Ordered logit regression on panel data

[Jacopo Michelon]

Dear readers,
I am using Stata 15. I am trying to run an ordered logit regression on panel data with the following characteristics:

. xtset teamid year
       panel variable:  teamid (unbalanced)
        time variable:  year, 1990 to 2005
                delta:  1 unit

The dependent variable is an ordinal variable with minimum value of 1 and maximum value of 6. I am using 8 independent variables (the first three variables are binary, while the others are discrete or continuous). When I run -xtologit-, I can see that "LR test vs. ologit model" gives me zero degrees of freedom. See below. I don't know how I can interpret the result of the test. Is there anything wrong that I am doing with the code or with the model?
I don't know why I get 0 degrees of freedom using 8 independent variables even if I have 423 observations in total. Is it because the minimum number of observations per group is 9? Given the following results, can I stick to the -xtologit- model or do I need to switch to a different model? Thank you.

 xtologit season_perf TMTMplayoffL1 TMTMconferencesemifinalsL1 TMTMnbachampionL1 yearsin
> team firstyearinteam PERprvs3ssn team_current_injury5 teamsimexp

Fitting comparison model:

Iteration 0:   log likelihood = -601.26536  
Iteration 1:   log likelihood = -500.94447  
Iteration 2:   log likelihood =   -494.676  
Iteration 3:   log likelihood = -494.52421  
Iteration 4:   log likelihood = -494.52414  
Iteration 5:   log likelihood = -494.52414  

Refining starting values:

Grid node 0:   log likelihood = -505.95414

Fitting full model:

Iteration 0:   log likelihood = -505.95414  (not concave)
Iteration 1:   log likelihood = -499.68727  (not concave)
...
Iteration 47:  log likelihood = -494.52414  (backed up)

Random-effects ordered logistic regression      Number of obs     =        423
Group variable: teamid                          Number of groups  =         29

Random effects u_i ~ Gaussian                   Obs per group:
                                                              min =          9
                                                              avg =       14.6
                                                              max =         15

Integration method: mvaghermite                 Integration pts.  =         12

                                                Wald chi2(8)      =     168.86
Log likelihood  = -494.52414                    Prob > chi2       =     0.0000

----------------------------------------------------------------------------------------
           season_perf |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-----------------------+----------------------------------------------------------------
         TMTMplayoffL1 |   .9466701   .2482169     3.81   0.000     .4601739    1.433166
TMTMconferencesemifi~1 |   1.369127   .2581991     5.30   0.000     .8630663    1.875188
     TMTMnbachampionL1 |   1.933308   .5662724     3.41   0.001     .8234347    3.043182
           yearsinteam |   .3753574   .2108604     1.78   0.075    -.0379213    .7886362
       firstyearinteam |  -2.427727   .8950941    -2.71   0.007     -4.18208   -.6733752
           PERprvs3ssn |   .3886468   .0881193     4.41   0.000     .2159361    .5613575
  team_current_injury5 |   .0761697   .0677475     1.12   0.261    -.0566131    .2089524
            teamsimexp |  -1.994576    1.06875    -1.87   0.062    -4.089288    .1001358
-----------------------+----------------------------------------------------------------
                 /cut1 |   4.386512   1.558613                      1.331687    7.441338
                 /cut2 |   6.155003   1.573321                      3.071351    9.238656
                 /cut3 |   7.336614   1.586068                      4.227978    10.44525
                 /cut4 |   8.300373   1.597244                      5.169831    11.43091
                 /cut5 |   9.201347   1.613779                      6.038399     12.3643
-----------------------+----------------------------------------------------------------
             /sigma2_u |   3.50e-32   7.76e-17                             .           .
----------------------------------------------------------------------------------------
LR test vs. ologit model: chi2(0) = 0.00                  Prob > chi2 =      .

[Anat Tchetchik]

Jacopo it seems to me that the problem is that for some panel units you don't have enough observations. You have 8 parameters (not to mention the thresholds parameters) which may be problematic in terms of df despite the overall large number of obs. I would start reducing the number of params. to say, two or three, just to see if this solves the problem.

[Jacopo Michelon]

Thank Anat for your response. I tried to reduce the number of independent variables from 8 to 5. Now the degrees of freedom of "LR test vs. ologit model" are 1. This is what I obtained:

xtologit season_perf TMTMplayoffL1 TMTMconferencesemifinalsL1 firstyearinteam PERprvs3s
> sn teamsimexp

Fitting comparison model:

Iteration 0:   log likelihood = -601.26536  
Iteration 1:   log likelihood = -506.63544  
Iteration 2:   log likelihood = -502.81306  
Iteration 3:   log likelihood = -502.80427  
Iteration 4:   log likelihood = -502.80427  

Refining starting values:

Grid node 0:   log likelihood = -511.35443

Fitting full model:

Iteration 0:   log likelihood = -511.35443  (not concave)
Iteration 1:   log likelihood = -505.22737  (not concave)
Iteration 2:   log likelihood = -502.99282  
Iteration 3:   log likelihood = -502.67803  
Iteration 4:   log likelihood =  -502.6736  
Iteration 5:   log likelihood = -502.67359  

Random-effects ordered logistic regression      Number of obs     =        423
Group variable: teamid                          Number of groups  =         29

Random effects u_i ~ Gaussian                   Obs per group:
                                                              min =          9
                                                              avg =       14.6
                                                              max =         15

Integration method: mvaghermite                 Integration pts.  =         12

                                                Wald chi2(5)      =     137.97
Log likelihood  = -502.67359                    Prob > chi2       =     0.0000

----------------------------------------------------------------------------------------
           season_perf |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-----------------------+----------------------------------------------------------------
         TMTMplayoffL1 |   .9726754   .2606753     3.73   0.000     .4617613    1.483589
TMTMconferencesemifi~1 |   1.537411   .2592291     5.93   0.000     1.029332    2.045491
       firstyearinteam |  -3.148207   .7788799    -4.04   0.000    -4.674784   -1.621631
           PERprvs3ssn |   .3763192   .0905499     4.16   0.000     .1988447    .5537938
            teamsimexp |  -2.024643   1.090424    -1.86   0.063    -4.161833    .1125484
-----------------------+----------------------------------------------------------------
                 /cut1 |   2.913362   1.441067                      .0889222    5.737802
                 /cut2 |   4.668063   1.453776                      1.818714    7.517412
                 /cut3 |   5.818683   1.466154                      2.945074    8.692293
                 /cut4 |   6.754031   1.479508                      3.854248    9.653813
                 /cut5 |   7.580767   1.494963                      4.650692    10.51084
-----------------------+----------------------------------------------------------------
             /sigma2_u |   .0531939   .1138834                      .0008008    3.533523
----------------------------------------------------------------------------------------
LR test vs. ologit model: chibar2(01) = 0.26          Prob >= chibar2 = 0.3046

It seems that by reducing the number of independent variables the problem was solved. Given the Prob >= chibar2 >0.05, can I run simply -ologit, vce (cluster teamid)-? Thank you.