In gsem, the log-likelihood remains 0 no matter what variables you put in and does not converge

[hizaki keiichi]

I am a beginner in Japanese stuttering and my English and software handling are not very good, but I would like to ask one question.

gsem (Nouse_not_necessary Nouse_satisfied no_app_bank Nouse_no_OD Nouse_private_scooter Nouse_travel_with_children Economic_reasons Low_competence Inlra Safety_reasons <-, logit)(C <- メスAge_3544 Age_4554 Age_5564 Age_65),lclass(C 3)

When the above code is run, the log likelihood remains at 0 with no sign of convergence. Since there is no sign of change when the number of variables used in the analysis is reduced, we must be doing something wrong.

The results of the analysis are output as follows

. gsem (Nouse_not_necessary Nouse_satisfied no_app_bank Nouse_no_OD Nouse_private_scooter Nouse_travel_with_children Economic_reasons Low_competence
> Inlra Safety_reasons <-, logit)(C <- Female Age_3544 Age_4554 Age_5564 Age_65),lclass(C 3)

Fitting class model:

Iteration 0: (class) log likelihood = -1345.7992
Iteration 1: (class) log likelihood = -1257.4052
Iteration 2: (class) log likelihood = -1255.4867
Iteration 3: (class) log likelihood = -1255.4838
Iteration 4: (class) log likelihood = -1255.4838

Fitting outcome model:

Iteration 0: (outcome) log likelihood = -23339.091
Iteration 1: (outcome) log likelihood = 0
Iteration 2: (outcome) log likelihood = 0

Refining starting values:

Iteration 0: (EM) log likelihood = -1255.4838
Iteration 1: (EM) log likelihood = -1255.4838

Fitting full model:

Iteration 0: log likelihood = 0 (not concave)
Iteration 1: log likelihood = 0 (not concave)
Iteration 2: log likelihood = 0 (not concave)
Iteration 3: log likelihood = 0 (not concave)
Iteration 4: log likelihood = 0 (not concave)
Iteration 5: log likelihood = 0 (not concave)
Iteration 6: log likelihood = 0 (not concave)
Iteration 7: log likelihood = 0 (not concave)
Iteration 8: log likelihood = 0 (not concave)


Iterations will continue until 8000 is exceeded, but no convergence will occur.

We apologize and ask for your patience.

Data is binary data in 1 or 2

From Japan at midnight
Hizaki

[Erik Ruzek]

A first problem is that you have variables with capital letters. sem and gsem treat any variables with a capital letter as a latent variables. You need to add the option nocapslatent to your gsem statement.

gsem (Nouse_not_necessary Nouse_satisfied no_app_bank Nouse_no_OD Nouse_private_scooter ///
                  Nouse_travel_with_children Economic_reasons Low_competence ///
                  Inlra Safety_reasons <-, logit) (C <- Female Age_3544 Age_4554 ///
                  Age_5564 Age_65),lclass(C 3) nocapslatent

Aside from that, the statement looks fine, but you may have to play with starting values to get gsem to converge. See this thread and look at posts by Weiwen Ng, who has done a ton of work sharing what they have learned about latent class analysis in Stata.

[hizaki keiichi]

Thank you for your instruction. However, the following situation does not seem to change even if I use that command... I will check the thread, but if you know anything, I would appreciate your comments.


. gsem (Nouse_not_necessary Nouse_satisfied no_app_bank Nouse_no_OD Nouse_private_scooter ///
> Nouse_travel_with_children Economic_reasons Low_competence ///
> Inlra Safety_reasons <-, logit) (C <- Female Age_3544 Age_4554 ///
> Age_5564 Age_65),lclass(C 3) nocapslatent

Fitting class model:

Iteration 0: (class) log likelihood = -1345.7992
Iteration 1: (class) log likelihood = -1257.4052
Iteration 2: (class) log likelihood = -1255.4867
Iteration 3: (class) log likelihood = -1255.4838
Iteration 4: (class) log likelihood = -1255.4838

Fitting outcome model:

Iteration 0: (outcome) log likelihood = -23339.091
Iteration 1: (outcome) log likelihood = 0
Iteration 2: (outcome) log likelihood = 0

Refining starting values:

Iteration 0: (EM) log likelihood = -1255.4838
Iteration 1: (EM) log likelihood = -1255.4838

Fitting full model:

Iteration 0: log likelihood = 0 (not concave)
Iteration 1: log likelihood = 0 (not concave)
Iteration 2: log likelihood = 0 (not concave)

The data is in the following format

Nouse_satisfied no_app_bank Nouse_no_OD Nouse_private_scooter Nouse_travel_with_children not_necessary Practical_reason Unabletouse_reason Totalban Banpavementriding
2 1 2 1 1 2 1 2 1 1
1 1 1 1 1 2 1 1 1 1
2 1 2 1 1 2 1 2 1 1
2 1 2 1 1 2 1 2 1 1