units = 7128
number of level 2 units = 43
 
Condition Number = 1.7360963
 
gllamm model
 
log likelihood = -4384.4693
 
------------------------------------------------------------------------------
  collcareer |     exp(b)   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       _cons |   1.408489   .0443982    10.87   0.000     1.324103    1.498252
------------------------------------------------------------------------------
Note: Estimates are transformed only in the first equation.
 
 
Variances and covariances of random effects
------------------------------------------------------------------------------

 
***level 2 (school_id)
 
    var(1): .61102282 (.0418296)
------------------------------------------------------------------------------

 
(est2 stored)

.
end of do-file

. lrtest est1 est2
test involves different estimators: gllamm vs. glm
r(498);

[Andrew Musau]

Try

lrtest est1 est2, force

If this does not work, present a minimal data example and I can provide a workaround.

[David Torres]

I thought of doing that, but didn't. It worked. Should have gone with my instinct.

Thanks,
DDT