Question about "Traj" command

[Radhouene DOGGUI]

Dear honourable members,

1# Please, could someone explain me the difference between “order” and “iorder” and on which bases their values are selected ?


2# One of the provided example in help file : traj, var(y*) indep(t*) model(zip) order(0 3 3) iorder(0 -1 0)
The “iorder” option takes “-1” as a value. However, in the STATA help file it is indicated that this option could range only from 0 to 3 [i.e. iorder(numlist) optional polynomial type (0=intercept, 1=linear, 2=quadratic, 3=cubic) for the zero-inflation of each group].
Could you explain more ?

3# A last question please, I don’t understand when it is useful to generated random starting values ?


Your support is really appreciated.

Regards,

[Tom Scott]

Radhouene DOGGUI

1) order sets the number of groups and their functions -- flat (0), linear (1), exponential (2), or cubic(3). iorder specifies that there is an extra zero probability in the model and the number represents the function of that probability (e.g., 1 = linear). If this is confusing, read more about zero inflation - https://statisticalhorizons.com/zero-inflated-models

2) -1 in iorder means there is no extra zero probability, so you are just running a Poisson model

3) Using random starting values helps ensure that the maximum likelihood estimator estimates the correct parameters by finding the global maximum of the likelihood function. It is a good thing to do. You can learn more by reading about maximum likelihood estimation.

[Tom Scott]

sorry 2 means quadratic not exponential

[Radhouene DOGGUI]

Tom Scott Thank you so much Tom. Its really helpful. Again one question: How do you select the most suitable iorder values ?

[Tom Scott]

Radhouene DOGGUI I think based on a mix of theory and looking at the data, but I haven't worked with ZIPs for awhile. See if the statistical horizons link I shared helps. You could also run the same model with each iorder value separately and compare the post-estimation information criteria (BIC and AIC) across models to see which model is a better fit to the data. Maybe search for something online like "identifying the function of zero inflation in zero-inflated Poisson", or something.

[Tom Scott]

This should help you: https://scholar.google.com/scholar?h...ong+Test&btnG=

[Radhouene DOGGUI]

Radhouene DOGGUI I think based on a mix of theory and looking at the data, but I haven't worked with ZIPs for awhile. See if the statistical horizons link I shared helps. You could also run the same model with each iorder value separately and compare the post-estimation information criteria (BIC and AIC) across models to see which model is a better fit to the data. Maybe search for something online like "identifying the function of zero inflation in zero-inflated Poisson", or something.

Thank you very much, Highly appreciated.

[Radhouene DOGGUI]

Tom Scott :

Hi Tom, please when I tried to run the following syntax I got an error message
traj if gender==1, var(sweeties*) indep(age*) model(zip) order(3 3) iorder(0) start(strt)


Incorrect number of start values, the correct number is 11.
r(199);

Do you have any feedback on that ?

Thanks.

[Tom Scott]

Radhouene DOGGUI is strt a matrix of starting values that you created? I'm pretty sure you need a starting value for each parameter you are estimating. So your matrix either has too many or too few starting values.

[Radhouene DOGGUI]

Tom Scott MANY Thanks TOM

Regards, Radhouene