Menbreg convergence issues

[Robyn Jones]

Hi,

I have a dataset which i'm trying to do some multi-level modelling with. Variable mmw is a count of health problems had by individuals idauniq over time variable year.

 xtsum mmw

Variable         |      Mean   Std. dev.       Min        Max |    Observations
-----------------+--------------------------------------------+----------------
mmw      overall |  2.561441   2.000645          0         15 |     N =   21712
         between |              1.79385          0     12.875 |     n =    2714
         within  |              .886408  -2.688559   9.061441 |     T =       8

Even running a basic unconditional means model creates convergence issues.

menbreg mmw || idauniq:, irr 

Fitting fixed-effects model:

Iteration 0:   log likelihood = -46055.077  
Iteration 1:   log likelihood = -43618.123  
Iteration 2:   log likelihood = -43352.396  
Iteration 3:   log likelihood = -43345.491  
Iteration 4:   log likelihood = -43345.488  

Refining starting values:

Grid node 0:   log likelihood = -38698.321

Fitting full model:

Iteration 0:   log likelihood = -38698.321  
Iteration 1:   log likelihood = -37772.338  
Iteration 2:   log likelihood = -35674.004  
......

Iteration 298: log likelihood = -35554.677  (not concave)
Iteration 299: log likelihood = -35554.677  (not concave)
Iteration 300: log likelihood = -35554.677  (not concave)
convergence not achieved

Mixed-effects nbinomial regression              Number of obs     =     21,712
Overdispersion: mean
Group variable: idauniq                         Number of groups  =      2,714

                                                Obs per group:
                                                              min =          8
                                                              avg =        8.0
                                                              max =          8

Integration method: mvaghermite                 Integration pts.  =          7

                                                Wald chi2(0)      =          .
Log likelihood = -35554.677                     Prob > chi2       =          .
------------------------------------------------------------------------------
         mmw |  Inc. rate   Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
       _cons |   1.960179   .0323571    40.77   0.000     1.897775    2.024635
-------------+----------------------------------------------------------------
    /lnalpha |  -19.38743          .                             .           .
-------------+----------------------------------------------------------------
idauniq      |
   var(_cons)|    .645755   .0226058                      .6029343    .6916169
------------------------------------------------------------------------------
Note: Estimates are transformed only in the first equation to incidence rate.
convergence not achieved
r(430);

when introducing year as a predictor and to the random part of the model convergence is still an issue. I have scoured statalist and generally the internet and tried the following:

1. rescaling mmw
2. using xtnbreg and mepoisson
3. using option difficult and changing the technique/algorithm used

Would anyone be able to advise?

Thanks for your help,
Robyn