[didem okmen]

Thank you very much, I will try again.

[Nikita Sergeev]

Hello!

I have a similar problem. I am trying to predict the number of goals, which two soccer teams will score in a match. I used Negative Binomial (NB) command "nbreg" and for some models, I received LR "not concave". My explanatory variables are attack and defense parameters which take values between 0 and 2 (continuous variables) and two dependent variables Final Total Home Goals and Final Total Away Goals.
- I thought it is because the data set is small (380 observations) but when I run it for larger sets (5700 observations) the only thing changes is that another model becomes "not concave" instead of the previous one.

Therefore, I have two questions:
1) Is there is a way to improve it/get rid of "not concave" and keep all variables?
2) I would really appreciate it if in your response you included a mathematical explanation of this problem too (in the lite of NB and/or regressions matter) along with SATA explanation if this is fine with you?

it is capable of running for these models,

nbreg FTHG HAStr ADStr, vce(robust)
nbreg FTHG HAStr HDStr ADStr, vce(robust)
nbreg FTHG HAStr HDStr AAStr, vce(robust)

nbreg FTAG HDStr AAStr, vce(robust)
nbreg FTAG HDStr AAStr ADStr, vce(robust)
nbreg FTAG HAStr HDStr AAStr ADStr, vce(robust)

However for this model, in particular, it is not concave

nbreg FTHG HAStr HDStr AAStr ADStr, vce(robust)

Fitting Poisson model:

Iteration 0: log pseudolikelihood = -531.35817
Iteration 1: log pseudolikelihood = -531.35386
Iteration 2: log pseudolikelihood = -531.35386

Fitting constant-only model:

Iteration 0: log pseudolikelihood = -657.79728
Iteration 1: log pseudolikelihood = -608.06597
Iteration 2: log pseudolikelihood = -608.0658
Iteration 3: log pseudolikelihood = -608.0658

Fitting full model:

Iteration 0: log pseudolikelihood = -539.81221
Iteration 1: log pseudolikelihood = -532.72613
Iteration 2: log pseudolikelihood = -531.68284
Iteration 3: log pseudolikelihood = -531.41772
Iteration 4: log pseudolikelihood = -531.36438
Iteration 5: log pseudolikelihood = -531.3551
Iteration 6: log pseudolikelihood = -531.35412
Iteration 7: log pseudolikelihood = -531.35392
Iteration 8: log pseudolikelihood = -531.35387
Iteration 9: log pseudolikelihood = -531.35378 (not concave)
Iteration 10: log pseudolikelihood = -531.35364 (not concave)
Iteration 11: log pseudolikelihood = -531.35355 (not concave)
Iteration 12: log pseudolikelihood = -531.35355 (not concave)
Iteration 13: log pseudolikelihood = -531.35355 (not concave)
Iteration 14: log pseudolikelihood = -531.35355 (not concave)
Iteration 15: log pseudolikelihood = -531.35355 (not concave)
.....and so on



Please feel free to ask for clarification and/or additional material.